0:00

This podcast is brought to you by

0:02

CD Key Offer.com. Use offer code Broken

0:04

Silicon to save 25% on all Microsoft

0:06

Software and also use offer code Broken

0:08

Silicon to save $30 on the Minies

0:10

Forum V3, 3, and 1 tablet. We'll

0:12

talk about these sponsors more later, but

0:14

for now, let's just go on with

0:16

the show. Welcome

0:44

to Broken Silicon, a gaming hardware podcast.

0:46

I am your host, Tom. And today,

0:48

the guest is someone who came on

0:50

almost half a year ago, I would

0:52

say, because I know you spoke with

0:55

Wendell at level one text, you spoke

0:57

with me, you were essential to really

0:59

breaking the news on the Intel Raptor

1:01

Lake Flau. So I really would recommend people

1:03

listen to that episode too. It's a

1:05

very good one, I thought, one of

1:08

the better ones. And I actually do.

1:10

want to start there just to kind

1:12

of for the people who didn't watch

1:15

that episode or maybe forgot that whole

1:17

saga that was all anyone was talking

1:19

about for two months I would say

1:22

maybe more than two months like just

1:24

what was the whole rapid like

1:26

flaw thing based on you know we're

1:28

months away from it to your memory

1:30

how is it ended up in your

1:32

mind and yeah what have you done

1:34

what I mean I think we all deserve

1:36

an update like you had tons of

1:39

cost imposed on your company by

1:41

these Raptor Lake failures. What has

1:43

become of all this for you? Sure,

1:45

sir. Yeah, with the Raptor Lake thing, I

1:47

had, we're talking hundreds of

1:49

thousands of dollars of problems

1:51

and damages and we've prepaid

1:53

for 14th-gen service for years

1:56

as well in advance, a lot of

1:58

companies will will prepay for this over

2:00

for a year. So that becomes a

2:02

particular problem where you find out all

2:04

the servers are defective. So it ended

2:06

up being, you know, six months of

2:09

finding out it was broken. And yeah,

2:11

ended up sending data to Wendell and,

2:13

you know, various other companies, gamers, and

2:15

reviewers and saying, hey, please cover this,

2:17

this is broken, Intel is not fixing

2:19

it. And we ended up softening everything

2:21

over to AMD and we're buying Zen4

2:23

servers like crazy now. And it's solid.

2:25

And we still have 10,000 crashes, you

2:28

know, a day from our player base

2:30

that still have these CPUs. But on

2:32

our side, we don't, we just have

2:34

them banned company one. You know, on

2:36

average, every five minutes, a raptor like

2:38

user crashes in our game still. Which,

2:40

please tell people what your games are

2:42

as well. Yeah, I work on part

2:44

of times as a dinosaur survival game.

2:47

You know, I've been working on five

2:49

years now. Because I remember, you know,

2:51

I don't want to regurgitate the conversation

2:53

we had too much, but you know,

2:55

I remember, God, it's already 2025, I

2:57

think it was late 2023, I had

2:59

a contact of mine in retail telling

3:01

me, by the way, we get a

3:03

lot of Raptor Lake returns, like, you

3:06

know, and then it started to boil

3:08

and boil and boil more and more

3:10

and more, and it's like, because at

3:12

first I think there was this thought

3:14

of like, well, you know, Raptor Lake

3:16

uses uses more energy, you know rocket

3:18

lake had more failures than comment lake

3:20

which had more failures than coffee lake

3:22

I was told but then over time

3:25

it was like oh no it's not

3:27

just like 20% more this is like

3:29

a lot a lot a lot a

3:31

lot lot lot more and you know

3:33

there's the oxygenation issue that I don't

3:35

even remember it really came of that

3:37

I pretty sure until has been being

3:39

sued over some of these things right

3:41

now and then it just I from

3:44

what I was told of people at

3:46

people at Intel I believe there really

3:48

is just a flaw on the extra

3:50

proved recently when we I've seen a

3:52

lot of gay people on Twitter like

3:54

so gang types like why is Arrow

3:56

Lakes ring bus clocked so low? It's

3:58

like yeah I was directly told that's

4:00

why Raptor Lake failed is because the

4:03

ring bus was clock too high. I

4:05

was told by people who worked on

4:07

Alder Lake that they were very worried

4:09

about that and that's why they didn't

4:11

push Alder Lakes ring bus crazy high.

4:13

So there you go on that one

4:15

but I guess my question is after

4:17

all has been said and done do

4:19

you think this information got to enough

4:22

people? Do you think that this has

4:24

been even remotely resolved resolved enough enough?

4:26

Yeah, so I think this is kind

4:28

of being buried and covered up by

4:30

Intel, because most systems are not going

4:32

to get that bias fixed, even if

4:34

it works, right? I don't know if

4:36

it's still being long enough to conclude

4:38

if it's perfect or not, but without

4:41

it being a Windows update forced thing

4:43

on people, or there's a message that

4:45

pops up in Windows that says, hey,

4:47

you're missing this critical fix, your PC's

4:49

going to die, which they don't want

4:51

to do for the PR, the PR

4:53

backlash. It's gonna keep people PC's is

4:55

just gonna keep dying on mass And

4:57

I am list I'm just I just

5:00

occur I'm gonna text someone retail right

5:02

now. Are you still getting tons of

5:04

Raptor Lake returns because I'm honestly curious

5:06

on that one. I mean, because I

5:08

know you in your games, you have

5:10

a thing pop up right if it

5:12

crashes. That's like, hey, just so you

5:14

know, it's because they're after like, please

5:16

don't spam us on Reddit, don't you?

5:19

Yeah, we started working on a tool

5:21

that will tech. We hard coded every

5:23

buyouts version in the tool to try

5:25

to detect every motherboard revision and every

5:27

vass revision to try to check if

5:29

someone's patched because we don't want to

5:31

annoy the user with an error message

5:33

or something that we've already patched. But

5:35

there's OEMs that have bias updates that

5:38

have not shipped the bias update to

5:40

customers or the bias update of locks,

5:42

you know, certain biases not flagged with

5:44

a certain OEM code from even updating.

5:46

So there's people that, you know, bought

5:48

it from a system integrator and then

5:50

they haven't got the fix. And games

5:52

just crashed. And when you talk to

5:54

users that have... this defect, I can

5:57

send them Windows video or a game

5:59

is an excess video, and they're like,

6:01

yeah, but only your game crashes. What

6:03

do you mean? My CPU is effective.

6:05

But that's not true. I believe there's

6:07

another developer that also has that sort

6:09

of warning, at least one other one.

6:11

You're not the only one to my

6:13

memory. Yeah, but to the end consumer,

6:16

it's, I didn't get a crash. I play your

6:18

game now. Now it crashes,

6:20

and you kind of have

6:22

to go into this. Well,

6:24

I'm religion five is super-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on-on

6:26

and you have to pull

6:28

to try to get them

6:30

help. And yeah, I have

6:32

people with Dell where they had

6:35

a Dell PC and they tried to get

6:37

an RMA on it was broken and Dell

6:39

just said, no, we're not RMAing this.

6:41

So I'm like, man, that's that's kind

6:43

of scamming the customer, right?

6:45

So like how many of these

6:47

RMA's and system OEMs, even though

6:50

our Intel's, you know, supposedly accepting

6:52

RMA's, how many Dell systems out

6:54

they are getting rejected because you

6:56

want a different process of buying new

6:58

PC? like we're not we're not arming

7:00

this. That's kind of rough. Yeah I just

7:03

got feedback from the person I

7:05

told you everyone here that I texted

7:07

and they said yep they still get

7:09

a lot of rapture like returns but

7:11

they don't sell them anymore. I wonder

7:14

why. proportionally I believe it's the same

7:16

so they still just seem to be

7:18

failing a lot they just don't carry

7:20

them anymore and there's a lot of

7:22

kind of especially like mom and pop.

7:25

pre-built companies that are kind of

7:27

done with them as well. Although I

7:29

assume the larger the company the more

7:31

you just can't do that because

7:33

you've got this huge contract with

7:36

Intel. Yeah, my several providers that

7:38

I was using buys their CPUs

7:40

directly from Intel and they were

7:42

saying that, hey, they just don't

7:44

want to carry until CPUs anymore

7:46

because of the way they handled

7:48

it. And as far as I know, they still

7:50

haven't got RMA's accepted on

7:53

some of these processes. You know,

7:55

you've wasted a whole bunch of time

7:57

suing them and you're not really providing

7:59

your game service? Was, you know, what's

8:01

the point? I mean, I guess, you

8:03

know, I see, Sean Yeager wrote in

8:06

with a question, but I think we

8:08

pretty much answered all this questions already

8:10

with our conversation. So I'll just try

8:12

to put a bow on this one

8:14

by asking, I assume you talked to

8:16

Intel, I mean, did they, has it

8:18

been followed up with a decent amount

8:21

or not? Because at first Intel acted,

8:23

you know, oh, we're earnest, we want

8:25

to help everyone. Yeah, so I had

8:27

a call with Intel and most of

8:29

this call is under NDA, so I

8:31

can't share too many details. But the

8:33

public info I can share is they

8:36

were horrible on the call, right? Like

8:38

there was an Intel gaming lead there

8:40

and they were talking it like half

8:42

the coal was disagreeing with each other

8:44

saying, oh, I've been working a client

8:46

Intel for 20 years and I think

8:48

you're wrong and this is the floor

8:51

is not here. And this is the

8:53

floor is not here. And it was

8:55

for a development relation standpoint, it was

8:57

horrible. It was like, this sounds as

8:59

bad as all my other vendors that

9:01

have really bad relationships with. You know,

9:03

they offered to buy, I think laptops

9:05

off me that had the rapidly floor

9:08

on laptop, because that is a still

9:10

a, underdeveloped story, because a lot of

9:12

those laptops still hasn't got fixes or

9:14

bias updates or, I guess, acknowledged on

9:16

mass. But I ended up not sending

9:18

that stuff to Intel, because I was

9:20

worried that. They wouldn't just be honest

9:23

with me. And, you know, because they

9:25

could just buy my laptops and say,

9:27

hey, their laptop's actually not wrapped with,

9:29

like, defect. It's something else. Put out

9:31

a press conference or like a press

9:33

release or whatever. And then say, but

9:35

don't worry, we sent Aldra, on games,

9:38

out of the generosity of our hearts,

9:40

you know, a bunch of free laptops

9:42

anyway, just in case, but also we're

9:44

not giving you the other ones back.

9:46

Yeah, well, well, they said, we'll buy

9:48

new. Intel laptops for me like I

9:50

don't want so. The main hour issue

9:53

is the bigger problem. I mean speaking

9:55

with again. speaking to people in retail

9:57

and pre-built and OEMs like they're like

9:59

yeah I mean the cost of the

10:01

CPU is a big deal but some

10:03

of these major companies get deals with

10:05

Intel well they'll be getting like I-7s

10:07

for under $200 in bulk and they're

10:10

like the problem is actually assembling the

10:12

PC and we're worried about how much

10:14

we're gonna have to pay people to

10:16

fix everything I assume for you it's

10:18

like well I just you just don't

10:20

want someone on your in your studio

10:22

to have to lose a bunch of

10:25

work right. Honestly, if we lost all

10:27

the money on the hardware itself, it

10:29

wouldn't matter. You know, you just buy

10:31

new equipment. If the time and wasted

10:33

money and, oh, the fix is coming,

10:35

it's going to be the next one,

10:37

we swear, and there's five more fixes.

10:40

And it's that kind of thing where,

10:42

you know, in the game of Nexus

10:44

Video, Intel was, you know, handling the

10:46

payoff where he'd asked him a question,

10:48

and Intel would silently edit their post.

10:50

with a different answer. Oh yeah, I

10:52

forgot until literally applying. Sir, you're just

10:55

dealing with someone that's hostile at that

10:57

point. And why would you, why would

10:59

you engage with that? Right. So I

11:01

ended up stop engaging with them. I

11:03

ended up sending the laptops to some

11:05

reviewers so they could independently confirm it.

11:07

But, you know, we're kind of avoided

11:10

Intel. I'm kind of at the point

11:12

now where I want to consider sending

11:14

a mass email out to my customers

11:16

about the Intel thing next or at

11:18

least. If we detect a crash, we

11:20

send the customer any while and say,

11:22

hey, you know, this dispute is dying,

11:24

please fix it. It's funny, when I

11:27

just like, you know, 10 minutes ago

11:29

brought this up to you on the

11:31

show, I was like, oh no, I

11:33

might be under-prepared. I probably should have

11:35

looked around and, but it doesn't seem

11:37

like I have been under-prepared, really? Like,

11:39

it just seems like there hasn't been

11:42

progress. There's no mass messaging to customers

11:44

to say go up there. And a

11:46

lot of our means are getting rejected

11:48

still. So until some new story kind

11:50

of loads up about, you know, the

11:52

lawsuit or until, you know, someone or

11:54

whatever, we're still, we're still kind of

11:57

in the stalemate. It's in the best

11:59

interest of everyone to just move on

12:01

and don't buy Intel product again. It

12:03

just, that was a mistake. My server

12:05

provider said, hey Matt, these 1400Ks can

12:07

do six gigahertz in Bruce mode and

12:09

they're really awesome. But the fix knaps

12:12

performance, right? I remember Intel telling me,

12:14

he's like, hey, well, we'll make sure

12:16

the fix doesn't know performance. Oh man.

