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

 

[jpiniero]

An 8GB 128bit card at that price point though will be absolutely destroyed because Aussie Steve will throw out an Indiana Jones at 1080p benchmark where the 12GB B580 is playable and the 8GB 9060 either does not load the game at all or is in the single digits.

Consider that marketing to encourage people to buy the 16.

 

[maddie]

I doubt they will do an 8GB SKU at $199 though. The 16GB SKU if it performs around the 7700XT / 6800 / 4070 would make a great card at $330 and it would be a direct replacement for the 7600XT.

The next step down below $330 is around $250 and as you say at that price point 12GB is a requirement.

Given the size of N44 I don't see yields being so bad that there is room for more than 2 SKUs.

So yea. If I was AMD 12GB 96bit @ $250 and it should roughly match the B580 which was well received and not have the driver overhead issue. That would be a winner.

An 8GB 128bit card at that price point though will be absolutely destroyed because Aussie Steve will throw out an Indiana Jones at 1080p benchmark where the 12GB B580 is playable and the 8GB 9060 either does not load the game at all or is in the single digits.

As another member suggested, keep producing the 8GB RX 6600 models on 6nm for the lowest end.

 

[Timorous]

As another member suggested, keep producing the 8GB RX 6600 models on 6nm for the lowest end.

N33 is on N6 as well so might as well just lower the price of the RX 7600 below $200 and that would do the same job.

 

[Timorous]

Consider that marketing to encourage people to buy the 16.

Given the 9060 XT 16GB is likely to be priced between the 5060 8GB and 5060Ti 8GB the marketing of it working in that title where those 8GB cards don't is even better. Also why waste the supply on the 8GB variants, they will just sit on shelves.

 

[Ghostsonplanets]

I assume the MSRP will be higher than the 7600 / XT was (which was $269 for the 8 and $329 for 16).

That's DOA for the 8GB model. Would be absolutely blasted by media.

So yea. If I was AMD 12GB 96bit @ $250 and it should roughly match the B580 which was well received and not have the driver overhead issue. That would be a winner.

I agree. A 96-but 12GB $249 would be a winner. Specially if it can match 6750XT/6700XT

An 8GB 128bit card at that price point though will be absolutely destroyed because Aussie Steve will throw out an Indiana Jones at 1080p benchmark where the 12GB B580 is playable and the 8GB 9060 either does not load the game at all or is in the single digits.

That's my thinking as well. 8GB VRAM would be destroyed by media. And upcoming mainstay benchmark titles like AC Shadows and Doom Dark Ages are requiring 8GB VRAM as minimum for 1080p Low settings.

N33 is on N6 as well so might as well just lower the price of the RX 7600 below $200 and that would do the same job.

I agree with this. N6 N33 should be plenty cheap to be made these days and AMD could afford a price drop on the 7600 to own the <$200 market.

 

[jpiniero]

That's DOA for the 8GB model. Would be absolutely blasted by media.

Maybe they won't sample to reviewers then, only the 16.

 

[marees]

Leaked reviews

https://twitter.com/x/status/1897278618587554059

 

[Timorous]

9070 XT stock at OCUK seems healthy.

we have several thousand cards in stock spanning across Sapphire, PowerColor and Asrock all of which have 1000+ in stock of each brand.

 

[PJVol]

Leaked reviews

Lol, they've got 9070 = 5070ti on average fps.

 

[Panino Manino]

IT BEGINS!

Sapphire Radeon RX 9070 XT Nitro+ Review - Beating NVIDIA

AMD's new RDNA 4 GPUs are launching today. The Sapphire Radeon RX 9070 XT Nitro+ is a fantastic custom design with powerful cooling and whisper-quiet fans. In terms of FPS, the RX 9070 XT offers competitive performance with similarly priced NVIDIA options in both raster and ray tracing.

www.techpowerup.com

 

[jpiniero]

Benchmarks look like all over the place.

 

[branch_suggestion]

Membw bound game, example A.

 

[Panino Manino]

Benchmarks look like all over the place.

It's a 5070ti on average with 4070ti RT.

A big improvement, 96CU performance with 64CU but again, it's only an average sized chip.

 

[Panino Manino]

FSR4 tested by DF:

EDIT: FSR4 Performance is better than DLSS4 CNN but worse than DLSS4 Transformer.

 

[branch_suggestion]

FSR4 is super low overhead, impressive.

 

[gdansk]

ComputerBase has the 9070 and XT leading by a large margin in fps/w when limited to 144 fps. If that's accurate then it is very disappointing these aren't in laptops.

 

[jpiniero]

Membw bound game, example A.

Sounds about right.

 

[exquisitechar]

Solid effort from AMD. Nvidia is still ahead, but they’re more competitive than they’ve been in forever without relying on a node advantage as a crutch.

 

[jpiniero]

without relying on a node advantage as a crutch.

Technically it's on N4P versus the "5 nm" 4N that Blackwell Gaming uses.

 

[GTracing]

FSR4 tested by DF:

EDIT: FSR4 Performance is better than DLSS4 CNN but worse than DLSS4 Transformer.

It's definitely really good. A night and day difference compared to FSR3. Just to add context for those that don't watch the video, the image quality is better, but it's more computationally heavy (they compared FSR4 performance vs DLSS performance). Overall DLSS3.8 vs FSR4 trade blows.

 

[SolidQ]

F1 2024 1% is good in RT mode

 

[sl0519]

It seems that the 9070 is being overlooked. It should've been the real star of the RDNA4 lineup. I assume AMD’s original plan was to position the 9070 to match the performance of the RX 7900 XT, but they somehow managed to push the clocks higher, bringing it closer to the RX 7900 XTX. As a result, they decided to introduce an additional SKU (9070XT). Based on TPU’s review, the 9070's efficiency is roughly on par with the RTX 5070 Ti, which suggests that the 9070 XT is clocked excessively high, leading to a significant drop in efficiency. A better approach would have been to increase the number of CUs while reducing clock speeds to maintain a more balanced power-to-performance ratio.

 

[inquiss]

It seems that the 9070 is being overlooked. It should've been the real star of the RDNA4 lineup. I assume AMD’s original plan was to position the 9070 to match the performance of the RX 7900 XT, but they somehow managed to push the clocks higher, bringing it closer to the RX 7900 XTX. As a result, they decided to introduce an additional SKU (9070XT). Based on TPU’s review, the 9070's efficiency is roughly on par with the RTX 5070 Ti, which suggests that the 9070 XT is clocked excessively high, leading to a significant drop in efficiency. A better approach would have been to increase the number of CUs while reducing clock speeds to maintain a more balanced power-to-performance ratio.

Nah, the xt is the non cut down product. So you've got this backwards. They'll have loads more xt than non xt

 

[Heartbreaker]

It seems that the 9070 is being overlooked.

The pricing makes it irrelevant until AMD fixes it.

 

[jpiniero]

The pricing makes it irrelevant until AMD fixes it.

Don't need to. Just as long as they manage supply appropriately.