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

 

[adroc_thurston]

Wouldn't be surprised by a 32GB "Budget AI, but we'll mark it up quite a bit higher than the cost of the memory" SKU at some point, if it's possible.

Radeon Pro's do have clamshells so yea.
But pedestrian pleb parts are getting 16G and they're gonna like it.

 

[Shmee]

I don't get all the hype for RT. I would rather have a good card like 7900XTX, without burning melty connector, affordable, and with 24GB of memory, and good performance.

 

[adroc_thurston]

I don't get all the hype for RT. I would rather have a good card like 7900XTX, without burning melty connector, affordable, and with 24GB of memory, and good performance.

There is no hype, it's just something to bash AMD over.

 

[Shmee]

That doesn't make any sense.

 

[adroc_thurston]

That doesn't make any sense.

It's one of the many ways people do post-factum purchase justification.

Well AMD sucks at RTRT, glad I bought Nvidia!

 

[Shmee]

Ah I see what you are saying. Yeah, it's kind of stupid IMO, I would rather have high FPS, but to each their own I guess.

 

[blckgrffn]

Ah I see what you are saying. Yeah, it's kind of stupid IMO, I would rather have high FPS, but to each their own I guess.

FRAME GEN!!! BOOM!!!

lol, I kid. I hear you.

Personally I am invested in FreeSync premium, the overlays and all that. That’s right, I am here for Team Red for the drivers and UI. Every time I tip back into the Nvidia side their software and log in requires GeForce Xtreme Xperiwnce slaps me across the face. I know they do offer some sweet game profiles and DLSS is the cats meow but I just want a solid gui and raster performance for the games I play today.

And yes, I do get frustrated when the AMD drivers aren’t playing nice too.

 

[igor_kavinski]

Why are we all acting like 5700XT part deux is going to be some sort of dark horse AI champ. Did I miss something?

They have supposedly gotten the power down to 220W so compared to 7800 XT, it should perform more efficiently and approach 7900 XT performance.

 

[poke01]

Every time I tip back into the Nvidia side their software and log in requires GeForce Xtreme Xperiwnce slaps me across the face

Ugh, I hated that. The new Nvidia is better but not as good as AMDs.

IMO, Nvidia has always been bad at app development, it’s always so clunky. Same with their website design.

 

[SolidQ]

Ah I see what you are saying. Yeah, it's kind of stupid IMO, I would rather have high FPS, but to each their own I guess.

Because NV brainwash a lot people with RT, and they need sell their cards. RT is fine tech, but still early and limited

 

[adroc_thurston]

but still early

We're a wee bit over 6 years of RTRT hardware being in HVM.

 

[gaav87]

Guy's RDNA4 = 4070ti super ~7900xt / between 5070 and 5070ti. ~+30% to 7800xt
Don't let in on the hype for 4080/7900xtx mess with you... Remember rdna3 guys...

 

[poke01]

We're a wee bit over 6 years of RTRT hardware being in HVM.

Not in games but in 3D rendering, hardware RT has a huge been improvement. Nvidia is the standard for 3D work because of this.

For games, RT won't matter till becomes even more mainstream. $500 consoles need to have 4080 Super levels of RT perf and that won't happen till PS6.

 

[misuspita]

Yeah, for once I'd like people calm the 🤡 down and stop the hype train from where they will step down inevitably disappointed.

You always overhype AMD products. Always. Every single product since ATI has been bought ended in a moonshot hype train

 

[SolidQ]

$500 consoles need to have 4080 Super levels of RT perf and that won't happen till PS6.

PS6 wil be more powerful, than 4080

 

[adroc_thurston]

Not in games but in 3D rendering, hardware RT has a huge been improvement. Nvidia is the standard for 3D work because of this.

Not a particularly relevant market segment at all.
RTRT progress in games has been lukewarm at best.

 

[poke01]

DLSS is the cats meow

Having used FSR2 and DLSS. AI based upscaling is much better, even XeSS is better than FSR2. It does make a difference to a lot of people. DLSS is what Intel and AMD need to beat, its mindshare is massive.

 

[adroc_thurston]

DLSS is what Intel and AMD need to beat, its mindshare is massive.

Shills will move the goalposts in 3 picoseconds.

 

[poke01]

Not a particularly relevant market segment at all.

Nvidia gaming cards have been used for Blender because of OptiX.

RTRT progress in games has been lukewarm at best.

Agree

 

[adroc_thurston]

Nvidia gaming cards have been used for Blender because of OptiX

Tiny market.

 

[poke01]

Shills will move the goalposts in 3 picoseconds.

So how hard is it to break Nvidia's mindshare? Do they need fail like Intel? I mean AMD did with Ryzen, who knows maybe will.

 

[adroc_thurston]

So how hard is it to break Nvidia's mindshare

You can't.

Do they need fail like Intel?

Pretty much lol.

 

[SolidQ]

So how hard is it to break Nvidia's mindshare?

Software - That should be big priority, a lot people i see choose because software/Cuda
RoCm to be similar Cude and better
Better Media Engine
Upscaler same quality or better, than DLSS
Fix Idle Power
more Vram for good Price

 

[poke01]

Software - That should be big priority, a lot people i see choose because software/Cuda
RoCm to be similar Cude and better
Better Media Engine
Upscaler same quality or better, than DLSS
Fix Idle Power

Well put. Software and efficiency is key.

AMD already provides a lot of VRAM which is great, just need to work on the above.

 

[SolidQ]

AMD already provides a lot of VRAM which is great, just need to work on the above.

if AMD gonna start N48 with 500$ + FSR4/RT is good, that gonna take some market share from NV. Live that was with RV770 cheap, but very good perfomance

efficiency is key.

Some generation AMD is good with efficiency, some is NV