12:18

That sounds like months of like months

12:20

of like, working internal engineers, engineers, work.

12:22

I don't want to mess with that.

12:24

I don't want to mess with that.

12:27

Yeah, so I guess the final thing

12:29

I would say then is I would

12:31

I don't know what I can really

12:33

do I guess there's actually it's occurring

12:35

to me I just texted someone we

12:37

were talking but there are a handful

12:39

of people I could follow up with

12:41

to see if there is progress on

12:44

this how they would summarize it how

12:46

they would summarize it I don't know

12:48

if that would be a video that

12:50

gets a ton of clicks because unfortunately

12:52

the way new cycles work is people

12:54

pay attention hyperventilate for a month and

12:56

then it just like it just like

12:59

can never happen and people forget and

13:01

people forget it so I think it

13:03

would at least be worth an update

13:05

on some broken silicone though as a

13:07

portion there. And hopefully, you know, like,

13:09

gamers' nexus was really angry in that

13:11

video, but I hope they will follow

13:14

up and let window and other people

13:16

will too because, you know, again, just

13:18

texting someone and they're like, oh no,

13:20

we still get tons of returns for

13:22

after, like, it's like, I guess this

13:24

is just, it would be unfortunate of

13:26

how this ends is, tons of them

13:29

break, keep breaking and that's the end

13:31

of the end of the end of

13:33

the end of the story. but successfully

13:35

for Intel successfully buried and they just

13:37

have to kind of deal with our

13:39

like being a mess you know currently.

13:41

Yeah it's funny to think about how

13:44

they rushed out arrow lakes assuming this

13:46

would be the thing of it that

13:48

saves everything but now people are almost

13:50

more excited for Bartlett Lake or something

13:52

so all right so I thought we

13:54

had to talk about that in the

13:56

beginning there but as you can probably

13:58

see by the light changing behind me

14:01

I think we have to talk about

14:03

Blackwell here. I mean, it was, of

14:05

course, the main reason that we delayed

14:07

recording this episode. You know, we were

14:09

originally gonna record Sunday like usual when

14:11

I record and then it was like

14:13

well you know there's an AMD and

14:16

an invidia press conference maybe we should

14:18

wait and then I mean I think

14:20

that was a very very good idea

14:22

that we did because I mean there's

14:24

a lot to discuss here and I

14:26

don't even know where to start so

14:28

when it comes to the 50-90 the

14:31

50-70-T-I and the 50-70-70-70-70 what are your

14:33

thoughts? Yeah, I guess it's kind of

14:35

interesting with comparing it to Intel's keynote

14:37

and Andy's keynote before that because it

14:39

was a very strong contrast between them.

14:41

You know, the Intel one kind of

14:43

just mentioning AI, the AMD one kind

14:46

of cut off and didn't really mention

14:48

our DNA that much. Well, I don't

14:50

know if it mentioned it at all.

14:52

They just put out a separate press

14:54

release. Yeah, there was a weird Xbox

14:56

thing in there where they talked about

14:58

kind of. vaguely and they're like, you

15:00

know, Matt Booty is promising that our

15:03

DNA full is going to be the

15:05

best, without mentioning any of it. Even

15:07

though they don't even have an elite

15:09

console for Xbox, so why would you

15:11

bring on Xbox to talk about that?

15:13

The thought I thought was, are they

15:15

going to have FSR4 for Xbox or

15:18

something? And that's why this guy's here.

15:20

That would make sense if they did

15:22

that. It didn't seem like it there,

15:24

right? With envy, this keynote. the interesting

15:26

thing that was mentioned was a there

15:28

was a price anchoring that took place

15:30

of a $10,000 PC. Jensen talked about

15:33

how you're going to buy the 40-90

15:35

and it's going to go into this

15:37

$10,000 PC and the 40-90 is kind

15:39

of cheap compared to your $10,000 PC.

15:41

So this is really good value and

15:43

you're going about to upgrade it and

15:45

then we saw the 50-90 with the

15:48

increased price. Yeah, I mean I had

15:50

to put that in the video I

15:52

put out last night too. I mean

15:54

that was hilarious to me. I'm reminded

15:56

of I think her name was Lucille

15:58

Bluth or something, the grandma and arrested

16:00

development, there's this scene where they're like

16:03

losing money and she's like, but she's

16:05

always been this rich spoiled woman. and

16:07

she goes, well, how much can a

16:09

banana really cost? $10? And then her

16:11

son goes, oh my God, you've never

16:13

bought groceries in your life, have you?

16:15

And it's like $10,000 PC. My PC

16:17

has a 40-90. And it's not the

16:20

craziest thing. It's like, I have like

16:22

Gen 5 SSDs or 99, 9800 X3D

16:24

or something, but I think the total

16:26

PC is probably below three grand. I

16:28

don't think you're gonna get to 10

16:30

grand unless this is a workstation. for

16:32

work. But we have thread rib risu.

16:35

Yeah, I was like, this is thread

16:37

ribber. But even now I'm like, should

16:39

I really upgrade? I mean, I have

16:41

to pull my water cooling leaves a

16:43

heart. But yeah, so, so those specs

16:45

released, there is a concern I had

16:47

about the 50-80 kind of being cut

16:50

in half, 50-90 in terms of like

16:52

the RAM is cut in half, a

16:54

lot of the court counts reduced and

16:56

stuff like that the gap between the

16:58

50-19, 90 and the 50-80s increasing increasing.

17:00

And it continues to increase every generation.

17:02

I mean, just to remind people, I

17:05

mean, God, how far back do I

17:07

really want to go? Like, invidious has

17:09

been trying to do this for a

17:11

while. Have the 80 actually be quite

17:13

far away from the TI, the 90,

17:15

the Titan, whatever it is. Ever since

17:17

I would, I mean, ever since they

17:20

could get away with it, frankly, I

17:22

mean, all the way back, like, 20

17:24

years ago, you can find weird expensive

17:26

cards they made, but it. where they

17:28

had the 980 and the 970 with

17:30

the 104 die, and then later, actually

17:32

the next generation, they released the 102

17:34

die with the, or whatever it was,

17:37

I think it, 110 die, maybe, I

17:39

don't know what they called it back

17:41

then, with the Titan, and then a

17:43

780, and then to Maxwell, you know,

17:45

again, they had an ATTI that was

17:47

the top die, then they go to,

17:49

you know, Pascal, and they keep, they've

17:52

been continuing to shrink. the 80 class

17:54

smaller and smaller relative to the top

17:56

one for the most part and now

17:58

we're at a point where yeah I

18:00

mean, it looks like it's less than

18:02

half, I mean, even if it's the

18:04

full die, so less than a half

18:06

the top die. And so you, you

18:09

bet it better be, you know,

18:11

like, like half the price of the

18:13

5090. Yeah, it gave me this

18:15

vibe that the 5090 is

18:17

really like this tighten cart

18:19

compared to the cop a

18:21

lot, because the differences. I

18:24

did notice that there was

18:26

a huge amount of claims

18:28

being made, like crazy. This

18:30

uses 50% less power consumption

18:32

as in somehow in video

18:35

is cooked up some magic

18:37

and it's 50% less power usage.

18:39

We saw stuff like, was

18:41

it the 50-70 being the

18:44

same performance as a

18:46

49 or something like

18:48

that? Like incredibly crazy

18:50

claims made and there was

18:52

a lot of, I guess, kind

18:54

of demo video footage of, you

18:56

know, DLS. It really looked like

18:59

a cinematic shot and then it

19:01

kind of turned into like a

19:03

game game clean play footage and there

19:05

was kind of crazy frame rate numbers

19:07

of on the left that have 30

19:10

FPS on the right that have 300

19:12

FPS or something crazy. Yeah, I'll probably

19:14

have Gerard show that video on screen

19:17

here as well because like when the

19:19

conference started, you know, All right, let's actually

19:21

get the weird stuff out of the

19:24

way first. I mean you mentioned the

19:26

there's the $10,000 PC thing, which tells

19:28

you Jensen's at a level of money

19:30

where there's no clue what things cost

19:32

anymore that are outside of his immediate

19:35

view. You had him weirdly walking on

19:37

stage with a shield. For me, the

19:39

weirdest thing actually was the first five

19:42

minutes where. A guy, like one of his

19:44

lackeys wats out and just praises him

19:46

as a guy who worked at Denny's

19:48

and he's got faith and drive and

19:50

it's like, it's like, wow, this is,

19:52

we're just starting with ego here. He's

19:54

like, he was a dishwasher and he

19:56

came from nothing and now he's a

19:58

super genius and it was. it was

20:00

a C.S. event organizer and this is

20:02

a thing that was super weird for

20:05

me because he was a C.S. person

20:07

not in video that makes it even

20:09

weirder because that means it kind of

20:12

like and I get this at G.D.C.

20:14

with game developers where depending on how

20:16

much money you pay you get a

20:19

different size booth right and you get

20:21

a different time slot. And I'm wondering

20:23

if invady has paid the most money

20:26

to see yes to be lost to

20:28

have the most time with his trying

20:30

stage and you have your CS lackey

20:32

come on and say that you're a

20:35

genius and you know those are intro

20:37

videos like invidious curing cancer and amusing

20:39

and that's how much I have security

20:42

it but yeah we're always curing it's

20:44

never happened though but like yeah like

20:46

let's talk about it when you do

20:49

it okay let's just cut to that.

20:51

when you actually accomplish it with your

20:53

AI. And that was a jacket, which

20:56

is, you know, interesting. When you see

20:58

stuff like that, you know, this is

21:00

a, just saying people, like, when you

21:03

see stuff like that, that looks that

21:05

silly, that means no one around Jensen

21:07

tells him no anymore. That's what that

21:10

tells you. So we're getting full ego

21:12

here, you know, going into the show.

21:14

None of that surprised me. There's just

21:17

more than I've ever seen before. They

21:19

start revealing the 50-90.90. The specs. what's

21:21

been leaked, nothing surprising to me. It

21:24

honestly does look very impressive. The first

21:26

red flag, yeah, 50-70 is strong as

21:28

a 40-90, but they don't tell you

21:30

how much V- RAM it has. That's

21:33

when the alarm bells just went crazy

21:35

for me. I'm like, no, it's not,

21:37

man, it isn't. And if you look

21:40

at the charts, and I analyze this,

21:42

like I actually measured their own bar

21:44

chart on their website and where's the

21:47

halfway point, as far as I can

21:49

tell the 50-70 in the one game

21:51

that they hand selected, mind you, that

21:54

still had some ray tracing turn on,

21:56

so we don't have a pure raster

21:58

comparison. It looked 20 to 35% faster

22:01

than the 4070, which places are, yeah,

22:03

where I was told, around a 4070

22:05

TI, maybe a little stronger. That's not

22:08

a 4090. The problem with that one

22:10

game that we have that's technically retraced

22:12

is they could have just upgraded the

22:15

retrace. Yeah, fuckrace. They could have just

22:17

upgraded the retrace in core and we

22:19

wouldn't really know a good reflection of

22:21

normal roster performance. Right. And I noticed

22:24

a laptop pricing looked pretty reasonable. The

22:26

price was for the whole laptop and

22:28

not the GP or anything so stuff

22:31

started immediately not adding up to me

22:33

whereas like power consumption 50% fast off

22:35

You know this thing's a 40-90 now

22:38

That didn't really add up and I

22:40

was looking at the the language Jensen

22:42

was using and he'd say stuff like

22:45

this wouldn't be possible with a out

22:47

AI and stuff like that so I'm

22:49

more getting the impression that DLS is

22:52

doing some heavy lifting. There is fake

22:54

frame generation happening as well with three

22:56

fake frames and one upscaled frame. So

22:59

I'm kind of concerned about the numbers

23:01

looking promising, but the actual context of

23:03

it being not really a huge upgrade.

