Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

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

soresu · May 15, 2024

Mahboi said:
I assume that RDNA 4 will be the occasion for a version bump and we're going to ROCm 7?

ROCm/HIP is mostly tied to their CDNA accelerators.

RDNA compatibility is mostly an after thought at the moment sadly.

Saylick · May 15, 2024

adroc_thurston said:
total VRAM death? daring aren't we

It will be a convenient way to obsolete previous generations of Nvidia GPUs.

That, and only offer Tensor Memory Compression on the latest Nvidia GPU. Check. Mate.

Mahboi · May 15, 2024

Saylick said:
Simple.

In-game AI chatbots. Nvidia will require game devs to implement them, so if your GPU doesn't have enough TOPS, your in-game NPCs will have room temperature IQs and then you can't enjoy the game. You ask game devs why they can't just have conventionally scripted NPC interaction but they'll keep hush or just say "because reasons", meanwhile they're collecting that Nvidia moolah behind their backs.

You can send the check to my mailbox, Nvidia. You're welcome.

But seriously nah, there's an already established plan: slightly tune up your images with ML, scream AIAIAAIAIAIAIAIAIAIAIAIAIAIAIAI for 2 years, and claim that it is Absolutely Peasant™ to play without the Superior DLSS 4 and its AI Master Race™ experience.

Just look at the stupid AI denoiser: they made it into a fancy name, when it's just a denoiser, and they acted like it was a whole new world of gaming. Same scam, same pigeons.

adroc_thurston · May 15, 2024

Saylick said:
It will be a convenient way to obsolete previous generations of Nvidia GPUs.

the new ones don't have moar VRAM.

Mahboi · May 15, 2024

soresu said:
ROCm/HIP is mostly tied to their CDNA accelerators.

RDNA compatibility is mostly an after thought at the moment sadly.

To be entirely fair?
As long as they keep working on making it better, cool.
Just...don't go back to 1 year to get proper consumer flagship support.

Rekluse · May 18, 2024

Joe NYC said:
What percentage of mobile dGPUs are <= 4070 in your opinion?

My guess is about ~75% units, if not more.

So there is more than enough market potential for Strix Halo to compete for. No point to even think about higher tier of dGPU market before enough of this ( <= 4070) market is gained.

The big question is, does Strix Halo give AMD as much margin as a intel/nVidia 4070 combo at a competitive price ? It's be a bit disheartening if it doesn't

inquiss · May 18, 2024

Rekluse said:
The big question is, does Strix Halo give AMD as much margin as a intel/nVidia 4070 combo at a competitive price ? It's be a bit disheartening if it doesn't

It's a margin product, but, for the OEMs too. Enables new use cases and power levels for the performance, saves on design complexity and lowers cooling needs. Once you build a chassis for it you're good for future versions too with the new form factor, should halo be a success

Abwx · May 19, 2024

NTMBK said:
I think you're misremembering- it was an Intel CPU that got the last minute switch (ditching an Athlon), not the Nvidia chipset. And ATi was an independent company back then anyway.

Yes, and since the plateform wasnt still adapted for an Intel CPU they made the public demo with the AMD CPU running the console...

Rekluse · May 20, 2024

inquiss said:
It's a margin product, but, for the OEMs too. Enables new use cases and power levels for the performance, saves on design complexity and lowers cooling needs. Once you build a chassis for it you're good for future versions too with the new form factor, should halo be a success

Outside of the handheld PC gaming wave pioneered by Steam. I feel OEM's have been reticent when it comes to AMD and "new form factors" . Intel seems to be the ones pushing dual screen laptops and NUCS etc rather than the OEM's themselves

marees · May 20, 2024

Rekluse said:
Outside of the handheld PC gaming wave pioneered by Steam. I feel OEM's have been reticent when it comes to AMD and "new form factors" . Intel seems to be the ones pushing dual screen laptops and NUCS etc rather than the OEM's themselves

Maybe a steam tablet / all-in-one / 2-in-one with a dock ??

Ideally should have been surface but for some reason, the surface team, don't seem to be a fan of AMD

tsamolotoff · May 20, 2024

soresu said:
From what I can gather DLSS/framegen and denoising are always going to be a problem while RT remains a focus.

So, maybe RTRT is sort of premature and needs a few orders of magnitude more of mem bw to actually work properly (as the current 'raytracing' is very sparse if even so)? I'm (and probably the majority of people) completely fine with baked lights, shadows etc, also the RTX(tm) titles still use it anyways (and since now you don't need to have an art director and lightning specialists thanks to the magic of RTX(tm), it also looks awful with or without pseudo-RT)

Mahboi said:
The PROBLEM is that Nvidia has no work to do in this regard, their rep is solid and their products are seen as the "default GPU brand you just buy

Yet they consistently do it at the most expert level. Just imagine that the Russian NV tech PR honcho (someone called Oleg Shkoda), was astroturfing his beloved corp day and night on b3d without ever indicating that he's a NV employee. They also had very specific guidelines for the ~~indentured slaves~~ press and tech bloggers with regards to how to review their products which most didn't mind (a shared former CIS thing, I guess)

soresu · May 20, 2024

tsamolotoff said:
So, maybe RTRT is sort of premature and needs a few magnitudes more of mem bw to actually work properly

Memory bandwidth is just part of the problem, the entire paradigm is fundamentally more compute intensive than raster graphics.

