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

 

[Josh128]

Non-XT might OC to near XT performance if the power limit can be raised high enough, especially if vram bandwidth is the main limitation.

Its missing 14% of the compute units, I dont think that is possible.

 

[SolidQ]

9 XFX models

https://videocardz.com/newz/xfx-preparing-9-models-of-radeon-rx-9070-xt-graphics-cards

I think for AMD would be best for 9070nonXT put 480\500$ to kill 5060ti 16GB/5070

 

[Hail The Brain Slug]

9 XFX models

https://videocardz.com/newz/xfx-preparing-9-models-of-radeon-rx-9070-xt-graphics-cards

I think for AMD would be best for 9070nonXT put 480\500$ to kill 5060ti 16GB/5070

Why did they even bother making redundant and less desirable non-magair models?

 

[SolidQ]

making redundant and less desirable non-magair

for people who have less money?

So 2xpin models have up 3ghz, 3xpin model 3.1, that mean if you want hard OC, need 3xpin models

 

[Hail The Brain Slug]

for people who have less money?

So 2xpin models have up 3ghz, 3xpin model 3.1, that mean if you want hard OC, need 3xpin models

I get that, but how does it help to make an identical model (mercury) without magair? Does adding the magair fans really bloat BOM so much that having an entire separate model without magair is cost-justified so they can sell it for what, $20 less? $30 less?

It makes sense the swift models are non magair. I mean the actual redundant models where there is both a magair and non-magair option with no other changes.

 

[blckgrffn]

I get that, but how does it help to make an identical model (mercury) without magair? Does adding the magair fans really bloat BOM so much that having an entire separate model without magair is cost-justified so they can sell it for what, $20 less? $30 less?

It makes sense the swift models are non magair. I mean the actual redundant models where there is both a magair and non-magair option with no other changes.

Mercury is the cooling pad setup, I think. That's why I got my Mercury 7900XTX vs other models. Of course, its also a magair but ¯\_(ツ)_/¯

Do fancy fans add more than $10 to the BOM? Are they eager to have enough models to add $100-$200 over "MSRP" and have a strategy to sell that to consumers? That' s my guess.

 

[SolidQ]

Does adding the magair fans really bloat BOM so much that having an entire separate model without magair is cost-justified so they can sell it for what, $20 less? $30 less?

We don't know what behind the scenes, maybe not enought magair components? Anyway we need to see difference in price.

 

[branch_suggestion]

I think for AMD would be best for 9070nonXT put 480\500$ to kill 5060ti 16GB/5070

In a just world NV's lower end parts wouldn't sell.

 

[Hail The Brain Slug]

We don't know what behind the scenes, maybe not enought magair components? Anyway we need to see difference in price.

I'm thinking with AMDs aggressive pricing push they are trying to come up with as many value adds to have models available at lower prices while still having higher margin models at the top end.

 

[gaav87]

Crazy 55C in FURMARK on linux mesa drivers +5800x3d
Scores above 4080

 

[Tuna-Fish]

Oh, the ROPS line got nuked, hmm...

What is interesting it's pixel fillrate is reduced, even vs 7800XT

The pixel fillrate conflicts with the ROP count in the same table. 2970MHz*64 rops = 190.08 GP/s.

So either there are only 64, there are 128 that run at half rate, or the 190.1GB/s figure is wrong.

 

[SolidQ]

FEB 28 lol

 

[gaav87]

 

[Bigos]

I would not compare radeonsi with Windows drivers.

 

[Saylick]

FEB 28 lol

Please tell me these are just placeholders…

 

[gaav87]

549$/599$ surprise.

 

[GTracing]

Its missing 14% of the compute units, I dont think that is possible.

Cut down SKUs often perform better than the core count would suggest. The 5080 has 20% more cores and higher clock speeds than the 5070 ti , but it's only 12-14% higher fps in games.

In TPU's testing, the 5070ti can get up to 12% more performance from overclocking.That puts it only 3% behind a stock 5080.

 

[coercitiv]

Please tell me these are just placeholders…

The news is that product might be up for sale/preorder on Feb 28. Prices are obviously irrelevant.

 

[Tup3x]

I wouldn't expect many MSRP cards now that even AMD doesn't plan to sell their reference cards.

 

[tsamolotoff]

One sample was hitting 3.35Ghz in games at 350-360W with roughly 6-8% more perf.

If it is power starved like N31 and if SPPT is also locked so PL and TDC can't be increased, this doesn't mean much

 

[adroc_thurston]

If it is power starved like N31

No.

and if SPPT is also locked so PL and TDC can't be increased

yes.

 

[gdansk]

I wouldn't expect many MSRP cards now that even AMD doesn't plan to sell their reference cards.

Yes, they are copying parts of the fake price strategy but not the part that leads to favorable but misleading initial reviews.

 

[gaav87]

No.

yes.

Please god no. I want more power tools 450W just for fun.

 

[PJVol]

View attachment 118209View attachment 118210

Linux OpenGL SP results are generally lower than DX11

 

[adroc_thurston]

Please god no. I want more power tools 450W just for fun.

Sorry brother you ain't touching that DVFS network.