23:06

Yeah, I mean again, like when they,

23:08

I don't even, I didn't even write

23:10

down half the claims because I know

23:12

they're all nonsense, but like you say

23:15

that they said like at half the

23:17

power or something, it's like, well, yeah,

23:19

I mean, but how are you counting

23:22

the frame? Is the 40-90 running without

23:24

the extra frame generation and then you're

23:26

camping at 240 hertz or something? Because

23:29

if that's what you're doing, it's using

23:31

half the power because you're generating half

23:33

the frames, you know. Yeah, if the

23:36

GPU stops doing stuff and waits for

23:38

the next rain to happen, it uses

23:40

less power. Like, I don't understand how

23:43

these claims were impressive, unless they're in

23:45

context that explains why the power, because

23:47

on a 49, you could log power

23:50

of consumption by 50% just doing some

23:52

changes to settings, right? So that's probably

23:54

the most concerning thing, and I'm seeing

23:57

a lot of benchmarks and posts about

23:59

comparing. frame gen with non-frame gen results

24:01

and saying, look how good the cut

24:03

is now when we know there's fake

24:06

frames don't take that much to produce

24:08

from the compute standpoint and it's very

24:10

misleading on the numbers and the fact

24:13

we only get one cherry-picked game that

24:15

has RT which is quite is concerning

24:17

to me. They could have had a

24:20

whole slide with Raster games as well

24:22

but that just was not a thing.

24:24

It's intentionally not true. Yeah and and

24:27

so again it's like I could... piece

24:29

apart all the specs. But I mean,

24:31

we just need to see third party

24:34

reviews. And this is no different than

24:36

before. It seems to be, well, I

24:38

guess I don't know actually, we're gonna

24:41

have to give it a week. This

24:43

was yesterday, I went to sleep at

24:45

like. five a.m. Actually, I think I

24:48

fell asleep at four a.m. on the

24:50

couch. Then my girlfriend woke me up

24:52

while she was going to work and

24:55

I went upstairs and went to bed

24:57

for another few hours. So it's not

24:59

like I've had the time at all

25:01

since this press conference that was at

25:04

night yesterday and started late to really

25:06

look at how people are thinking about

25:08

this. All I can say is when

25:11

it first happened, it feels like the

25:13

reactions I saw on like, read it,

25:15

the Discord, YouTube comments. The initial reaction

25:18

seemed to be taking Invidia at their

25:20

word more than two years ago, but

25:22

I think we have to remember two

25:25

years ago. Invidia claimed that the 47

25:27

ETI, which at first they called the

25:29

4812 gigabyte, would be four times faster

25:32

than a 3080 TI. It was 10%

25:34

faster. And I don't know. I mean,

25:36

so when I see them claim of

25:39

5070 is two or three or whatever

25:41

they say times faster than a 4070.

25:43

it might be 20 to 30% faster.

25:46

And I think that's kind of where

25:48

we have to leave it besides, you

25:50

know, this unknown of if their neural

25:52

compression, which I have a picture draw

25:55

I can put up as well, where

25:57

they like compress something, I think it

25:59

was, actually let me just open it,

26:02

since I can put it up for

26:04

myself. So they have an example where

26:06

they compress some texture that's 4096 by

26:09

4096 from 256 megabytes to, I think

26:11

that's 3.8 megabytes. Okay, yeah, I mean,

26:13

that's a lot of compression. I can

26:16

actually tell it though, it doesn't look

26:18

lossless at all, although it's a small

26:20

part of a texture, I guess. So

26:23

they might argue you don't notice. You

26:25

know, they say like neural face or

26:27

AI face. I don't even know what

26:30

that means. Like they had all these

26:32

different ways that. you know, AI is

26:34

going to compress and enhance the graphics

26:37

without actually taxing the GPU. But we

26:39

don't know if that will just work

26:41

or not, how many games that will

26:43

work in. And again, like, I think

26:46

people need to remember that Starfield, which

26:48

didn't have DLS right away, there was

26:50

a hack to add DLS. It added

26:53

a lot of artifacts, and that's what

26:55

happens. Dev's do, and you're developers, this

26:57

is where I kind of want you

27:00

to chime in your eventually like. You

27:02

do have to do your own work

27:04

to make sure DLSS and FSR don't

27:07

have artifacts, right? Even though invidia acts

27:09

like it just works. Yes, the concern

27:11

for me was, there was compute shaders

27:14

that was mentioned about compute shaded doing

27:16

machine learning. We haven't really seen how

27:18

many games are going to add that

27:21

in and how cod that is to

27:23

support. But it looks interesting, but. without

27:25

the context or seeing a game where

27:28

that it's kind of hard. For the

27:30

upscaling and frame gen and DLS stuff,

27:32

there's kind of a split the industry's

27:35

going in where one side is like

27:37

upscale, we don't care about artifacts or

27:39

noise, we're going to render a 1080P

27:41

and upscale to 4K, you know, super

27:44

sample everything. At the other side of

27:46

the traditional rendering side is image quality

27:48

as king, we're going to render in

27:51

4K, everything to look perfect. We don't

27:53

need any anti-aliasing. And there's a huge

27:55

split and invidious led the industry with

27:58

it's okay if we do retracing because

28:00

we're going to downscale the resolution by

28:02

50% to make the retraceable. And oh,

28:05

it's okay if we do, you know,

28:07

DLSS, because we're going to add an

28:09

video reflex and that's going to fix

28:12

a latency. And we've seen stuff like

28:14

reflex two, which is going to preemptively

28:16

predict where the next frame's going to

28:19

be, where your mouse is moving, which

28:21

is a little bit weird from my

28:23

standpoint, because unless you waste a whole

28:26

bunch of time retracing stuff, your input's

28:28

going to be really quick. So if

28:30

you need to make four fake frames,

28:32

for example, your input latency is going

28:35

to be really high. So they're going

28:37

to make another fake frame that has

28:39

the camera in a different spot to

28:42

make the input latency less. It really

28:44

gave these Google Stady on negative agency

28:46

vibes. Where I'm like... I forgot about

28:49

that. Yeah. This sounds like some kind

28:51

of... I'm sure in some cases it

28:53

might be cool. But your game has

28:56

to be very wasteful on resources to

28:58

need the latency fixed thing from Mvideo.

29:00

And it's kind of like this house

29:03

of cards where at the bottom you

29:05

have a 1080P render and then we

29:07

add retracing which loads it down and

29:10

then we add DLS to try to

29:12

get the performance back and then we

29:14

add reflex or reflex two and then

29:17

we need this neural network to downscale

29:19

the textures because the performances are bad.

29:21

And I'm wondering at what point people

29:23

realize, oh, this isn't bad. This is

29:26

a good for image quality. You know,

29:28

there's artifacting, this blur. This is a

29:30

bad direction to get in. But because

29:33

AMD and Intel are so far behind

29:35

and they're copying invidious features, right? If

29:37

the invidious got this black well pulling

29:40

people in, it's like, games already added

29:42

retracing, so your be your GP better

29:44

support. faster reach. You better have DLS

29:47

and you better use it or ray

29:49

tracing is going to give you really

29:51

low frames unless you're using DLS. So

29:54

and all these features if you use

29:56

if you use DLS by itself on

29:58

a non-ray trace game. your performance is

30:01

actually lower in a lot of cases

30:03

because the game isn't bottlenecked. Only when

30:05

you bottleneck the game is the performance

30:08

increased when you're using DLS. I noticed

30:10

that with some games with PSSR as

30:12

well. So you'll, you know, I'll be

30:15

in part of Excel 2, I'll be

30:17

like, okay, let's upscale from 10-8-4-K and

30:19

my FPS is down versus rendering natively

30:21

in 4-K. So like, what's the point

30:24

of using DLS-I add retracing? and really

30:26

slow the game down. Right? Yeah. Maybe

30:28

people really like retracing and don't care

30:31

about image quality, in which case this

30:33

GP was amazing and great. But if

30:35

you care about image quality and this

30:38

stuff makes you sick and your input

30:40

latency is really bad and you don't

30:42

like some of this rendering stuff, this

30:45

is not going to be the great

30:47

direction to move forward. And it really

30:49

depends how many game has noticed or

30:52

care. or you know, experience motion sickness

30:54

from this, the stuff, right? Yeah, I

30:56

mean, I, we've already seen with like

30:59

the 4090 versus the 7900xDX that obviously

31:01

on average overall, it's, you know, a

31:03

lot faster. I mean, it can be

31:06

over 30% faster, it's like 15 to

31:08

20% faster unless you crank up ray

31:10

tracing. But, you know, there are games

31:12

where the XXX matches or beats the

31:15

4090, depending on how the game chooses

31:17

to render things, and it just. Amy

31:19

doesn't have a competitor of the 50-90

31:22

this time, but it's, I really, it

31:24

would not surprise me if there are

31:26

some games, you know, especially ones heavily

31:29

influenced by invidia, that use a ton

31:31

of ray tracing, DLS, and all these

31:33

features, and the performance looks incredible. And

31:36

then there's games where it only beats

31:38

the 40-90 by 20 or 30 percent,

31:40

like, which. It's crazy to me because

31:43

it has like what double the bandwidth

31:45

or like almost like it's got an

31:47

absurd amount more bandwidth and cuda cores

31:50

but again to the people that think

31:52

I'm acting crazy here I mean I'm

31:54

looking at a chart that in video

31:57

has provided themselves of the 50-90 over

31:59

the 40-90. and it looks like it's

32:01

like 30% faster in far cry

32:03

six than the 40-90. That's really

32:05

really low to me even just based on

32:08

the specs on paper, but they chose

32:10

to show that game, which is odd.

32:12

So it could be all over the

32:14

place depending on the game you're playing.

32:17

So the weird thing is that's

32:19

their cherry-picked game that might be

32:21

the best case scenario. So are

32:23

we going to see some games

32:26

where it's a 5 or 10%

32:28

15% performance uplift in raster performance

32:30

for 49D and it's so much more

32:32

expensive that it's like what

32:35

is the point of buying this

32:37

for gaming at least like unless

32:39

you're doing the retreecing and

32:41

unless you're doing all these

32:44

effects that are better on the new

32:46

GPU. This piece of content is brought

32:48

to you by CD Key Offer.com.

32:50

Just in the past few weeks,

32:52

Dan built a new PC with

32:54

a fresh install of Windows 11

32:56

on a blazing fast Gen5 and

32:58

VEMESD. And of course, that code

33:01

for Windows 11 came from CD

33:03

Key Offer.com. And we've been using

33:05

CD Key Offer.com at Moore's Law

33:07

and for a very long time.

33:09

I use it for my office

33:11

software on both my desktop and

33:13

my... laptop, and that's because they're

33:15

reliable. They've always been there for

33:17

us and they offer reasonable prices

33:20

for Microsoft products that are usually

33:22

set in my opinion at monopolistic

33:24

pricing. And you know, they're actually

33:26

having a Christmas. super sale right

33:28

now. So if you're looking to

33:30

get the best pricing on Microsoft

33:32

software or games or other products

33:34

on their website as well, please

33:36

show support for a company that's

33:39

shown support for the Moore's Law

33:41

is dead team in community for

33:43

years by using offer code broken

33:45

still on to get 25% off

33:47

all Microsoft software and die string

33:49

to get 3% off everything else. To

33:51

support Moore's Law is dead by checking

33:53

out CDK Offer.com today. Um.

34:00

And I guess you kind of answer

34:02

it, but I just want to try

34:04

to get a concrete one to make

34:06

sure it is answered. Like this neural

34:08

compression and the way they're going to

34:10

predict the next frames, what's your gut

34:13

reaction to how well that will work

34:15

though? Because on paper, they could argue

34:17

that an eight gigabyte card is like

34:19

a 24 gigabyte card now. I just

34:22

have a hard time believing that's going

34:24

to work. The compression thing looks interesting,

34:26

but for... For our games, we already

34:28

use like Google texture compression. It's already

34:30

really good. So I wonder if that

34:33

compression comes in more on rate tracing

34:35

workloads and it's not as suitable on

34:37

other workloads. The whole running out of

34:39

V RAM thing people keep going on

34:42

about is I turn on retracing and

34:44

retracing destroys my V RAM and I

34:46

run really high texture quality textures and

34:48

you just run out of space. So

34:51

it looks like it could be good.

34:53

But on the other side, the latency

34:55

in a video reflex too, or what

34:57

it was called, it concerns me that

34:59

they're like, hey, we need to predict

35:02

where the camera's going to be next

35:04

frame, because our frame takes 50 milliseconds

35:06

to render. Like I'm used to making

35:08

games where at 60 APX, it's 16

35:11

milliseconds, and there's not a lot of

35:13

delay, right? And how accurate is this

35:15

thing going to predict where the next

35:17

frame is? If you're turning the corner

35:20

and there's a bad guy there, the

35:22

machine learning is not going to know

35:24

the person there. It's going to get

35:26

a bunch of pixels that may or

35:29

may not be there. Do people really

35:31

want a competitive shooter where they turn

35:33

around a corner and maybe... Like it

35:35

reminds me of this text where they're

35:37

using satellites to kind of scan the

35:40

earth, you know, to figure out to

35:42

map it. Right. And there's two companies.

35:44

There's a company that's like we're going

35:46

to... use an algorithm to predict what

35:49

the pixel could be accurately, and there's

35:51

an AI company saying we're going to

35:53

run machine learning and have it guessed.

35:55

And it will put planes in there

35:58

where there's no planes in the images.

