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

 

[DisEnchantment]

You sure it's RDNA 4 and not RDNA 3.5?

Oh yes, RDNA3.5 needed less than a dozen patches to support the new HW, RDNA4 needed more than a hundred patches.

LLVM

Pull requests · llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - Pull requests · llvm/llvm-project

github.com

Mesa got a massive patch from Marek too

https://gitlab.freedesktop.org/mareko/mesa/-/commits/amd-gfx12-initial?ref_type=heads

 

[Panino Manino]

Battlemage is looking good.
AMD getting punched from above and bellow.

 

[marees]

AMD's desktop sales crashed by almost half.
Otoh, Nvidia managed to maintain their numbers

Die Grafikchip- und Grafikkarten-Marktanteile im ersten Quartal 2024 | 3DCenter.org

Die Marktforscher von 'Jon Peddie Research' berichten in zwei Artikeln – No.1 & No.2 – zu den weltweiten Marktanteilen für Grafikchips im abgelaufenen ersten Quartal 2024. Der interessantere Teil ist hierbei immer der AiB-Report (AiB =

m-3dcenter-org.translate.goog

So AMD should be having lots of unsold inventory

 

[marees]

AMD's desktop sales crashed by almost half.
Otoh, Nvidia managed to maintain their numbers

View attachment 100655

Die Grafikchip- und Grafikkarten-Marktanteile im ersten Quartal 2024 | 3DCenter.org

Die Marktforscher von 'Jon Peddie Research' berichten in zwei Artikeln – No.1 & No.2 – zu den weltweiten Marktanteilen für Grafikchips im abgelaufenen ersten Quartal 2024. Der interessantere Teil ist hierbei immer der AiB-Report (AiB =

m-3dcenter-org.translate.goog

So AMD should be having lots of unsold inventory

In other news, TSMC wants to increase chip prices
& the move is welcomed by Jen Hsun Huang

This is going to crush Radeon's (16%) margins
(Intel probably has tonnes of shareholder money to burn thru further)

 

[moinmoin]

AMD's desktop sales crashed by almost half.

Crashed? While I know what you refer to, there is a reason why Y/Y comparisons are generally preferred over Q/Q comparisons.

 

[blckgrffn]

Crashed? While I know what you refer to, there is a reason why Y/Y comparisons are generally preferred over Q/Q comparisons.

Yes, business is almost always YoY measurements, whether that is months or quarters.

Looks like they got a solid 42% growth in units shipped YoY, which is... a lot. Basically fully absorbed the Intel market share back.

 

[marees]

Yes, business is almost always YoY measurements, whether that is months or quarters.

Looks like they got a solid 42% growth in units shipped YoY, which is... a lot. Basically fully absorbed the Intel market share back.

What you are saying is true would be true in general circumstances, however mindfactory numbers show that AMD sales have crashed in the last 1 month (compared to first quarter figures)

Check these numbers

https://twitter.com/x/status/1796965478939713684

 

[nurturedhate]

Side note:

here ya go, been running I think 3 or 4 days now
View attachment 100638
Son did wow raid yesterday on this, I only play D4.

Obviously T1 and T3 are glitched because I wouldn't be typing on this if it hit 224C

Same. Sapphire 7900 xtx Nitro+ in WoW Classic Era running around Stormwind for 5 min. Close enough to 3k. Plenty of games bump it above 3k.

 

[nurturedhate]

More WoW

 

[blckgrffn]

What you are saying is true would be true in general circumstances, however mindfactory numbers show that AMD sales have crashed in the last 1 month (compared to first quarter figures)

Check these numbers

https://twitter.com/x/status/1796965478939713684

That's... not really that much data though. A grain of sand on the beach.

The earlier data, if accurate, is much more convincing in terms of volume, given the source is trusted. And that one paints a a picture of relative equilibrium and bouncing back from an epically terrible PC hardware year industry wide.

I co-own an ecommerce business and trust me I know all about recency bias because I am constantly fighting my urge to act on it. Yesterdays sales were up 200% YoY? I am having a great evening. Yesterdays sales 50% of the same day YoY? I am sad, pack it up. This is all blatantly preposterous, I know that on one level but I feel it on another.

Let's take a little longer view and let the chips fall.

Tying this to the battlemage thread, I am excited about the hype but Intel has burned me out a bit over the last few years. Let's see them deliver - and get back to the glory of shipping enough units to count for more than a percentage point - then I will get back on the hype train vs standing on dubious hill where I am still standing.

 

[SolidQ]

however mindfactory numbers show that AMD sales have crashed in the last 1 month

Super series is out. 4070S > 7800XT. 4070ti ~ 7900XT, but NV cards have better Upscaling and RT.
That AMD is missed. FG is ok, but need RT and Upscaling, that why alot people choose NV

 

[adroc_thurston]

but need RT and Upscaling, that why alot people choose NV

The shill goalposts will move overnight if they win that.
What they need is a shotgun.

(Intel probably has tonnes of shareholder money to burn thru further)

Intel is hilariously cash-poor lmao.

 

[Ajay]

Battlemage is looking good.

I'll believe that when I see benchmarks.

 

[SolidQ]

I'll believe that when I see benchmarks.

There is claiming up 50%. I don't think it compete even with RDNA4

The shill goalposts will move overnight if they win that.

They need TIE at least for start, also make something new for Radeon cards and heavily marketing it

 

[Mahboi]

Battlemage is looking good.
AMD getting punched from above and bellow.

Call me a conspiracy theorist, but I'm starting to severely doubt everything Intel showed at Computex.

