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

 

[SolidQ]

So was RDNA2, it was a major PPW and PPB improvement with no additional die area, on the same node.

Yeah based on TPU

40%+ perf than GRE

AMD will be able to dial it in to outperform 5060Ti no problemo.

Should be like 7800XT perf? or possible ~7900GRE?
What if AMD gonna surprise us 9060XT with 16GB/9060 is 12GB, both cut down N48 and let 9500 series for N44

 

[adroc_thurston]

N44 is where things get interesting again IMO.

It's gonna be a horribly mutilated abortion. Please ignore.

40%+ perf than GRE

Yep. N31 perf-ish, but with the same caveats as AD103 (gonna lose in the same games).

 

[GodisanAtheist]

It's gonna be a horribly mutilated abortion. Please ignore.

- You've just made it that much more interesting

 

[SolidQ]

but with the same caveats as AD103 (gonna lose in the same games)

in bandwidth limit games? That mostly older games?

 

[adroc_thurston]

- You've just made it that much more interesting

seems like they're keeping the tradition of x4 dies being crimes against humanity.

in bandwidth limit games? That mostly older games?

More like older (or more traditional) engines.

 

[gdansk]

An abortion like N44 was designed for laptops with iGPUs they could use for media acceleration or like the shader arrays and caches are different?

 

[SolidQ]

More like older (or more traditional) engines.

But usually there already a lot fps, 150 or 180 doesn't matter

 

[linkgoron]

- I think RDNA 2 gets brownie points for sort of coming out of nowhere.

5700xt was already competitive with 2070 and better than the 2060S.

Everyone expected 6800XT = 3070 let's go home thanks to AMD's abysmal track record and suddenly BAM competitive stack.

How would that work? 5700xt at ~250mm^2 would be competitive with 2070 but 6800XT with a big die and new uarch was going to competitive with 3070?

AND just to show that I'm not talking in hindsight, here are posts that I made in September/October 2020:

Just above a 3070 would be a total failure for a 500mm^2 card and a new architecture. The 5700xt is around 2/3 of a 2080ti performance-wise in 4K. Getting a double the size card with a new arch only gain ~50%-60% would be pretty terrible.

So AMD targeted big Navi (500+mm^2 7nm card) to fight a 3070 (390mm^2 8nm) which itself is slower than the 2080ti card? They also only gained 50% performance with a new architecture and double the size card? I find that hard to believe. This would put RDNA2 at Vega levels of fail.

I don't see how AMD could've expected anything less than 2080ti + 30-40% at least, for a die shrink.

 

[adroc_thurston]

An abortion like N44 was designed for laptops with iGPUs they could use for media acceleration or like the shader arrays and caches are different?

The former, they're keeping the tradition intact.

AND just to show that I'm not talking in hindsight, here are posts that I made in September/October 2020:

Well yeah plebbit wants every AMD gen to be Vega.
It gives a nice post-purchase justification for every NV slop owner.

 

[Tuna-Fish]

Do desktop ryzens have an usable media engine to go with the minimal igpu? I honestly don't even know.

 

[Win2012R2]

But usually there already a lot fps, 150 or 180 doesn't matter

1) FPS based on old "traditional engines" (that are not CPU limited) is a good proxy for real perf
2) yes, it matters!

 

[gaav87]

If the top skus are 1 euro cent above 1k euro im gona riot

 

[Hail The Brain Slug]

If the top skus are 1 euro cent above 1k euro im gona riot

I think if AMD's first party comparison is against the 7900GRE, then pricing in line with the 7900GRE is probably where it's going to end up.

If AMD wanted to price it at 7900XT or 7900XTX tier, they would be directly comparing it to those SKU's instead.

 

[ToTTenTranz]

seems like they're keeping the tradition of x4 dies being crimes against humanity.

8GB, PCIe x4 and no video codecs?

 

[ToTTenTranz]

40% faster than GRE is the claim from AMD apparently

Why would they compare it to the GRE that costs $550 if the current rumor is pointing to $750 which is about the same as the 7900XT?

 

[Hail The Brain Slug]

Why would they compare it to the GRE that costs $550 if the current rumor is pointing to $750 which is about the same as the 7900XT?

Seems to me like it would lend credence to the $579 price rumor that's been around a while.

"40% more performance for 40% more money" is not exactly a rock solid marketing strategy.

 

[DAPUNISHER]

Do desktop ryzens have an usable media engine to go with the minimal igpu? I honestly don't even know.

From the TPU review -

The iGPU is identical to the one on Ryzen 7000—it is based on the RDNA 2 architecture, with just two CUs or 128 cores. In terms of codecs you get encode support for H.265 and H.264 and decode support for AV1. VP9. H.265 and H.264.

 

[gaav87]

Well the CU are brutal FP16 machines.

 

[ToTTenTranz]

From the TPU review -

Yes, I get the lack of codecs is passable assuming people aren't putting it inside next to an AM4 CPU (though if they do.. auch? imagine not being able to play videos).

But if they castrate the PCIe lanes and cut on the VRAM again, they'll be killing off the GPU entirely. Might as well just not release it at that point.

 

[SolidQ]

Why would they compare it to the GRE that costs $550 if the current rumor is pointing to $750 which is about the same as the 7900XT?

Original price(China) for GRE how many it's was 600-650?

 

[SolidQ]

Well the CU are brutal FP16 machines.

Non XT?

 

[RnR_au]

304W is the official spec but MBA won't be released

Seems to be confirmed...

https://videocardz.com/newz/amd-confirms-radeon-rx-9070-xt-reference-design-wont-be-available-for-purchase

 

[gaav87]

Non XT?

YES non-xt xD
I did some quick napkin math 9070 xt should have higher geekbench ai score than 5080 and 4090 That benchmark is prolly not worth anything but still hilarious xD

 

[gdansk]

Seems to be confirmed...

https://videocardz.com/newz/amd-confirms-radeon-rx-9070-xt-reference-design-wont-be-available-for-purchase

That sucks. Now I have to go look through TPUdb for a card small enough to fit in my case.

 

[gaav87]

That sucks. Now I have to go look through TPUdb for a card small enough to fit in my case.

reaper