Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

 

[poke01]

https://twitter.com/x/status/1877028029576147419

FSR4 only on RDNA4. Well if you want good FSR4 at the minimum.

 

[jpiniero]

Tlogic measured 12.3 by 28.4 so yeah 350 or below.

Even 350 is kinda rough if they are going to price it compare to the 5070. GB205 is way way smaller.

 

[gdansk]

GB205 is way way smaller.

Do you know the size? 280mm²?

 

[Heartbreaker]

https://twitter.com/x/status/1877028029576147419

FSR4 only on RDNA4. Well if you want good FSR4 at the minimum.

I have no issue with that. It's likely a real limitation, unlike NVidia making FG 40 and 50 series only... Note: I don't actually care about FG, it's the principal of the thing.

Plus you can be assured that every FSR 4 game will fall back to FSR 3 where it's lacking the AI HW.

 

[jpiniero]

Do you know the size? 280mm²?

I don't think it's revealed but probably closer to 240.

 

[MrTeal]

I mean for gamers

https://twitter.com/x/status/1876624801600532711

Maybe it's going to FSR4?

Here's to hoping. AMD and FSR is what gives me the option for Frame Gen in CP2077 on my 3080. Thanks Nvidia!

 

[Meteor Late]

Is there even FSR FG latency testing? because when I searched it, all I found is either nothing or data with the first revision that was broken in some aspects (like not working with VRR).

 

[Saylick]

We're going to infer inference. Why actually calculate 4 whole bits when we can calculate 1 and predict the other 3. And since we're predicting the least significant bits it has little impact, really.

In 10 years Jensen will just sell people an Nvidia branded blindfold for $2000 and he’ll tell us to just close our eyes and imagine high fps, fully path traced games, all powered by Nvidia Neural Rendering and Nvidia Reflex 4, which cuts out the whole rendering pipeline since light doesn’t even need to travel into your eyes. Super low latency!

 

[MrTeal]

In 10 years Jensen will just sell people an Nvidia branded blindfold for $2000 and he’ll tell us to just close our eyes and imagine high fps, fully path traced games, all powered by Nvidia Neural Rendering and Nvidia Reflex 4, which cuts out the whole rendering pipeline since light doesn’t even need to travel into your eyes. Super low latency!

It's the future of gaming.

 

[lucasworais]

Huge oof if that die size is real. It’s time to Radeon have Arc margins I guess lol

 

[JustViewing]

Not sure it is already posted here

 

[gdansk]

Huge oof if that die size is real. It’s time to Radeon have Arc margins I guess lol

It's better than Arc margins even with a "huge oof". ~380mm² at $400 and $500.
But yes that's 350 and 500 less than Nvidia. Even if GDDR7 was twice as expensive it can't dent that margin much.

 

[Kepler_L2]

Do you know the size? 280mm²?

272mm² IIRC

 

[lucasworais]

It's better than Arc margins even with a "huge oof". ~380mm² at $400 and $500.
But yes that's 350 and 500 less than Nvidia. Even if GDDR7 was twice as expensive it can't dent that margin much.

The worst thing is, it’s competing against a much smaller die

GB205:

272mm² IIRC

 

[gdansk]

The worst thing is, it’s competing against a much smaller die

I was already comparing like size parts.

 

[Kepler_L2]

The worst thing is, it’s competing against a much smaller die

I think it will end up a bit faster than 5070 in raster, but AMD won't be able to charge more because of RT/DLSS/Reflex 2/etc

 

[iLLusiveMan]

What are your thoughts on this ?

 

[beginner99]

What are your thoughts on this ?

COD favors AMD heavily so AMD beating a 4090 isn't all that magical as it sounds. Will be very different in other games especially NV favoring ones with heavy RT.

If the die size is indeed around 350mm2 or bigger (vs the leaked 270mm2 ish) and the huge ass coolers + 3 pin options pointing to high power use = high clocks, that performance doesn't seem that unrealistic. I'm thinking 3.2ghz clocks or more.

 

[iLLusiveMan]

COD favors AMD heavily so AMD beating a 4090 isn't all that magical as it sounds. Will be very different in other games especially NV favoring ones with heavy RT.

If the die size is indeed around 350mm2 or bigger (vs the leaked 270mm2 ish) and the huge ass coolers + 3 pin options pointing to high power use = high clocks, that performance doesn't seem that unrealistic. I'm thinking 3.2ghz clocks or more.

Agreed. I mean the only "inflation" going on with the IGN article is that I think they mislabeled the 9070XT as a normal 9070.

 

[CastleBravo]

It's better than Arc margins even with a "huge oof". ~380mm² at $400 and $500.
But yes that's 350 and 500 less than Nvidia. Even if GDDR7 was twice as expensive it can't dent that margin much.

IMO, there is no way we are getting an uncut ~380mm^2 RDNA4 GPU for only $500.

 

[gdansk]

IMO, there is no way we are getting an uncut ~380mm^2 RDNA4 GPU for only $500.

There is one way: bad performance per area. And the forecast is looking likely.

 

[gdansk]

What are your thoughts on this ?

View attachment 114503View attachment 114501

It's bad idea to extrapolate anything from a COD screenshot as it doesn't really list all settings like many other benchmarks do. Don't extrapolate from alpha drivers. Or a single game in general.

Really I know there's nothing to go on but it is best to ignore the IGN article entirely.

 

[iLLusiveMan]

It's bad idea to extrapolate anything from a COD screenshot as it doesn't really list all settings like many other benchmarks do. Don't extrapolate from alpha drivers. Or a single game in general.

Really I know there's nothing to go on but it is best to ignore the IGN article entirely.

Why wouldn't AMD take that article down if it wasn't accurate ? It's been online for almost a day now. Do they like dealing with PR nightmares at launch ?

 

[gdansk]

Why wouldn't AMD take that article down if it wasn't accurate ? It's been online for almost a day now. Do they like dealing with PR nightmares at launch ?

Not their job to control the media after the fact. But they let them run one (1) single benchmark. AMD set themselves up for it again.

 

[Bryo4321]

What are your thoughts on this ?

View attachment 114503View attachment 114501

That last bit about AMD “wanting” this out seems delusional to me. Have these guys EVER worked at a company before? AMD is not some singular puppet master entity, lol. It was just some dude from ign sneaking a bench in at a booth, and he didn’t even seem to know there was a 9070 and another xt model.