36:00

And if this is a military contract

36:02

company, do you really want a

36:04

fake plane in your image, right? So,

36:06

I mean, no, people have probably seen

36:08

footage of like helicopter gun, gunners.

36:10

And it's like the most grainy, horrible

36:13

looking camera you've ever seen. But it

36:15

is accurate. And that's all they want.

36:17

So, you know, if there's a thing

36:19

that and I guess this also depends

36:22

how many frames are you're guessing, like

36:24

a lot of people don't know.

36:26

And media has a setting in the

36:28

driver. 16 frames of their warning, but

36:30

they've set it to be 5, right?

36:33

And generating a new frame isn't that

36:35

impressive, compared to rendering it normally.

36:37

Like it's hard to use it as

36:39

a comparison. And I think depending on the

36:41

play as personal view, if frame gen is

36:43

not a issue for you, you're going to

36:45

like the deep view. But as a developer,

36:47

that I want to control everything

36:50

on the frame. And if there's a bug,

36:52

I want to make sure that the, if

36:54

there's flickering in the corner somewhere somewhere

36:56

somewhere somewhere somewhere, Am I

36:58

going to go to one of my

37:01

developers and say, oh, that's just the

37:03

Nvedi thing. Good luck fixing that

37:05

bug. It's in a model where

37:07

you can't edit or control. And

37:10

that's something I wanted to touch

37:12

on too. Like, as a developer,

37:14

does this worry you? Because I

37:16

think, does this worry you, because

37:18

I think, again, already, like, this

37:20

is, I've had you on Brian

37:22

Heemskirk, someone who both worked at

37:25

Sony Santa Monica and I believe

37:27

now, Very big games, some of them,

37:29

all of that on Unreal Engine Devon.

37:31

One thing that comes up from

37:33

you and every developer I speak

37:36

with is how often gamers are

37:38

like, why doesn't this have FSR

37:40

or DLSS? You know, because it's

37:42

not just DLS, it's like why

37:44

doesn't Metro Exodus have FSR support?

37:46

Why does it? And it's like

37:48

the developer has their game,

37:50

they've finished it, and they're worried

37:53

if they add another one, it will

37:55

break it. Right. It almost feels like this is

37:57

a Pandora's box of things you're going to have to

37:59

worry about. How much is that, were you?

38:01

Yeah, so with us, we upgraded

38:04

to a real 5.5 recently, and

38:06

our committee is like, what do

38:08

you mean you change your engine?

38:10

Where's our Nanite, Lubin, Megalites, DSSR,

38:12

ifSR, XESS, reflexes, reflex two neural

38:14

temporal material thing? Like my God,

38:16

that sounds like there's 30 features

38:18

I need to add that are

38:20

constantly moving that might break the

38:22

game, that MD might break in

38:24

a driver update next week. That

38:26

sounds really complicated and if we

38:28

add in DLS and our frame

38:30

rate doesn't go off because we're

38:32

not using rate tracing, it's going

38:35

to be pretty pointless, right? That's

38:37

a concern. So game is, it's

38:39

not really communicated in the marking

38:41

material what this frame gen thing

38:43

actually is and how to fix

38:45

your game very well. And no

38:47

one really understands, like the frame

38:49

rate kind of goes up, that

38:51

must mean it's better. If we

38:53

added a button to the game

38:55

that just shared a higher number

38:57

on the EPS counter without it

38:59

doing more EPS, some people would

39:01

say, oh, my game's better. The

39:03

number says 200 instead of 120

39:06

now. If the game actually better

39:08

there. And as a developer, it

39:10

does feel like someone coming into

39:12

your game and messing with the

39:14

pixels and frames and adding problems

39:16

and bugs that weren't there before.

39:18

And if I have to beg

39:20

and video, please fix your model,

39:22

I have this one bug, one.

39:24

one area that's closing people to

39:26

have seizures. Am I going to

39:28

get support for him? I don't

39:30

know. It sounds like a mess

39:32

to me. Yeah, so, and you

39:34

would agree, we just don't know.

39:37

We don't know if there will

39:39

be issues. We don't know how

39:41

much of this will. It just

39:43

works. You know, we don't know

39:45

how much of this is going

39:47

to require to have input until

39:49

we have review day. you know

39:51

and we're watching hardware and boxed

39:53

or something or gamers nexus and

39:55

they're like these are the five

39:57

or hopefully all like 100 games

39:59

that actually already support it there's

40:01

no issues or if that's not

40:03

going to be the case but

40:05

you know again what I can

40:08

say about performance is when I

40:10

look at the action chart hand-picked

40:12

by and video, the hand-picked game,

40:14

the one game without like full

40:16

ray tracing or one of the

40:18

only full games is far cry

40:20

six from three years ago and

40:22

it looks like most of these

40:24

cards are 20 to 40% faster

40:26

than their predecessors, you know, not

40:28

doubling, which again makes me two

40:30

minds about it. On the one

40:32

hand that's within what I expected,

40:34

although on the lower end for

40:36

some of them, and so I'm

40:39

like, well, okay, so it's on

40:41

the lower end of the expected

40:43

performance. the 50-90s pretty much exactly

40:45

where I expected it to cost

40:47

50-70-50-70-T-I slightly cheaper, 50-80 is cheaper

40:49

than I expected it to, although

40:51

if you look at the memory

40:53

speeds and stuff, it kind of

40:55

seems like they use cheaper components

40:57

than they might have been considering

40:59

and they're making it possible to

41:01

be, but if DLS4 is not

41:03

all it's hyped up to be,

41:05

It doesn't really impress me more

41:07

than Lovelace did. If it's black

41:10

magic, then sure, this seems crazy,

41:12

but it would have to be

41:14

black magic. And I hope that

41:16

it is for gamers, but also

41:18

just like, if what Jensen just

41:20

showed on stage, none of it

41:22

happens. I think we're at the

41:24

peak of the AI bubble. Like,

41:26

that's what tells me, all right,

41:28

like you said, there's a Google

41:30

Stadium presentation that wowed everyone that

41:32

didn't happen. Like, the guy walked

41:34

out in this jacket that no

41:36

one told him looks ridiculous because

41:38

no one tells him no anymore.

41:41

He held up a shield and

41:43

acted like a gladiator and then,

41:45

you know, pets.com. Like, that's what

41:47

I'm gonna say happened if this

41:49

doesn't deliver as well as they

41:51

say it. It is hard for

41:53

me to see how the competition

41:55

catches up to this any time

41:57

soon. I'll say that. I'm sure,

41:59

I'm sure that it might be

42:01

a subset of stuff that get

42:03

delivered like 25% of the stuff

42:05

he says, or 50% of the

42:07

stuff he says, but without marking

42:09

it in the context of the

42:12

reason why it uses 50% less

42:14

power. is because of the LSS.

42:16

And without the LS, it doesn't

42:18

use us panel. But we're going

42:20

to find all those answers out

42:22

in the reviews. All right, so

42:24

quick jumper rights in and says,

42:26

what is the current competitive landscape

42:28

in your opinion between DLS3, FSR,

42:30

I suppose XC Super Samling 2,

42:32

in terms of game adoption and

42:34

implementation quality? With FSR 3, reportedly

42:36

reaching over 50 supported games. How

42:38

does DLS3's adoption compare? And what

42:40

patterns do you see emerging in

42:43

real-world quality of frame generation implementations

42:45

between these technologies? Sir, the thing

42:47

I like about... DLS 3 and

42:49

FSR 3.1 is supposedly DLS 4

42:51

and FSR 4 will be a

42:53

in place upgrade without a loaded

42:55

developer work. Like that sounds great,

42:57

that sounds very future proof. I'm

42:59

pretty disappointed that they had this

43:01

whole MD streamline thing that was

43:03

supposed to be a code once

43:05

and then magically your game will

43:07

support all three pipelines. But it

43:09

seems like... Intel and Tom Peterson

43:11

does not like streamline and you

43:14

know didn't bother integrating it with

43:16

invady or getting it to work

43:18

just disagreements. So the thing that

43:20

I hate the most is I

43:22

have to add three different GPU

43:24

vendors code into my game using

43:26

three different sets of plugins. These

43:28

vendors are not unreal engine experts

43:30

like they don't go to plug-in

43:32

that is unreal engine perfect. It

43:34

is And video doesn't know how

43:36

to use Unreal, so they made

43:38

a plug-in that might work or

43:40

might not work. That's kind of

43:42

the vibe I get from more

43:45

theory of them. So there's different

43:47

bugs and code standards between all

43:49

three plugins. And sometimes they fight

43:51

with each other too. So we'll

43:53

put the DLS plug-in and it'll

43:55

break PS-5, even though our PS5

43:57

doesn't use DLS. Of course, yeah.

43:59

Which you're like trying to talk

44:01

to an DVD and you're like,

44:03

hey, peace. Yeah. But in terms

44:05

of the future sets, I think

44:07

the LSS is getting a lot

44:09

better. I think FSR was nice

44:11

because it was super open, but

44:13

I'm concerned about it being closed

44:16

off now. It was a closed

44:18

off thing. And Intel released Battle

44:20

Mage, marketing XESS2. It's in one

44:22

game so far that. not running

44:24

it through for well and there's

44:26

not released to developers. And I

44:28

emailed Intel about it and they

44:30

said it was a launching in

44:32

January and we still haven't heard

44:34

about it yet. So I'm generally

44:36

concerned about Intel not following up

44:38

on their promises. And not to

44:40

mention there's a whole bunch of

44:42

comments made by Intel about it

44:44

being open-source and open platform and

44:47

they're going to support it for

44:49

everybody and that kind of disappeared

44:51

and doesn't run on Linux and

44:53

stuff like that. So it's kind

44:55

of. kind of concerning. Yes, and

44:57

that was something that stunned me.

44:59

I believe that was actually one

45:01

of the things we were talking

45:03

that made me say, hey, maybe

45:05

you should come on again soon

45:07

is, you're saying XE Super Samling

45:09

too, like Intel isn't talking to

45:11

developers at all about it. Yeah,

45:13

so there's one game that supports

45:15

it, like it's been what a

45:18

month since Battle Mage came out?

45:20

There's the so-called launch and it's

45:22

been the beast. Was it 870

45:24

or B5? What was it? The

45:26

one, the one was from that

45:28

B570? No. Oh, there is a

45:30

B570 that should launch in a

45:32

week. So that's launching a week

45:34

from now and we still haven't

45:36

seen XES2. No one's adding support

45:38

for it for games except one

45:40

game. I don't think that's going

45:42

to get a lot of adoption.

45:44

If AMD is struggling with FSR

45:47

adoption Intel is over in the

45:49

corner eating glue over here and

45:51

I'm very concerned that This is

45:53

going to be a mess to

45:55

maintain of three different plugins. I've

45:57

got to go buy every single

45:59

GPU now like I have to

46:01

go buy a 50-90 over add

46:03

support for whatever crazy thing Jensen's

46:05

booked up for every GPU vendor

46:07

This is a crazy amount of

46:09

support like it's not like I

46:11

implemented direct X12 API and all

46:13

these features work for all the

46:15

GPUs or all the ones that

46:18

support automatically It's it's kind of

46:20

like Android development where I need

46:22

to go by I have this,

46:24

uh, I guess I don't have

46:26

you see it, this like rack

46:28

of currency. Oh yeah, I do.

46:30

Yeah. In a wood box here,

46:32

where I've bought like 50 Android

46:34

phones and I have to go

46:36

test every single one. This snaptragged

46:38

one supports this, this other ones

46:40

about this. This one does retracing,

46:42

this doesn't, and it's a maintainability

46:44

mess. So I think the standard

46:46

that will win is not the

46:49

one that looks the best. But

46:51

the one that is most open

46:53

to developers and most supported, and

46:55

if I email an video next

46:57

week, am I going to get

46:59

a reply from the disabled saying,

47:01

we're here to support you on

47:03

adding this new feature among, right?

47:05

Probably more likely than Intel emailing

47:07

you back. Intel, I emailed them,

47:09

I'm thinking, man, are these guys

47:11

going to impede me because I

47:13

called them out on the reptile

47:15

thing? Am I even going to

47:17

get a reply? I even said,

47:20

hey, look, I'll sign an NDA.

47:22

I'll add X ESS to, you

47:24

know, no, Roseabout it. No, I

47:26

can't get access to it right

47:28

now. So it's kind of kind

47:30

of wrong. I mean, this is

47:32

the part where I mean, I

47:34

ask like, how, I mean, how

47:36

different is it to support Battle

47:38

Mage over? I mean, I mean,

47:40

I mean, I mean. not even

47:42

just in video, right? There's Blackwell,

47:44

Lovelace, Ampere, and then Turing's really

47:46

entirely different from Ampere. Turing is

47:48

certainly entirely different from Pascal, which

47:51

you still have to support, of

47:53

course, like how much work goes

47:55

into not just supporting Battlemage versus

47:57

Invidia, but also supporting five-in-video architectures.

