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

 

[itsmydamnation]

The what

you know how retailers had cards in NOV/DEC of 2024.

 

[adroc_thurston]

you know how retailers had cards in NOV/DEC of 2024.

Yeah cards that did not have FSR4.

 

[itsmydamnation]

Yeah cards that did not have FSR4.

i agree waiting for FSR4 was the right thing to do. I just asking which date do we count from

 

[poke01]

i agree waiting for FSR4 was the right thing to do. I just asking which date do we count from

Unless a new FSR comes for RDNA 5 we can say late 2026 to early 2027

 

[basix]

Mid to late 2027

That seems to be very late. I can see pros and cons for that, but more cons:
- Xbox Next release end of 2026 contradicts with that
- APU implementations with Zen 6 in Q1...Q2/2027 contradict with that
- PS6 release by end of 2027 could support that
- >2y after RDNA4 release contradicts with that. They had running RDNA4 cards at CES 2025 in January

If I had to speculate on an RDNA5 release date I would suggest Q4/2026 or Q1/2027. With tendency to Q4/2026 (Xbox Next + AT2 release).

 

[marees]

If I had to speculate on an RDNA5 release date I would suggest Q4/2026 or Q1/2027. With tendency to Q4/2026 (Xbox Next + AT2 release).

Going by precedents it would be after Nvidia. So Q1 2027 is my best guess

 

[regen1]

 

[poke01]

HIP Ray Tracing - AMD GPUOpen

HIP RT is a ray tracing library for HIP, making it easy to write ray tracing applications in HIP.

gpuopen.com

RDNA4 support. Hopefully we see improvements in Blender

 

[Win2012R2]

Going by precedents it would be after Nvidia. So Q1 2027 is my best guess

These synced releases only help Nvidia and suppress AMD sales before release.

So AMD needs to get out of this sync releases - much better release much better card with no competition, then later Nvidia releases something better (unlikely by much), but then if everybody knows AMD will bring even better card in a year then why buy Nvidia?

 

[RnR_au]

why buy Nvidia?

Because noone buys AMD... I mean AMD's marketshare is getting close to being indistinguishable from Linux's marketshare...

 

[Win2012R2]

Because noone buys AMD...

I'll speak for myself - I am Nvidia since Riva128 but I will buy top end AMD RDNA5 if it's quicker than 6090, likely buy if it's close (within 5%) too - as long as power not as crazy, and ray tracing is about the same. AMD already appears to have an edge in UE5 games, if Nvidia does not fix driver overhead (which I doubt they will so quickly given their current focus), then this is likely be deal breaker for me.

 

[RnR_au]

... if it's quicker than 6090, likely buy if it's close (within 5%) too - as long as power not as crazy, and ray tracing is about the same.

From my understanding of the rumours, this is unlikely to happen.

 

[adroc_thurston]

From my understanding of the rumours, this is unlikely to happen.

Ah, they're both 192 core parts.
It's gonna be a boring samurai duel.

 

[SolidQ]

Ah, they're both 192 core parts.
It's gonna be a boring samurai duel.

looking at RDNA4 vs Blackwell, usually around same perf 32CU vs 36SM, 64CU vs 70SM. So RDNA5 and Rubin should be ~same

 

[adroc_thurston]

looking at RDNA4 vs Blackwell, usually around same perf 32CU vs 36SM, 64CU vs 70SM. So RDNA5 and Rubin should be ~same

Oh they're very very different things.
Don't think about it too hard.

 

[poke01]

Ah, they're both 192 core parts.
It's gonna be a boring samurai duel.

Nah, those are the best. I wanna see who dies, even if it’s by 1fps

 

[Kepler_L2]

Nah, those are the best. I wanna see who dies, even if it’s by 1fps

Even if AT0 beats GR202 in existing games it won't matter, with how much silicon they're dedicating to AI perf NVIDIA will move the goalposts to some neural rendering BS where AMD is half the performance on NV-sponsored games.

 

[poke01]

Even if AT0 beats GR202 in existing games it won't matter, with how much silicon they're dedicating to AI perf NVIDIA will move the goalposts to some neural rendering BS where AMD is half the performance on NV-sponsored games.

Yeah that’s annoying. Especially if game devs use Nvidia UE5 stack too. This is what monopolies do…

But I will and most others here will judge by pure rasterisation uplift and performance per watt.

 

[RnR_au]

Ah, they're both 192 core parts.
It's gonna be a boring samurai duel.

You have a weird definition of 'boring'

 

[SolidQ]

Especially if game devs use Nvidia UE5 stack too. This is what monopolies do…

Exclusive NV features maybe, but overall in UE5 AMD is fine, and should be also in UE6

Spoiler

 

[basix]

Even if neural rendering stuff gets pushed into games by Nvidia, I expect RDNA5 to be no slouch on that front as well. AMD will for sure add FP4 support and might also double matrix core width. That is not as extreme as 8x width as on Rubin CPX (my expectation: gaming cards will likely get cut-down to 4x) but already very decent for many neural rendering use cases. Nvidias card might be faster, but we are talking about a few percent in probably most cases (and by far less than 2x).

Why AMD will likely extend matrix core performance:

• Neural rendering is kinda new but there are papers out there since at least 2021 (the original Neural Radiance Caching paper) and AMD will bring their own "neural rendering" stuff with FSR Redstone
• Neural rendering techniques can cut down cost. E.g. neural texture compression allows less VRAM and hard disk size. If we extend "neural rendering" to SR, FG and RR it gets even more obvious: You can use a smaller chip to get to similar visual and performance results
• AMD, Microsoft and Sony should look far into the future towards PS7 and Xbox-Next-Next. The more "neural rendering" is supported with strong matrix core acceleration, the easier will be crossgen of PS6 with PS7
  • This trend is kinda obvious, already today: Usage of reural rendering techniques will get more and more prevalent in the future (at least for some parts of the rendering pipeline)

 

[adroc_thurston]

You have a weird definition of 'boring'

Yeah, the non-boring approach is AMD shipping twice as much shaders.

 

[lucasworais]

Ah, they're both 192 core parts.
It's gonna be a boring samurai duel.

Full AT0 will be coming to client?

 

[adroc_thurston]

Full AT0 will be coming to client?

no lol.

 

[lucasworais]

no lol.

So the 192cu part is a cutdown?