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

 

[gdansk]

If these are accurate, why would RDNA3.5 be a long-lived architecture?
RDNA4 seems like a pretty big improvement over that.

 

[adroc_thurston]

If these are accurate, why would RDNA3.5 be a long-lived architecture?

Because things like MedusaDT/gator range exist.

RDNA4 seems like a pretty big improvement over that.

5 even moreso.

 

[SolidQ]

5 even moreso.

will 5 fix PT perf or that another big jump in RT?

 

[adroc_thurston]

will 5 fix PT perf or that another big jump in RT?

what the hell is "PT perf".

 

[SolidQ]

what the hell is "PT perf".

Path Tracing. Weakest point for AMD

 

[coercitiv]

what the hell is "PT perf".

Path Tracing, the big carrot all the media likes to chase because we're the ones paying for it, not them.

 

[Tuna-Fish]

what the hell is "PT perf".

Just pure RT, with no raster at all.

 

[sandorski]

Path Tracing. Weakest point for AMD

Seems likely something they improved.

 

[adroc_thurston]

Path Tracing. Weakest point for AMD

That's just ray tracing. Just more of it, with more divergence.

Path Tracing, the big carrot all the media likes to chase because we're the ones paying for it, not them.

Snooze.

Just pure RT, with no raster at all.

That's just RT.
RT with raster is hybrid rendering.

 

[gdansk]

Path Tracing. Weakest point for AMD

The feature which reduces two different $1600 Nvidia GPUs to ~17ms frametimes at 1920x1080? For some reason I'm not too worried about needing to turn that crap on. Even in the near future.

 

[SolidQ]

Seems for RDNA4 also PT weak point. AMD didn't compare GRE vs 9070 in Cyberpunk/Alan Wake 2

Will have RDNA4 similar to NV SER, that question

 

[adroc_thurston]

Seems for RDNA4 also PT weak point. AMD didn't compare GRE vs 9070 in Cyberpunk/Alan Wake 2

They're sponsored titles lmao.

Will have RDNA4 similar to NV SER, that question

Hopefully, no.
That's a really really really bad crutch that sits in a way of programmable RT pipelines.

 

[SolidQ]

They're sponsored titles lmao.

They compare in RT mode, but not in Overdrive.
Why not show like 3x times faster than GRE in overdrive mode

 

[adroc_thurston]

They compare in RT mode, but not in Overdrive.
Why not show like 3x times faster than GRE in overdrive mode

Because it's a sponsored title.

 

[SolidQ]

White Red Devil

https://videocardz.com/newz/powercolor-preparing-radeon-rx-7650-gre-reaper-spectral-white-edition-white-rx-9070-xt-red-devil-also-planned

 

[Heartbreaker]

The best way to compare costs across nodes is to multiply the die area by the cost-per-wafer.

Any comparison with a 50%+ margin for error is useless.

Actual full costs per/wafer are nothing but guesses stacked upon guesses.

 

[Jan Olšan]

White Red Devil

https://videocardz.com/newz/powercolor-preparing-radeon-rx-7650-gre-reaper-spectral-white-edition-white-rx-9070-xt-red-devil-also-planned

Nice name. Right up there with Sapphire Radeon RX 7650 GRE Black Diamond :>

 

[SolidQ]

Because it's a sponsored title.

They need own sponsored tiles for Path Tracing.

 

[adroc_thurston]

They need own sponsored tiles for Path Tracing.

they need a $3k SRP target halo lol.

Actual full costs per/wafer are nothing but guesses stacked upon guesses.

they're not guesses.
N4 is like $15.5k a wafer.

 

[Heartbreaker]

Navi 21 N7 ~27B transistors
Navi 48 N4 ~53B transistors, almost 100% increase

I agree that tracking transistor costs is useful, but using it for die costs estimates of different designs might lead to very weird results. 6900 XT launched for $1000, and it looks like N48 needs even higher price for similar margins. Even if we take 6800XT MSRP as guideline, the math still looks bad.

Using die size would likely be equally problematic (even including the guesses about wafer costs).

I sadly expect that the Navi 48 won't end up very profitable for AMD, and die costs are likely why they are avoiding going larger.

 

[adroc_thurston]

I sadly expect that the Navi 48 won't end up very profitable for AMD

no it's bretty cheap.

and die costs are likely why they are avoiding going larger.

no, everything 'larger' was a chtonic SoIC monstrocity.

 

[gaav87]

no it's bretty cheap.

no, everything 'larger' was a chtonic SoIC monstrocity.

128cu 112cu glued comming ?

 

[adroc_thurston]

128cu 112cu glued comming ?

Uh. No? Nope.

 

[gaav87]

Uh. No? Nope.

Crazy take glued n48 on gddr6 + glued defects with cut down phys also gddr6 possible ?

 

[jpiniero]

they're not guesses.
N4 is like $15.5k a wafer.

I'll take the over.