48:00

Yeah, so the hard bit from the

48:02

gaming side is, unless and video and

48:04

these GP vendors go out to game

48:06

developers and say, hey, we're going to

48:08

send you money, we're going to give

48:10

you support, we're going to add, please

48:12

add these features. Like, adding a feature

48:14

that just works on the 50-90, for

48:16

example, would suck, because only a small

48:18

percentage of people buy 50-90s, right? So

48:21

they have. That's what they're going to

48:23

do again. We're kind of betting on

48:25

what kind of adoption of adoption

48:27

of a GP. If Intel sells a

48:29

hundred battle-made GPS, is it

48:32

worth supporting it? That'd be a

48:34

big month for them. Right? That is

48:36

the question. Now, I would like to

48:39

see as a developer, I want to

48:41

support everything. I want to support everything,

48:43

even if it's not popular, because I

48:45

wouldn't want someone to play my game and

48:48

have a bad time. But there is an economics

48:50

factor of... You could just support

48:53

invidia, and you've got 73% of the

48:55

steam hardware survey. It's really hard to

48:57

justify doing the others. We kind of

48:59

do them because we care, not because.

49:01

And so if you're like half a

49:04

percent, although does Battle Mage being in

49:06

Lunar Lake make a difference? Although I

49:08

guess I have to admit, Arrow Lake,

49:10

which is probably the overwhelming majority

49:13

of laptop CPUs will be actually,

49:15

I believe is meteor, like what

49:17

is that? So then alchemus plus

49:19

or alchemis plus. It's some variant

49:21

of an enhanced alchemist that doesn't

49:23

use Battle Mage. But like does.

49:25

So actually a battle mage might be like

49:27

the least ubiquitous Intel gen ever

49:29

you think about it because I bet celestial

49:31

is in a lot more APU is than

49:33

battle mage will be. But like there's battle

49:35

mage I guess being in lunar lake and

49:38

were it to have been an arrow like

49:40

would that have meant oh well we you

49:42

know there's no question we're going to support

49:44

this a lot or is the difference between

49:46

like. Like alchemist and arrow like an

49:48

alchemist media like an alchemist on desktop. Are

49:50

those different enough? Or no, it's not that

49:52

simple of like, but we have to support

49:54

it because it's an Intel's integrated graphics which

49:57

are everywhere. The dedicated one really is its

49:59

own beast. Yeah, so they're probably ran

50:01

into with Battle Mage and some

50:03

of the other Intel IGPUs is

50:05

a lot of our customers will

50:07

show up with a 6-gen Intel

50:09

CPU or a, you know, 10-gen

50:11

Intel CPU even. And Intel abandoned

50:13

the drivers for them, right? So

50:15

we get like thousands of users

50:17

a month, for example, they show

50:19

up, they don't get any individual

50:21

drivers. Some people have a CPU

50:23

that is two years old. and

50:25

the GP driver is already abandoned

50:27

by Intel, right? So there's a

50:29

bad track record of this. You're

50:31

pointing that out to me too,

50:33

behind the scenes, because you saw

50:35

a lot of tech tubers for

50:37

some reason going, oh, because Tom

50:39

Peterson, you know, he said it

50:41

was going to last a long

50:43

time, who was never like before,

50:45

by the way, said that it's

50:48

going to get a lot of

50:50

support, we're just going to believe

50:52

it, even though Intel's canceled drivers

50:54

after two years. Yeah, so for

50:56

Ross, when we see a driver

50:58

canceled, the game runs at, you

51:00

know, 45 FPS on an Intel

51:02

IGPU, and it's perfectly fine. The

51:04

only thing is Intel abandoned or

51:06

driver, and now the driver crashes.

51:08

And so we go to get

51:10

support for that, and they just

51:12

say, hey, it's abandoned. Right? You

51:14

can imagine the most popular thing

51:16

that's released that's the most available

51:18

is abandoned. How the hell do

51:20

you think battle majors get a

51:22

go? Right? Do you like to

51:24

have to hope that like if

51:26

I spend, you know, developer time

51:28

adding battle major point, is Intel

51:30

going to help me and is

51:32

it going to work? And the

51:34

issues I run in battle majors,

51:36

I'll pull out my GPU and

51:38

the, you know, connect a Xbox

51:40

controller and the Y button for

51:42

Xbox will be the wrong color.

51:44

And I'm like, how does it,

51:46

how does a 2D image, like,

51:48

let's say the A button's green,

51:50

on Battle Major's just like, like,

51:52

Orange, right? And you're like, how

51:54

does this 2D image of a

51:56

button that we're just showing as

51:58

a button prompt? It completely... the

52:00

Lauren color and Burkin because of

52:02

the driver. And you're like, how

52:04

much else in the game is

52:06

going to be like this flickering?

52:08

There's all stuff to lead problems.

52:10

Right. So you're opening to support

52:12

battle major opening up a can

52:14

of worms of like all of

52:16

these tickets, you're going to have

52:18

to open with Intel. That's like

52:20

thousands like, you know, and getting

52:22

one ticket like we have a

52:24

bug with the DVD where they

52:26

have shadows. Flicko. Since we've operated

52:28

on religion five. and would be,

52:30

let's say a bug in the

52:32

driver, we've fixed it on AMD

52:34

because AMD's drive is a lot

52:36

more open, but the invidious driver

52:38

is a very locked down. And

52:40

so if I'm looking at these

52:42

tickets like, okay, battle major is

52:44

going to be, you know, months

52:46

of work, and I'm going to

52:48

have to open up 200 tickets

52:50

with Intel, and Intel is not

52:52

the best of communicating. And we

52:54

could spend like a single ticket

52:56

could take two months to fix

52:58

for one developer who's been doing

53:00

this with 10 years, right? Unreal

53:02

engine does not care about battle

53:04

age, right? So you have this

53:06

problem where you're now going uphill

53:08

against like fixing epic stuff on

53:10

to work with battle range. And

53:12

you know, I play, you know,

53:14

I have a B580 and I've

53:16

run games like disonid. And I'm

53:18

seeing artifacts in that game, right?

53:20

It's like a multi-year-old game in

53:22

2016. I've run a game, disordered

53:24

from 2013 or whatever it is,

53:26

and it's broken on battle age.

53:28

And I'm seeing these hardware and

53:30

books video saying, performance on all

53:32

CPUs is really bad. And you're

53:34

thinking, man, I kind of regret

53:36

buying this thing. This thing's a

53:38

mess. Like, all the reviewers had

53:40

positive reviews on this on this.

53:42

Well, I want to say this

53:44

too. It's not just on older

53:46

CPUs. You need a 98. 800

53:48

X3D to consistently be to 4060.

53:50

Like we're not, we're talking, and

53:52

this is the most eye-opening for

53:54

me, like you can go to

53:57

Zen 4 and there's games where

53:59

there's just 20% performance swings. And

54:01

what's gonna happen when next-gen games

54:03

come out with higher CPU demands?

54:05

Like I think we're gonna start

54:07

to see soon, some games on

54:09

PC go, all right, consoles have

54:11

had eight core Zen two for

54:13

a while, that's the minimum requirement

54:15

now. What is that gonna do,

54:17

then when. Like when you, I

54:19

think, I guess what I'm saying

54:21

is it's not just older CPUs.

54:23

I would make the argument anyone

54:25

who is realistically considering Battle Mage

54:27

does not have a CPU probably

54:29

good enough. I mean, if you

54:31

kind of followed a chance of

54:33

DPU, you're not buying the most

54:35

nice and CPU. And it's not

54:37

like Intel testing this thing with

54:39

Ayr Lake, which has even worse

54:41

performance on Amanda. I forgot about

54:43

that. Yeah. went out of their

54:45

way to crap all over hourly.

54:47

My eyes actually glazed over during

54:49

that, because I'm like, and I

54:51

didn't talk about in the video

54:53

I put out yesterday. I guess

54:55

by the time this comes out,

54:57

I guess, maybe two days ago,

54:59

or but it's like, yeah, I,

55:01

Aorlake lost, I know. And I

55:03

don't even know if that's the

55:05

fixed version of hourly that didn't

55:07

really get fixed, or if it's

55:09

the worst version from earlier, but.

55:11

But yeah, I just want to

55:13

say this disclaimer too. So like,

55:15

guys, I've gotten a lot of

55:17

hate for saying it seems like

55:19

Battlemage is really not a good

55:21

option. Here's a person who actually

55:23

works on it. This is their

55:25

opinion. I talk to people and

55:27

that's where I get my opinions

55:29

from. You know, I don't just

55:31

benchmark it on a $500 CPU

55:33

and then go, it's great. Like,

55:35

this is where my opinions come

55:37

from. It's actually trying to understand

55:39

like, like, will you be able

55:41

to be able to buy it?

55:43

saying it's going smoothly. To me

55:45

does not seem like it is,

55:47

but um... Let me ask this,

55:49

because this actually dovetails a little

55:51

bit into RDA-4, but I think

55:53

we could tie this to Battle

55:55

Mage. Transpose 1195 writes it and

55:57

says, one of AMD's goals for

55:59

RDA-4 is greater market share, and

56:01

therefore greater developer support. Is there

56:03

a market share percentage above which

56:05

support for things like FSR start

56:07

to become less optional? Can Amy

56:09

get there the strong showing this

56:11

year? I also want you to

56:13

answer that question just in context

56:15

of battle-mage. Do you have to

56:17

support all AMD features? I know

56:19

you don't have to, but I

56:21

think you would agree. You feel

56:23

like you have to support all

56:25

DLSS and invidious features because of

56:27

the market. Is it common sense?

56:29

Like of course you would feel

56:31

that way. So there is an

56:33

argument to be had for game

56:35

developers that even though, say, Intel

56:37

has, you know, what is it

56:39

less than 1% market share or

56:41

something like that? It might be

56:43

worse supporting because your game's going

56:45

to get some publicity from benchmarks

56:47

and Intel might advertise your game.

56:49

And so like when I asked

56:51

for XESSS Intel was like, hey,

56:53

well, mention social media, just add

56:55

XE support. So there is a

56:57

bit of a novelty factor of

56:59

even if AMD doesn't have market

57:01

share, you can be in their

57:03

driver and store and you can

57:06

be in there. adrenaline somewhere and

57:08

they could upload a video to

57:10

your YouTube channel. So a lot

57:12

of game... Yeah, when you're updating

57:14

your AMD drivers, it'll be like

57:16

dishonored to or you're installing or

57:18

something, you know. So there is

57:20

a bit of incentive from game

57:22

developers where this is hidden incentive

57:24

of if we support AMD or

57:26

we support and video, and video

57:28

is going to help us out.

57:30

You know, maybe they'll reply to

57:32

our tickets more faster, maybe they'll

57:34

assign dedicated developers to support us.

57:36

So there's a bit of a,

57:38

even if this feature makes no

57:40

sense to add in, it still

57:42

might be worth adding in just

57:44

to check this box, where you're

57:46

going to get more support. You

57:48

know, so that's kind of a

57:50

baseline incentive for game developers. As

57:52

you content creator might use you

57:54

and more benchmarking. years because you're

57:56

the first game that has retracing

57:58

or you're the first game on

58:00

PS5 pro, even if it doesn't

58:02

have low sales. Yeah, I remember

58:04

for a while hardware boxed called

58:06

Shadow, what is it, something of

58:08

the singularity, ashes of the singularity,

58:10

called it ashes of the benchmark

58:12

because there's like one of those

58:14

first games to use a new

58:16

engine and so it was just

58:18

in every every average for GP

58:20

reviews. Yeah, so, so that's ignoring

58:22

that baseline incentive. It needs to

58:24

push enough units where your customers

58:26

are coming in and saying, hey,

58:28

I have a battle mage, I

58:30

need this thing, right? And typically

58:32

with the whole 73% in video

58:34

thing, that's pretty hard. If I

58:36

could see it going to like

58:38

60% in video or 50% in

58:40

video, that would definitely open the

58:42

up, open the option up, but

58:44

to more aggressively support. But if

58:46

you need to pick like true

58:48

vendors, it's just am. in video,

58:50

and you just don't have to

58:52

bother with the computer 1% unless

58:54

you really care about your play

58:56

base. You're not doing it for

58:58

the numbers, you're doing it for

59:00

the, your game's going to be

59:02

a better thing. Yes, but so

59:04

it sounds like you do think

59:06

that is something people should be

59:08

concerned about. Like, you're paying, I

59:10

would suggest maybe you're painting a

59:12

picture where it's like, look, battle

59:14

major supported on most games when

59:16

it came out. seems like they

59:18

probably worked, you know, over time

59:20

for a couple months to make

59:22

sure this launchments smoother than the

59:24

previous one. But one month out,

59:26

they're still not replying to any

59:28

Devs about XE Super Samling too.

