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

 

[GodisanAtheist]

The 70 in 9070 suggests this is a 1440P class GPU. 64 ROPs is fine for 1440P, but likely means the card will drop off pretty hard at 4K and above, which normally would be fine because there would be a 96/128 ROP card sitting above it.

 

[SolidQ]

The 70 in 9070 suggests this is a 1440P class GPU. 64 ROPs is fine for 1440P, but likely means the card will drop off pretty hard at 4K and above

one moment.
" Our focus with this generation is to address the highest volume portion of the enthusiast gaming market with our new RDNA 4 architecture. RDNA 4 delivers significantly better ray tracing performance and adds support for AI powered upscaling technology that will bring high quality 4K gaming to mainstream players when the first Radeon 9070-series GPUs go on sale in early March” - Dr. Lisa Su

At least we need confirmation how many ROPS.

 

[coercitiv]

one moment.
" Our focus with this generation is to address the highest volume portion of the enthusiast gaming market with our new RDNA 4 architecture. RDNA 4 delivers significantly better ray tracing performance and adds support for AI powered upscaling technology that will bring high quality 4K gaming to mainstream players when the first Radeon 9070-series GPUs go on sale in early March” - Dr. Lisa Su

At least we need confirmation how many ROPS.

No need, I'll translate corporate slang for you:

• enthusiast -> $500 and above
• 4K gaming -> 4K FSR4 Quality/Performance

 

[GodisanAtheist]

No need, I'll translate corporate slang for you:

• enthusiast -> $500 and above
• 4K gaming -> 4K FSR4 Quality/Performance

- Never forget that the RX 7900XTX was AMD's first 8K GPU

 

[gaav87]

28th... 1 week after 5070ti lol

https://twitter.com/x/status/1890102891119276284

 

[jpiniero]

28th... 1 week after 5070ti lol

They really are going to decide on the price based upon the 5070 Ti benchmarks. Granted it's going to be very questionable as to whether any 5070 Ti will be sold at $749, so...

 

[gdansk]

They really are going to decide on the price based upon the 5070 Ti benchmarks. Granted it's going to be very questionable as to whether any 5070 Ti will be sold at $749, so...

Many 5070 Ti reviews will still talk about this imaginary pricing.

Honestly, AMD has to do the same. A low, low MSRP is a must even if they have no intent to honor it.

 

[linkgoron]

28th... 1 week after 5070ti lol

https://twitter.com/x/status/1890102891119276284

Will be pretty funny if they still don't announce pricing on that date.

" Our focus with this generation is to address the highest volume portion of the enthusiast gaming market with our new RDNA 4 architecture. RDNA 4 delivers significantly better ray tracing performance and adds support for AI powered upscaling technology that will bring high quality 4K gaming to mainstream players when the first Radeon 9070-series GPUs go on sale in early March” - Dr. Lisa Su

It, sadly, clearly states that 4k will be possible thanks to "AI powered upscaling technology".

 

[SolidQ]

Should be call Sapphire RX 9070 XT Brick Edition, instead Sapphire Nitro

 

[gaav87]

Should be call Sapphire RX 9070 XT Brick Edition, instead Sapphire Nitro

Even the GPU is sad its using 12v xD

 

[MrTeal]

The cutout and magnetic cover look great. That's going to let you do some really nice cable routing.

I really wouldn't be worried about 12x6 in a 300W card the same way I would in a 575W card, especially if it's implemented more like a 3090 Ti than a 5090.

 

[tajoh111]

Either 285k bottlenecks hard in monster hunter or its f'in bearly faster than 6800xt and 4070s

View attachment 116978

According to videocardz, that is high settings with FSR and frame generation enabled.

 

[Bryo4321]

https://twitter.com/x/status/1890123828367601764

So there goes that rumor.

 

[SolidQ]

So there goes that rumor.

9800 then?

 

[jpiniero]

9800 then?

I was gonna say XTXXX but that works too.

 

[Bryo4321]

9800 then?

Maybe, sounds like it would either be pro or a different sku. 9070 XTX?

 

[gaav87]

According to videocardz, that is high settings with FSR and frame generation enabled.

No its no high settings u can change language to japanese and high looks diff. its extreem high setting but its missing from the regular benchmark and cba downloading the beta to do 1 bench.

 

[gdansk]

It's the same as last generations. Double memory on the workstation card. So there will likely be a 32GB Navi 48 but it won't be affordable nor available to consumers.

 

[reaperrr3]

It's the same as last generations. Double memory on the workstation card. So there will likely be a 32GB Navi 48 but it won't be affordable nor available to consumers.

Why would AMD waste this opportunity, though?

The only way they can get a meaningful advantage out of a 32GB card is if it's significantly cheaper than the 5090.

The idea here is to have a card sit above the 9070 XT that they can ask 300-500$ more for even though it's barely faster and only like 50$ more expensive to make, but takes away "local AI LLM that fit into 32GB" desktop marketshare from the 5090.

 

[gdansk]

but takes away "local AI LLM that fit into 32GB" desktop marketshare from the 5090.

Will it? 🤔
And what stops Nvidia from making a 32GB GB203 as a counter? Fear of cannibalizing the 5090? Well they can't make enough of them anyway so what's to fear.

 

[RnR_au]

It's the same as last generations. Double memory on the workstation card. So there will likely be a 32GB Navi 48 but it won't be affordable nor available to consumers.

Does this make sense though? The last generation workstation card did 48GB. Why go backwards in a pro card in an AI era that demands more vram?

 

[gdansk]

Does this make sense though? The last generation workstation card did 48GB. Why go backwards in a pro card in an AI era that demands more vram?

They went backwards in performance and memory interface. And so also memory too.

 

[RnR_au]

They went backwards in performance and memory interface. And so also memory too.

Not sure I follow... the W7900 vs the W6800 increased membw and vram. Just going by the techpowerup db.

 

[gdansk]

Not sure I follow... the W7900 vs the W6800 increased membw and vram. Just going by the techpowerup db.

N48 is the best they can do this generation so there is no other option. Unless they want to keep N31 in production (the lack of stock suggests they do not). If there is a 32GB variant floating around I assume it is for the W9070 (or whatever they call it) rather than some consumer-focused card.

 

[MrTeal]

Will it? 🤔
And what stops Nvidia from making a 32GB GB203 as a counter? Fear of cannibalizing the 5090? Well they can't make enough of them anyway so what's to fear.

Nothing, Nvidia probably will make a RTX 5000 Blackwell with 32GB and a RTX6000 Blackwell with 64GB. RTX 5000 Blackwell will certainly be more expensive than a pro version of Navi 48.