First there was Skymont's "workloads" benchmarks. That's a funny benchmark, "workloads". People accused AMD of having published fake numbers for Zen 5 and used Geekbench AES encryption as an example of "highly cherry picked", but highly cherry picked at least had names on the cherries. Skymont could just as well be the most obtuse workloads ever.

Also, Intel has a tendency to overstate their results, while AMD has the opposite. I'm very much expecting to see Zen 5 do closer to 17 or 18%, while Skymont may be quite a bit below 38%. Wait and see, or so I thought.

But now, we're hearing about "Battlemage being great". Ok. No. This is where I stop believing what Intel says.
Alchemist was a textbook example of doing 25 things at the same time, somehow their first gen GPU based off Intel UHD was going to also implement the biggest raytracing implem there ever was. The drivers were a mess, the general quality was a mess, area was huge, the MSI Claw is ridiculously bad at power draw AND stability while fighting borked RDNA 3.

And all this time, we heard nothing about BMG area, nothing about BMG at all really, just nothing. Drivers getting slowly better but still far from fully reliable, as the Claw showed about 2 months ago.

And now, supposedly, after ACM was a vile bust, and BMG is even more late vs its targeted competitors (RDNA 3 and Lovelace), we apparently should have "high expectations".

I don't buy it.
And all the BMG talks is making me (a little) doubtful of even Skymont.
More than ever, wait for reviews.

 

[gdansk]

There is claiming up 50%. I don't think it compete even with RDNA4

They need TIE at least for start, also make something new for Radeon cards and heavily marketing it

50% over MTL which was already quite a bit more efficient and while using less power.
I think it is about 2.5x ppw on N3B compared to Alchemist.

 

[adroc_thurston]

I think it is about 2.5x ppw on N3B compared to Alchemist.

Wide and very slow config on the nicest node besides N3e is a rather bad comparsion point to the DT chungus@like 2.2 on desktop.

 

[gdansk]

Wide and very slow config on the nicest node besides N3e is a rather bad comparsion point to the DT chungus@like 2.2 on desktop.

I only had random screenshots of LNL to compare to. So it's hardly a good guess in anycase.

But just wanted to point out that 50% is most probably an understatement of what it can achieve relative to Alchemist.

 

[adroc_thurston]

But just wanted to point out that 50% is most probably an understatement of what it can achieve relative to Alchemist.

Oh yeah, but ACM was piss-poor on every PPA metric in its big boy incarnation while N5 iGP configs fared (relatively) a lot better.
Basically you're shooting birdshot in the dark.

 

[tsamolotoff]

Plenty of games bump it above 3k.

N31 is extremely PL limited so you get high clocks in not that demanding games (including mostly CPU-limited MMOs) and in the other games and stress tests it drops to rock bottom or starts to eat upwards of 800W (check evc2se TS/TSE OC results at OCN). Either AMD aimed at 600W tdp initially or something is really broken either in some raster-related shit inside the CUs or the fixed part of the pipeline (whatever it is called now)

 

[Mahboi]

https://twitter.com/x/status/1798718324639187078

Yet again I don't really buy the Wccfmeme report on RDNA 4 by 2025.
Seems way too late when they should be beyond ready by September. And unsurprisingly RDNA 3 is very, very, VERY much spent.

 

[SolidQ]

I don't really buy the Wccfmeme report on RDNA 4 by 2025.

Same. I don't see reason launch it with RTX 5070/5060

 

[ToTTenTranz]

There is claiming up 50%.

+50% in performance per Watt, and the GPU topped at ~200W.

They can launch a 300W GPU with 2.25x the performance of the Arc A770, which would place it between a 4070 Ti and a 4070 Ti Super. Perhaps closer to the 4080 in raytracing performance since Battlemage seems to have significantly upgraded its RT units.


Is that a groundbreaking product? No, but if it's price correctly then it may be a very appealing product for the 20% of DIY PC gamers that haven't superglued their eyes to the green brand.

 

[Mahboi]

"The whisper on the show floor"
Ok so...a big rumour that unless the source gets specified, could just be some troll that threw that out for funs?
Oh wait, we have RGT confirming it, clearly going to be true.

 

[DaaQ]

You are running a FX-8350 in 2024. You have a 7900xtx and are telling me how things are while running it with a FX-8350??? Is that even PCI-3? Those numbers mean nothing. Your link speed is 5GT's which is PCI-2 (PCI express generation 2.0). I see the 3.2ghz numbers but I do not believe those are real world GPU core clock speeds. For starters it list 3.2ghz speed with 0.0% utilization in one of the tiers.

You have to use a benchmark or play a game with Adrenaline reading the GPU clock speed. Your system is heavily CPU bottlenecked, so there probably is not point to this. You need to upgrade your CPU to at least a Zen 3 or Zen 4 with PCI-4.0 to take advantage of a 7900xtx. That requires a new motherboard.

Does not the GPU utilization in hwinfo64 not suffice? would steel nomad suffice? PCIE feature test suffice? Did 20 loops of speedway, I will only do FSE or yltra without cpu test because yea it is 2fps on that test, Ill do the 2 gpu tests.

What does 100% GPU usage and 80% CPU usage mean?
EDIT:
BTW you are saying 20-30% of the chip is not functioning, then end your post with you are speculating.

This card performs alot different than the Dec 2022 reviews. Put your money out and test one.

2nd EDIT: look at the 3dmark site, you can see mine in there also, REALLY easy to look up. CPU GPU search.

3rd edit here ya go.