59:30

And so there is a real

59:32

concern. Can they keep up that

59:34

driver support for more than like

59:36

one month? Like, I mean, you

59:38

just test this dishonored and work,

59:40

you know, your game, like, multiple

59:42

games are coming out already that

59:44

seem to have issues. So yeah

59:46

the question when buying a GPU

59:48

people should be oxing and reviewers

59:50

should be oxing is can this

59:52

GPU vendor make a commitment you're

59:54

going to get eight years of

59:56

updates or you're going to get

59:58

10 years of updates, or you're

1:00:00

going to get five years of

1:00:02

day one game support. Because it

1:00:04

seems like the reviewers say, hey, buy this

1:00:07

G.P. year, maybe you'll keep it for a

1:00:09

couple of years, maybe it was

1:00:11

supported, but there's technically no commitment

1:00:13

from any of the vendors saying,

1:00:15

this is going to be a five year

1:00:17

thing. Would you buy an Android phone? They

1:00:19

say, you're going to get major Android updates

1:00:21

for three years, and you know, this is the

1:00:23

last version I've entered I'm going to get.

1:00:25

But for a battle mage, it could be

1:00:27

10 years, it could be two years, the

1:00:30

driver team could get late off next week.

1:00:32

It could be supported, but technically

1:00:34

like a skeleton crew that's not

1:00:36

really making it work. I would suggest

1:00:38

that's probably what it would happen. I don't

1:00:41

think so like is a new game that

1:00:43

comes out a year from now gonna run

1:00:45

on your B580 card if they've given

1:00:47

up on some of the support and

1:00:49

they're focusing on the next card, right?

1:00:52

That's the unknown, unknown, and I'd much

1:00:54

rather if people push Hold GP

1:00:56

event as accountable and say, envy, yeah,

1:00:58

you need to guarantee me X

1:01:00

number of use of updates as

1:01:02

a written public commitment. And if you

1:01:04

don't do that, I'm going to get a

1:01:07

response. Or like, that needs to be the

1:01:09

standard, but it just seems like reviewers

1:01:11

are not pushing for that or not

1:01:13

really big factor. But from the game dev

1:01:15

side, and video, sorry, Intel shipped a

1:01:18

broken driver update for the IGPUs

1:01:20

and then abandoned support for it.

1:01:22

So the very latest driver driver

1:01:24

is for it. For the one you

1:01:26

mentioned, yeah. Yeah, they do this thing

1:01:28

that's like, oh, it's only going to

1:01:30

get security fixes. Great, right? How's my

1:01:33

game's going to run on it? And

1:01:35

so as a game developer, we have

1:01:37

to do all these checks in the

1:01:39

code that's like, if Battle Maids this,

1:01:41

do this fix, else if this Jeep,

1:01:43

you do this fix, and it's not

1:01:46

very fun. And sometimes these bug

1:01:48

fixes are just so complicated

1:01:50

that, um... it's just a house of cars and

1:01:52

falls apart and we just kind of just put

1:01:55

a message that says, have you have this G.P.

1:01:57

or we're just not bothering to support

1:01:59

you anymore. And I'm looking at this,

1:02:01

oh wow, I forgot the 10th

1:02:03

gen even was, because I'm looking

1:02:05

at the driver support ending, that

1:02:07

includes Ice Lake. Ice Lake had

1:02:09

a good integrated graphics. I remember

1:02:11

actually being very impressed with Intel

1:02:13

Ice Lake and thinking you really

1:02:15

didn't need a dedicated graphics card.

1:02:17

I think it was around as

1:02:19

strong as an MX or somewhere

1:02:21

around there, like, you know, that

1:02:23

was decent. That's an interesting blunder

1:02:25

I would say by them. because

1:02:27

whenever you look at the steam

1:02:30

hardware surveys, I mean you have

1:02:32

to admit, Intel graphics is actually

1:02:34

huge if you count integrated. And

1:02:36

if you signal to developers, by

1:02:38

the way, after two years, or

1:02:40

whatever, like we might just crush,

1:02:42

and again when you say two

1:02:44

years, it's like, um, oh yeah,

1:02:46

end of life products. Yeah, so

1:02:48

they ended that life in 2022.

1:02:50

I'm sorry. I'm just reading this

1:02:52

too. I'm talking to you. Right.

1:02:54

Yeah. And so they were selling.

1:02:56

They stayed for that TV year

1:02:58

in 10th Gen or 11th Gen.

1:03:00

It's like all of these work

1:03:02

had their driver support ended after

1:03:04

two years. But there were models

1:03:06

that came out and two years

1:03:08

later, they got rid of driver

1:03:10

support. Like if you're a developer,

1:03:12

you're like, well, I've been putting

1:03:14

all of this effort into just

1:03:16

white knuckling it, making it barely.

1:03:18

So my game runs in 720

1:03:20

P. 30 frames per second because

1:03:22

I know there. you know, whatever

1:03:24

the game is. And then if

1:03:26

you're like, well, it doesn't matter,

1:03:28

you put in all that effort

1:03:30

optimizing for this week, I.G.P.O. Like,

1:03:32

yeah, that's a weird thing to

1:03:34

signal to developers is I would

1:03:36

argue Intel's integrated graphics market share

1:03:38

is probably the biggest reason you

1:03:41

see battlemates. It's more important than

1:03:43

battlemates, even, right? And they could

1:03:45

be taking valuable lessons from still

1:03:47

maintaining that driver and rolling it.

1:03:49

we're still going to learn the

1:03:51

lessons from battle major and apply

1:03:53

it to this, but they're not

1:03:55

even bothering to do that. And

1:03:57

I consider that, you know, mistreatment

1:03:59

of the customers that bought your

1:04:01

your processor, right? And not all

1:04:03

these processes, some of these processes

1:04:05

run laptops, not like you can

1:04:07

plug a GPU in that's upgradeable,

1:04:09

you're screwed, you're going to buy

1:04:11

a new laptop and your laptop

1:04:13

might be fine. Sucks. All right,

1:04:15

let me see here, this is

1:04:17

a random one, but I want

1:04:19

to ask you about it. Clean

1:04:21

Sweep writes in, why do you

1:04:23

think Sony would design PSSR to

1:04:25

use XES code and functionality as

1:04:27

shown here? So he linked to

1:04:29

a video, it is a time

1:04:31

stamp, I clicked on it, I

1:04:33

clicked on it. It's pretty quick.

1:04:35

This seems to be a video

1:04:37

about something else where someone just

1:04:39

stumbled into seeing XESS code in

1:04:41

the PSSR log or something. Maybe

1:04:43

you have nothing to say, but

1:04:45

I included that. You could probably

1:04:47

have the editor put this on

1:04:50

screen. It does look like they're

1:04:52

using the exact same variable names,

1:04:54

right? Like I said, yes, PSSR

1:04:56

and XES both use enabled optimal

1:04:58

screen percentage and quality as exact

1:05:00

values. and they're pretty much also

1:05:02

identical. They even call quality quality,

1:05:04

quality plus. So PSSR has ultra

1:05:06

quality plus and AXES also has

1:05:08

ultra quality plus. That could be

1:05:10

weird. They could have looked at

1:05:12

the excess excess plug-in as a

1:05:14

reference even for like, how do

1:05:16

we make an unreal engine plug-in

1:05:18

that does upscaling? It just swept

1:05:20

all the calls out to be

1:05:22

fans. Because I see super sampling

1:05:24

as open source, right? Yeah, you

1:05:26

can just download off get help

1:05:28

and look at the card. So

1:05:30

that's probably my guess. Otherwise there

1:05:32

might be a naming commission on

1:05:34

a real. But developers will do

1:05:36

this. Like if you're, if you're

1:05:38

making it on real engine plug-in,

1:05:40

you're certain. meet on real plug-in

1:05:42

before. So you're looking at you're

1:05:44

looking at examples. If you're looking

1:05:46

at FSR you're looking at XC

1:05:48

super sampling and you're like, this

1:05:50

seems to work. I wish I

1:05:52

wish these were more unified. Something

1:05:54

else I wanted to mention was

1:05:56

when adding FSR support and media

1:05:58

and these men is use different

1:06:01

naming for everything and they have

1:06:03

different setting names and it's kind

1:06:05

of a mess for the u.

1:06:07

Right. I guess in video had

1:06:09

this idea where developers in their

1:06:11

game would have an in-video tab

1:06:13

and it would just have in-video-LS

1:06:15

and video reflects and video this

1:06:17

and video that. For a game

1:06:19

developer, we don't want that. We

1:06:21

just want a tab that's like

1:06:23

enable upscaling and it just does

1:06:25

whatever the vendor or specific thing

1:06:27

is. So there's definitely some... you

1:06:29

know, on the U.S. side. They're

1:06:31

organizing it from an invidious-centric mindset

1:06:33

where they're like, here's all the

1:06:35

invidious stuff. You're saying, don't do

1:06:37

that, put your invidious stuff where

1:06:39

it's already supposed to go, right?

1:06:41

Yeah, like instead of like a,

1:06:43

sir, you know, I got a,

1:06:45

you know, guidelines from DLS for

1:06:47

invidious marketing, go buy an invidious

1:06:49

settings into it. They want that

1:06:51

in the game setting. So when

1:06:53

you have like video, display, audio,

1:06:55

invidious settings, I think I have

1:06:57

seen that in a game or

1:06:59

two. Yes, right. So when we're

1:07:01

adding this settings in, we had

1:07:03

an invidious settings panel, and we're

1:07:05

like, hang on, are we just

1:07:07

marketing invidious gfues and that game

1:07:09

is like an ad? Like, could

1:07:12

we just call this upscaling or

1:07:14

whatever? Is invady gonna get upset

1:07:16

at us if we? Don't market,

1:07:18

you know, if we just call

1:07:20

it like you can use the

1:07:22

way DLS or up scaling, but

1:07:24

you can use it in our

1:07:26

town for it and not the

1:07:28

envious system, right? And that's confused

1:07:30

me before because I've been like,

1:07:32

this game says it has DLS,

1:07:34

support, where is it? Right, like

1:07:36

DLS, is calling it ultra-performance, performance,

1:07:38

balanced, and ultra-performance is a completely

1:07:40

different setting on FSR and they

1:07:42

might call a different name. But

1:07:44

if you go over to PSSR,

1:07:46

that's kind of confusing for developers.

1:07:48

We want unification. We want to

1:07:50

have low, medium, high upscaling as

1:07:52

a setting that works on every.

1:07:54

GPUs, that is standardized and we're

1:07:56

not advertising some GPUs. Like me

1:07:58

putting X, AX, X, 2 at

1:08:00

my game might be advertising Battle

1:08:02

Mage more than, you know, Intel's

1:08:04

marketing team is, right? Like this

1:08:06

channel, you know, Mosler's Dead has

1:08:08

more subscribers than Intel's or on

1:08:10

gaming channel, right? So they're like

1:08:12

not putting a lot of effort

1:08:14

in to the heard marketing for

1:08:16

gaming. Well, so it's funny. So

1:08:18

it sounds like you want, like,

1:08:21

like, like, like, Cut the bullshit,

1:08:23

AMD, AMD, invidia, just whatever invidious

1:08:25

calling it, which I believe is

1:08:27

like quality, balanced, performance, can we

1:08:29

all just call the upscaling that?

1:08:31

Can you please just call yours

1:08:33

that so we know what you're

1:08:35

targeting when you do that? That's

1:08:37

what you're saying too. Yeah, just

1:08:39

unify everything and, you know, if

1:08:41

you want mass game support, make

1:08:43

it easy to add and unify

1:08:45

everything as much as possible. You

1:08:47

know, like, like, everyone like resizable

1:08:49

bars called Sam for some reason

1:08:51

just to confuse everyone. I don't

1:08:53

know if it's worth it, right?

1:08:55

Well, technically they said it first.

1:08:57

So that is where you think,

1:08:59

but I guess, it should be

1:09:01

a standard of just call it

1:09:03

the same thing so we know

1:09:05

what it is and it's unified.

1:09:07

I don't want to add three

1:09:09

different UI menus for a GPU.

1:09:11

It's kind of a mess. Well,

1:09:13

speaking of copying what your competition

1:09:15

calls their stuff, let's talk about

1:09:17

the 97DX. We have to talk

1:09:19

about our DNA4. It was. Not

1:09:21

in the press conference, they put

1:09:23

out a separate press release. I

1:09:25

suppose from that perspective, we should

1:09:27

mention that we both agree that

1:09:29

their excuses are very, very silly.

1:09:32

I'll even just say, I am

1:09:34

like 99% sure that AMD just

1:09:36

decided last minute, we shouldn't announce

1:09:38

our DNA for fully until we

1:09:40

fully understand what we're competing with.

1:09:42

Let's delay it a week. And

1:09:44

actually on that note I was

1:09:46

actually messaged today by a contact

1:09:48

that was just briefed by AMD.