That's the thing - with raster it takes more code, but can give very good results with lower end compute hardware.

With RT you can use less code, but you need higher end compute hardware to get a good result, and if you want it in real time then denoising is basically unavoidable for shadowed areas and indirect light (global illumination/GI).

marees · May 20, 2024

soresu said:
denoising is basically unavoidable for shadowed areas.

So nvidia is a de-noising company that sells bundled hardware ala apple ??

soresu · May 20, 2024

marees said:
So nvidia is a de-noising company that sells bundled hardware ala apple ??

Nah they are a hardware company that uses software to leverage their market position.

Without CUDA they wouldn't be nearly so competitive in the professional space.

ToTTenTranz · May 20, 2024

tsamolotoff said:
So, maybe RTRT is sort of premature and needs a few orders of magnitude more of mem bw to actually work properly (as the current 'raytracing' is very sparse if even so)? I'm (and probably the majority of people) completely fine with baked lights, shadows etc, also the RTX(tm) titles still use it anyways (and since now you don't need to have an art director and lightning specialists thanks to the magic of RTX(tm), it also looks awful with or without pseudo-RT)

From what I've read, RTRT is premature because it needs a lot more of everything and on top of that it needs more versatility and proper tools/features for compartmentalization and LODing. Current solutions consist of brute-forcing too much which then brings $2500 GPUs to their knees.

But for Nvidia that's okay because that $2500 GPU is brought to its knees a bit less than the competition and Nvidia's previous $2500 GPU. So not only do they get to shine the competition in a worse light, they also do planned obsolescence on their older graphics cards.

adroc_thurston · May 20, 2024

soresu said:
With RT you can use less code, but you need higher end compute hardware to get a good result, and if you want it in real time then denoising is basically unavoidable for shadowed areas.

the thing with hybrid renderers we have now is that they have both more code and more hardware.
what a pickle.

Saylick · May 20, 2024

adroc_thurston said:
the thing with hybrid renderers we have now is that they have both more code and more hardware.
what a pickle.

Convenient for Nvidia, who want to always sell you a bigger chip while marketing their software solutions as the best.

Oh wait, but they were the ones who pushed for hybrid rendering in the first place. *pondering*

adroc_thurston · May 20, 2024

Saylick said:
Convenient for Nvidia, who want to always sell you a bigger chip while marketing their software solutions as the best.

Well so does AMD lmao.

Saylick said:
Oh wait, but they were the ones who pushed for hybrid rendering in the first place. *pondering*

Well no ImgTec did.

Aapje · May 20, 2024

adroc_thurston said:
the thing with hybrid renderers we have now is that they have both more code and more hardware.
what a pickle.

It leaves a sour taste in the mouth.

adroc_thurston · May 20, 2024

Aapje said:
It leaves a sour taste in the mouth.

it is what it is, 5 years and RTRT hasn't left the gimmick stage.

Aapje · May 20, 2024

adroc_thurston said:
it is what it is, 5 years and RTRT hasn't left the gimmick stage.

True, but that doesn't change the fact that my comment was really just a pickle joke.

beginner99 · May 21, 2024

marees said:
Ideally should have been surface but for some reason, the surface team, don't seem to be a fan of AMD

Maybe they get money from intel. MS surface looks "cool" but the hardware is always outdated and the price on macbook levels. At that point I can just buy a mac and get double the efficiency vs a 3 gen old intel cpu.

adroc_thurston · May 21, 2024

beginner99 said:
MS surface looks "cool" but the hardware is always outdated and the price on macbook levels.

you dropped this banger the day MS actually refreshed it with relevant h/w for once.

branch_suggestion · May 21, 2024

adroc_thurston said:
you dropped this banger the day MS actually refreshed it with relevant h/w for once.

Van Gogh/Rembrandt were relevant hardware too...

adroc_thurston · May 21, 2024

branch_suggestion said:
Van Gogh/Rembrandt were relevant hardware too...

Well ugh VGH was supposed to be in a Surface.
But MS gotta MS.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Platinum Member

Diamond Member

Senior member

Platinum Member

Senior member

Member

Member

Lifer

Member

Member

Junior Member

Platinum Member

Member

Platinum Member

Member

Platinum Member

Diamond Member

Platinum Member

Golden Member

Platinum Member

Golden Member

Diamond Member

Platinum Member

Senior member

Platinum Member