1:09:50

This context said that they should

1:09:52

unveil RDA4 fully or at least

1:09:54

close to fully. Just like a

1:09:56

week from now, I think I

1:09:58

was told January 15th, so I'll

1:10:00

have that quote put on screen

1:10:02

too. So that's interesting. It still

1:10:04

sounds like it might launch in

1:10:06

January or something. It really doesn't

1:10:08

sound like something went wrong so

1:10:10

much as AMD didn't know if

1:10:12

they're ready to talk about RDA4

1:10:14

until they fully understood the competition.

1:10:16

That is really what I believe.

1:10:18

happened with it being cut there

1:10:20

but I don't know what do

1:10:22

you think about AMD saying they

1:10:24

only had 45 minutes. The narrative

1:10:26

thing like I watched the hardware

1:10:28

and books video and they said

1:10:30

oh CS has got a 45

1:10:32

minute timer like you know it's

1:10:34

busy we can't like videos wasn't

1:10:36

they the interesting thing about AMD's

1:10:38

keynote was how many vendors they

1:10:41

had there. promising they're going to

1:10:43

ship AMBCP's like they had they

1:10:45

spent time on Dell and Xbox

1:10:47

and all to be fair I

1:10:49

think Dell's a big deal for

1:10:51

AMD for investor reasons I would

1:10:53

agree the Xbox thing I think

1:10:55

we were talked about that is

1:10:57

complete nonsense I mean unless Xbox

1:10:59

announced in a next-gen console that

1:11:01

uses already in a four I

1:11:03

don't know maybe they're handheld that's

1:11:05

rumored will use already was already

1:11:07

in a four maybe that's why

1:11:09

they were there but they should

1:11:11

have announced it if PlayStation who's

1:11:13

using some of the architectural features

1:11:15

of RDA4, I think like the

1:11:17

BBH reversal thing for ray tracing

1:11:19

from it, and the PS5 Pro,

1:11:21

that would make more sense because

1:11:23

they're actually using it. Point being,

1:11:25

if they had to cut some

1:11:27

time, maybe the Xbox part, and

1:11:29

you can just put that out

1:11:31

on your channel separately, I don't

1:11:33

know. I think the Xbox thing

1:11:35

goes like Xbox is a publisher

1:11:37

even, which is super weird, because

1:11:39

like Xbox is a PC games

1:11:41

publisher, which we're not really used

1:11:43

to that much. But we didn't

1:11:45

ever go told what FSL for

1:11:47

was, I guess they told the

1:11:49

press before the keyner. Yeah. But

1:11:52

the, it's definitely a bit of

1:11:54

a mess of a, for a

1:11:56

event like this, you'd plan for

1:11:58

weeks. right, like, just randomly cutting

1:12:00

it off or, or, or moving

1:12:02

something off because of some 45

1:12:04

minute thing. I think it's crazy,

1:12:06

but if it was, if the 45

1:12:08

minute thing is true, why didn't AMD pay

1:12:10

more money to see us to get a bigger

1:12:12

slot at a different spot or a different

1:12:14

stage or whatever the hell they needed to

1:12:17

do? Like, it's kind of like the,

1:12:19

depending on how much money you have big,

1:12:21

your company is, you get a biggest lot

1:12:23

kind of vibe kind of vibe going on.

1:12:25

And I think AMD. could have just

1:12:27

got a bigger slot or something like

1:12:30

or put the key dirt off to

1:12:32

the video or something like that. But

1:12:34

at least or just be out front

1:12:37

and say, hey, we're going to launch

1:12:39

this thing, you know, a week from

1:12:41

now or say something like that, that

1:12:43

doesn't imply this weird

1:12:46

shutdown castle kind of thing

1:12:48

vibe going on. Well, or, you

1:12:50

know, half of the time you

1:12:52

spent talking about cracking. Remove that

1:12:54

we know just show us what it is

1:12:56

and then move on like there's so many

1:12:58

things Cut the Xbox thing in half.

1:13:01

There's so many things that they could

1:13:03

have cut down That I saw a

1:13:05

joke that was like the audio issues

1:13:07

in the beginning of it is why

1:13:09

we didn't get the art in a

1:13:11

four announcement I don't but I don't

1:13:13

think so I don't believe any of

1:13:15

that. I think I don't know who

1:13:17

came up with that excuse. It's a

1:13:19

ridiculous excuse again. I think they were afraid

1:13:21

to announce pricing before in video.

1:13:23

They caught wind of how insane in video's

1:13:25

presentation is going to be. I might also

1:13:28

say from a more pessimistic point of view,

1:13:30

maybe AMD saw the generated frame stuff and

1:13:32

said we need our charts to have more

1:13:34

fake frames on them. They may have said

1:13:37

we need to go away and do that.

1:13:39

I don't know. And by the way I

1:13:41

also think then in video I went, oh so

1:13:43

AMD's not announcing pricing, fine, we'll just go to

1:13:45

our minimum price where we're willing to do for

1:13:48

these cards. See how AMD likes making us go

1:13:50

first. That is really what I think. They could

1:13:52

have mentioned the cards without pricing. Or they could

1:13:54

have teased a new event. Like if it's happening

1:13:56

in a week from now, just tell us the

1:13:59

events in a week. The vibe I

1:14:01

get is like we don't care

1:14:03

about our GPs that much, right?

1:14:05

Because like compared to CPUs and

1:14:07

there's not, if you want to

1:14:09

grow the GPU market, you need

1:14:11

to show you care about the

1:14:13

GPUs, like announce it without pricing,

1:14:15

because like power color did this

1:14:17

tweet, whether like, hey guys, we're

1:14:19

at CES too, we could add

1:14:21

GPUs or whatever, but they weren't

1:14:23

even in the keynote, right? You're

1:14:25

like, man, I feel bad for

1:14:27

this. For this. their GPUs being

1:14:29

announced to show them off. And

1:14:31

they're like, hey, we're here too.

1:14:33

And it got caught. So like,

1:14:35

I honestly feel bad for the

1:14:37

OEM partners more than AMD on

1:14:39

the. Oh, yeah. I mean, I'm

1:14:41

sure AMD is doing at least

1:14:43

as much, if not significantly more

1:14:45

Apple research than I'm doing for

1:14:47

my YouTube channel. So I assume

1:14:49

they expected the, you know, to

1:14:51

be a certain price they expected.

1:14:53

DLS4 to be interesting, but they

1:14:55

didn't know and video is going

1:14:57

to claim AI face stuff and

1:14:59

all this craziest things. And I

1:15:01

think they said, well, we can't

1:15:03

completely remove a press announcement because

1:15:05

we literally have AIBs at the

1:15:07

show floor showing off 16 gigabyte

1:15:09

cards. So like, what's the minimum

1:15:11

we can show without showing anything

1:15:13

so that it doesn't look completely

1:15:15

ridiculous? There's AIBs there. It just

1:15:17

seemed to me like organized chaos

1:15:19

because we had, and video for

1:15:21

example, release the, there was a

1:15:23

press release from, was it razor

1:15:25

and MSI about new laptop with

1:15:27

50 series GPUs when the 50

1:15:30

series GPUs hasn't been announced yet.

1:15:32

Like it just seems like that's

1:15:34

a weird kind of chaotic launch

1:15:36

where we're finding out random bits

1:15:38

of info from random people. And

1:15:40

maybe that's intentional though, maybe invidious

1:15:42

like teased the 50-90 before we

1:15:44

announced it. information bias where the

1:15:46

first thing they hear is treated

1:15:48

more seriously. So I wonder if

1:15:50

kind of doing this bait and

1:15:52

switch hide thing is more dangerous

1:15:54

to the brand than just announcing

1:15:56

it without pricing. compared to making

1:15:58

it disappear. Yeah, I actually, you

1:16:00

know, we'll see if I'm wrong.

1:16:02

My suspicion is it doesn't matter

1:16:04

actually that like no one's talking

1:16:06

about the Intel ring bus flow

1:16:08

anymore even though the C fees

1:16:10

are still breaking. My suspicion is

1:16:12

that no one will even remember

1:16:14

this happened a month from now.

1:16:16

We'll know what the pricing is.

1:16:18

They'll have launched or be about

1:16:20

to launch. We'll know how it

1:16:22

stacks up. You know, it'll be

1:16:24

like, ah, it's a little better

1:16:26

than the 50-70, but it doesn't

1:16:28

have DLS, but it's a little

1:16:30

cheaper with more RAM is what

1:16:32

I'm guessing. And that's where I

1:16:34

think AMD is calculating, you know.

1:16:36

And I guess you can put

1:16:38

this light up or whatever, but

1:16:40

the other part was they made

1:16:42

their CPU naming convention less by

1:16:44

calling everything Per Max Plus. So

1:16:46

I'm getting this signal where they're

1:16:48

like, we care about the naming

1:16:50

so much guys, we've changed our

1:16:52

entire naming convention seemingly every other

1:16:54

card, but we've made the CPU

1:16:56

naming more complex. And I wonder

1:16:58

if like AMD having confidence in

1:17:01

their own product and just staying

1:17:03

with the cause. is stronger for

1:17:05

the GP brand instead of solving

1:17:07

out the naming committee. Like, could

1:17:09

you imagine if video just changed

1:17:11

their naming convention? Every other gen.

1:17:13

Just to mess with the AMD

1:17:15

order. Like it just seems like

1:17:17

if you want to build an

1:17:19

audience of people that buy AMD

1:17:21

GPS every generation, make it consistent

1:17:23

and make it like, stick it

1:17:25

out even if it's uncomfortable to

1:17:27

build the brand slowly of time.

1:17:29

No, but yeah, your point about

1:17:31

the CPU versus GPU naming is

1:17:33

interesting because, you know, going from

1:17:35

what? 7700xT to 97xT. I know

1:17:37

what that means. I think most

1:17:39

enthusiasts will instantly know what they're

1:17:41

referencing and who cares. But if

1:17:43

you're going to do that to

1:17:45

try to set like, why is

1:17:47

the CPU then? I'm looking at

1:17:49

it now, AMD Rise and AI

1:17:51

Max plus 395. Like, you have

1:17:53

not made this, so like, but

1:17:55

there are different marketing teams, and

1:17:57

this proves it. Like you have

1:17:59

on the one side, they're like,

1:18:01

9070, 90, 9060, 60, 60, 9050,

1:18:03

and we're done, we're going home.

1:18:05

And over here, we have the

1:18:07

AI Max 365 plus triple X.

1:18:09

I mean, I don't know. There's

1:18:11

a lot of inconsistency here with

1:18:13

AMDA and marketing, and I just.

1:18:15

I don't know. It still just

1:18:17

doesn't to me feel like they're

1:18:19

getting it together like they were

1:18:21

telling me they might. It's kind

1:18:23

of baffling from the consumer fan

1:18:25

point of seeing like how hot

1:18:27

is it to get the naming

1:18:30

consistent. You know what I mean?

1:18:32

Like it doesn't. They went to

1:18:34

9,000. So presumably they're going to

1:18:36

have to start over again with

1:18:38

UT&A or something and unless they

1:18:40

go to 10,000, which maybe they're

1:18:42

getting it. I think it's at

1:18:44

this point we've got I've got

1:18:46

to declare it as being done

1:18:48

intentionally to confuse consumers. Right. Like

1:18:50

we want to put Zen 3

1:18:52

and Zen 4 CPUs out there.

1:18:54

They can't have the old name

1:18:56

because no one will realize they're

1:18:58

old and buy them. We need

1:19:00

to rebrand them and keep reselling

1:19:02

them. And Intel does the same

1:19:04

thing where if you bought a

1:19:06

14th gen, it might not be

1:19:08

rapidly. It might be the one

1:19:10

before. And so it's just swap

1:19:12

the naming every gen to confuse

1:19:14

consumers. so they don't know what

1:19:16

they're buying and it's not good.

1:19:18

Like the pro back thing stuff,

1:19:20

I kind of see vaguely of

1:19:22

like, Apple's doing it, or maybe

1:19:24

it's like working. I can get

1:19:26

it to a certain degree, yes,

1:19:28

but then why do you have

1:19:30

AIHX? Like, it's like you wanted

1:19:32

to use both names at the

1:19:34

same time. Yeah, so we have

1:19:36

this challenge in our game where

1:19:38

we, you know, we hire someone

1:19:40

and say, hey, repeat all the

1:19:42

names of the dinosaurs in our

1:19:44

game backwards or if I had

1:19:46

or whatever, right? But just imagine

1:19:48

going to an AMD employee and

1:19:50

say, can you just list off

1:19:52

like the different types of CDs

1:19:54

on what they're named for the

1:19:56

last couple of gems? fail. Whereas

1:19:58

if you go to invidia, oh

1:20:01

easy, easiest thing to do in

1:20:03

the world. Count it, count it,

1:20:05

count it, count it, to prove

1:20:07

it. This piece of content

1:20:09

is brought to you by the Minies

1:20:11

Forum v3. The Minies Forum v3 is

1:20:13

a Reasonably priced alternative to a sort

1:20:15

of form factor that usually costs twice

1:20:17

as much and yet despite costing under

1:20:20

a thousand dollars It comes with a

1:20:22

lot of unique features. It can work

1:20:24

as its own independent display to be

1:20:26

used with game consoles or really another

1:20:28

PC if you want to use it

1:20:30

for that and that display also is

1:20:33

high refresh rate and is 1600 The

1:20:35

thing has Hawk Point, I mean I

1:20:37

actually did a full review for this,

1:20:39

it works fantastically, can play any AAA

1:20:41

game that I own. And then, you

1:20:43

know, when it comes to battery life,

1:20:46

it's, you know, a bit middle of

1:20:48

the road, but the fact is, it

1:20:50

can use just about anything to charge.

1:20:52

A lot of laptops that I've come

1:20:54

across require, I don't know, 45 watt,

1:20:57

minimum power sources to charge. This thing's

1:20:59

minimum is 15 watts. You can plug

1:21:01

it into a random port at the

1:21:03

airport at the airport. keep working and

1:21:05

I've been using it diligently throughout 2024

1:21:07

as I've flown a lot and it

1:21:10

actually fits on the plane can keep

1:21:12

going forever with battery packs and I've

1:21:14

got a lot of work done on

1:21:16

the go and then when I get

1:21:18

back to the hotel room I start

1:21:20

to play because again it has gaming

1:21:23

shops. So if you're interested in this

1:21:25

product, go to the link in the

1:21:27

description to use offer code Broken Silicon

1:21:29

to save $30 on this fantastic laptop

1:21:31

tablet display hybrid device. If you use

1:21:33

Broken Silicon you'll be helping the channel

1:21:36

even just click on the link helps

1:21:38

as well to support Moore's Law is

1:21:40

dead by checking out the minisform v3

1:21:42

today. B.

1:21:45

Jones 1794 writes in says, I'm

1:21:47

incredibly annoyed at Amy, literally not

1:21:49

even trying in their CES keynote

1:21:51

by not announcing GPUs or pricing

1:21:53

or even bringing out Lisa Sue.

1:21:56

It felt like they just gave

1:21:58

up before they started, but if

1:22:00

the. 9070XT really is 7900XT-H performance,

1:22:02

which I know they didn't announce

1:22:04

performance, but they did show a

1:22:06

chart comparing, you know, the 9070XT

1:22:08

to the 7900XT directly. So do

1:22:10

you think that performance at a

1:22:12

$400, $450, or maybe $500 price

1:22:14

point is realistic, and could it

1:22:16

possibly stand a chance against the

1:22:19

5070 once drivers are finalized? And

1:22:21

reviewers actually benchmark results and compare

1:22:23

the two. Well, I'll jump in

1:22:25

first and say. This absolutely could

1:22:27

cost $500 or less. I think

1:22:29

they could get away with $400

1:22:31

if they wanted to accept fairly

1:22:33

low margins because the $7,800 XT

1:22:35

was $500. It used basically the

1:22:37

same node for most of the

1:22:40

die 200 millimeter squared and then

1:22:42

it had four like 36 millimeter

1:22:44

squared six nanometer dies and then

1:22:46

those so you have like a

1:22:48

total die size right of like

1:22:50

330 or 50 or something. And

1:22:52

then those were all had to

1:22:54

be packaged and put together and

1:22:56

that cost money. Art and A4,

1:22:58

my understanding, is like a 250

1:23:01

to 350 millimeter squared, probably not

1:23:03

even 350 to 300 millimeter squared,

1:23:05

total monolithic 4 nanometer, I think

1:23:07

it costs less to make than

1:23:09

the 7800xD probably. So I think

1:23:11

there's no doubt in my mind

1:23:13

that 499 they can do if

1:23:15

they want to, and I could

1:23:17

see cheaper. Yeah,

1:23:19

I think I'm a little bit

1:23:21

concerned about the the keynote retraction

1:23:23

kind of kind of showing their

1:23:26

hand a little bit here on

1:23:28

the on the like were they

1:23:30

about to announce something stupid. I

1:23:32

also wonder if you know, if

1:23:34

the AMD GPU traditional Rasser performance

1:23:36

is better, I would come out

1:23:38

and say, hey, these are real

1:23:40

frames and we're, you know, rendering

1:23:42

this much faster at this present

1:23:44

like if they can catch and

1:23:46

video in the video in the.

1:23:48

By the way, our raster performance

1:23:50

isn't actually that faster from last

1:23:52

gen. They could have a angle

1:23:54

there from the media standpoint, but

1:23:56

if people really buy this fake

1:23:58

frame stuff and stuff like that

1:24:01

they might be in serious trouble.

1:24:03

Even if the card is a

1:24:05

lot faster, is there features gonna

1:24:07

be better than DLS and some

1:24:09

of the other stuff where it's

1:24:11

at now? And that's the real

1:24:13

question. So I just now in

1:24:16

my head sort of putting

1:24:18

myself in AMD's shoes and

1:24:20

hear me out everybody, I can

1:24:22

now actually kind of see

1:24:24

why they don't know what to do

1:24:26

because we know. Blackwell's

1:24:29

launching in three weeks or

1:24:31

something and I know that in

1:24:33

real performance the 50-70s probably a

1:24:36

little stronger than a 40-70-T-I seems

1:24:38

like the 50-90 will be 35

1:24:40

to 50-something percent faster we'll see

1:24:43

I mean they only showed a

1:24:45

couple games that they had selected

1:24:47

faster than the 40-90 and so

1:24:50

on and so forth but that's

1:24:52

not what envy portrayed and so

1:24:54

you've got the situation where If

1:24:57

your AMD and Invidia was

1:24:59

honest about real-world performance outside of

1:25:01

using fake frames, then you could

1:25:04

just go, oh, well, we're going

1:25:06

to compare the 90-70XT to the

1:25:08

4080, or the 4070 TI Super,

1:25:11

because we know that invidia admitted

1:25:13

the 50-70s around that performance. But

1:25:16

they haven't. Invidia claim the

1:25:18

50-70s as strong as a 40-90,

1:25:20

even though it isn't, and we're

1:25:22

not going to have real benchmarks

1:25:24

of it. for three weeks. So

1:25:26

what do you announce if you're

1:25:28

AMD? Do you announce? Like let's

1:25:31

say it's even better than

1:25:33

the 7900XT, all right? But I

1:25:35

think it'll be around there. Like

1:25:37

even if it was XTX

1:25:39

performance, what do you do?

1:25:41

Say better than the 4080? Because

1:25:44

then all the press is gonna

1:25:46

go, but that doesn't matter. The

1:25:48

5070 is stronger than the 4090.

1:25:51

So actually now that I'm thinking

1:25:53

about the fake. frames and the

1:25:56

crazy claims that they're making.

1:25:58

We've got to figure out... a strategy

1:26:00

and how to like accurately portray the

1:26:02

value we think we're bringing without people

1:26:05

just being misled or something. Also if

1:26:07

they want to go the angle of

1:26:09

well fake frames are bad if they

1:26:12

have if they added fake frames in

1:26:14

their FSL 4 they can't exactly go

1:26:16

and say hey fake frames are bad

1:26:18

but we added them in FSL 4

1:26:21

they're kind of stuck in this trap

1:26:23

of seeing if invidious like is there

1:26:25

a gap between rational performance versus fake

1:26:28

frames and upscaling and they might be

1:26:30

really hard stock on waiting until because

1:26:32

like how do you even market this

1:26:35

thing right like emphasis for great like

1:26:37

let's say they added three generated frames

1:26:39

instead of four or two instead of

1:26:42

three they're going to be behind on

1:26:44

the numbers so badly and if reviewers

1:26:46

trade like if they plainly trust that

1:26:49

the EPS number which has been done

1:26:51

to last ten years people just go

1:26:53

based on the EPS number On the

1:26:56

paper, they might be screwed, but the

1:26:58

card might actually be not bad if

1:27:00

it is properly marketed and properly explained.

1:27:02

But they're kind of caught in this

1:27:05

thing of we might be doing frame

1:27:07

generation, but NPD is doing more frame

1:27:09

generation. How do people compare this traditionally?

1:27:12

And that's going to be where it's

1:27:14

stuck. Until that 50-90 performance of the

1:27:16

50-70, until that gets debunked. How are

1:27:19

you supposed to announce? I'd like you

1:27:21

to release shots or whatever, but your

1:27:23

shots are going to be comparing lost-gen

1:27:26

because you don't have access to the

1:27:28

next one. So you're stuck with shots.

1:27:30

And invidious, actually, and like, even the

1:27:33

50-60 though, that's not an ounce or

1:27:35

somehow faster than a 40-90. You can't

1:27:37

compare it to last-gen. Honestly, I'm sorry

1:27:40

to realize, maybe that as dishonest as

1:27:42

it is, maybe that was a brilliant

1:27:44

move by invidious, if... a lower cut

1:27:47

is faster than a 40%. So they

1:27:49

have to prove the lie and it's

1:27:51

like an video handpick the best charts

1:27:53

to put them in the worst position.

1:27:56

Like either the charts didn't make an

1:27:58

video look good or they intentionally were

1:28:00

careful to avoid the empty comparison. So

1:28:03

it's kind of hard to market a

1:28:05

card like that if you're in this

1:28:07

trap, like a perfectly constructed trap. Yeah,

1:28:10

and now I'm, yeah, I think that's

1:28:12

why Amy delayed their announcement. And here's

1:28:14

what I worry. I really worry. And

1:28:17

again, I don't know. how good FSR

1:28:19

for is. I honestly know, or at

1:28:21

a minimum, I haven't asked anyone or

1:28:24

read anything about it for many months,

1:28:26

but let's say AMD was gonna add

1:28:28

fake frames and let's say they were

1:28:31

gonna have, like you said, three times

1:28:33

instead of four times, is AMD now

1:28:35

going back? And they're like, just add

1:28:37

a stupid mode that six frames, like

1:28:40

even if it's horrible, add the six

1:28:42

frame mode, and I'm just worried they're

1:28:44

gonna do that. Come back in a

1:28:47

week and go, oh well we have

1:28:49

six times the frames and it's actually

1:28:51

50% faster than a 7900xx if you

1:28:54

use this crazy mode. That is what

1:28:56

I suspect they're going to do because

1:28:58

that's the only way I see out

1:29:01

of this trap is to fight fire

1:29:03

with fire. And if they don't have

1:29:05

the reflex two or whatever attack that's

1:29:08

supposed to counter the leads and sea

1:29:10

problem. they could have some fake frames

1:29:12

and they picked a less number of

1:29:15

fake frames to go with because they

1:29:17

don't have a fix for the latency.

1:29:19

And not to mention the whole neural

1:29:21

sheater thing. We really don't know if

1:29:24

they have an answer for that. Like,

1:29:26

they might be. Or if they need

1:29:28

to, does it even work? Like, you

1:29:31

know, in video, when they announced the,

1:29:33

because I want to remind people of

1:29:35

this, when in video announced the 4060TII

1:29:38

8 gigabyte, they said because it has

1:29:40

GDDR 6 or something, 8 gigabytes was

1:29:42

enough. Well, it wasn't. And then they

1:29:45

just, like, didn't even respond to an

1:29:47

email after that. respect major respect to

1:29:49

the AMD team, engineers working late in

1:29:52

the night trying to figure out how

1:29:54

the hell we even announced this thing

1:29:56

and you know I hope for the

1:29:59

best of them. because all is totally

1:30:01

of sure. Yeah, again, just for everyone

1:30:03

listening, as dumb as I think

1:30:05

it is what AMD did, and

1:30:07

there's no defense of the 45-minute

1:30:09

excuse, I mean, it's just stupid, but I

1:30:11

now get why they probably deleted a

1:30:13

week, because do we show more fake

1:30:15

frames? Do we not show more fake frames?

1:30:18

Do we announce that we're also

1:30:20

making some competitor to neural compression?

1:30:22

Even though we don't even know

1:30:24

if their thing will work, deals

1:30:26

that's one, how do we know that will work?

1:30:28

And if I were at AMD, I would say, add

1:30:30

more fake frames to the graph. We'll have

1:30:32

two charts, two different slides. The

1:30:34

first slide will be, you know, the real

1:30:37

comparison, but we'll have the dumb thing

1:30:39

so we can compare it to invidious

1:30:41

and say faster than 49. Like, I

1:30:43

guess we have to pretend just like

1:30:45

in video is. I really think that's

1:30:47

the only option. I think I'll skip over

1:30:49

some of the ray tracing talk. I'll bring

1:30:51

it very quickly, though. Like, like,

1:30:53

there's questions we get these we get

1:30:55

these all the first slides. how important

1:30:58

is ray tracing in deciding what card

1:31:00

you buy. At this point it seems

1:31:02

like the opinion of me and Dan