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

 

[tsamolotoff]

Could be that the mobile versions are pushed back, who knows. Although some $3000 price therapy might force some people to rethink their credo of 'radeon exists to lower prices on geforce'

 

[marees]

To me, it has not so much sense. If reason is price, you can always adjust it later. If the product is ready, then delaying it means only losing potential sales. A real reason would be excess inventory of older cards that must be cleaned up before launching the new ones (which could be).

This is shocking yes. But people had started speculating a month back (after Lisa's Guidance for less GPU/console sales in H2 2024)

The 3 things that make me worried that could delay RDNA 4 are:

1) AMD is readying FSR 4 with AI upscaling & wants to launch RDNA 4 along with it
2) overstock of RDNA 3 (should not have happened if AMD had learnt lessons from RDNA 2 overstock debacle)
3) unsold RDNA 2 — for ex: the 6800 selling for $360 in NewEgg. (Should not be a major concern as most Navi 21 is sold out.)

On the pro side:
1) the 7900 GRE is being discounted to $510. Surely it is more profitable to sell navi 48 for $550-$600 (unless there is tons of 7900 GRE /navi 31 sitting around ??)
2) already the 7700xt has been discounted to $350 a few times. That means AMD can sell both navi 44 (=4060 ti for $300) & navi 48 (=7900xt/4070 ti super for $550 to $600), even if navi 32 remains in stock

 

[Kaluan]

I sure hope that it is just

View attachment 100540

being Wccftech.

Because another 4 months of completely pointless delay for a low midrange product which already has its mesa updates + Valve compiler updates out is just ridiculous.

In one of the X870/X870E motherboard previews (forgot which outlet covered it), when talking in a pre-event Q&A about the new mandatory 16x PCIE 5.0 GPU + PCIE 5.0 lanes for NVMe component, they said AMD hinted at "PCIE 5.0 GPUs" coming out "soon" (in the context that more 5.0 for GPU future proofing for is nice and all, but when can we actually get graphics cards that support it).

IMO that can only be interpreted as RDNA4 SKUs, their own new line of GPUs. Now, what exactly "soon" mean, who knows. But as earlier people said, SIGGRAPH or Gamescom seem like a good bet. Or another AMD exclusive livestream event.

 

[jpiniero]

In one of the X870/X870E motherboard previews (forgot which outlet covered it), when talking in a pre-event Q&A about the new mandatory 16x PCIE 5.0 GPU + PCIE 5.0 lanes for NVMe component, they said AMD hinted at "PCIE 5.0 GPUs" coming out "soon" (in the context that more 5.0 for GPU future proofing for is nice and all, but when can we actually get graphics cards that support it).

IMO that can only be interpreted as RDNA4 SKUs, their own new line of GPUs. Now, what exactly "soon" mean, who knows. But as earlier people said, SIGGRAPH or Gamescom seem like a good bet. Or another AMD exclusive livestream event.

Could be Blackwell.

 

[del42sa]

RDNA3.5 gets a 500-600Mhz clock increase.
RDNA4 is an unrelated microarchitecture.

I would like to know where did 600 MHz clock increase vanished ?

 

[reaperrr3]

I would like to know where did 600 MHz clock increase vanished ?

Assuming you're referring to Strix Points' max GPU clock of 2.9 GHz, max clock != average/sustained clock, and Strix IGP has 33% more WGPs on virtually the same process.

There's also that without significantly more cache, it's probably quite bandwidth-constrained for the 16 WGP version, so going higher on clocks would do little else than reduce perf/w and take power budget away from the CPU cores.

Higher clock despite 33% more WGPs suggests that at identical config, RDNA3.5 could indeed hit 500-600 MHz more at the same power budget, there's just no benefit to it in mobile.

It's also possible they want to reserve breaking the 3 GHz barrier for RDNA4 and Strix Halo for marketing reasons.

 

[del42sa]

Higher clock despite 33% more WGPs suggests that at identical config, RDNA3.5 could indeed hit 500-600 MHz more at the same power budget, there's just no benefit to it in mobile.

RDNA3.5 architecture being only used in mobile or NUC

 

[inquiss]

RDNA3.5 architecture being only used in mobile or NUC

Sure, but that doesn't negate the point. If there is a config with fewer WGPs where it's less memory constrained then it may be worth pushing up the clock curve. Why bother if you're memory constrained. Doesn't mean it couldn't do it.

 

[gdansk]

I figured one of the new RDNA4 would launch in late November or December like N31.
It is weird to see so much activity in open source for something further out than that.
But discrete RDNA4 has to be one of AMD's lowest priorities now so who knows.

 

[eek2121]

I figured one of the new RDNA4 would launch in late November or December like N31.
It is weird to see so much activity in open source for something further out than that.
But discrete RDNA4 has to be one of AMD's lowest priorities now so who knows.

AMD’s focus is data center without a doubt. I just hope gaming products continue to be developed. We will see what leaks in the coming days/weeks. I suspect there were roadmap changes. Have not heard/read anything yet, however.

 

[blckgrffn]

AMD’s focus is data center without a doubt. I just hope gaming products continue to be developed. We will see what leaks in the coming days/weeks. I suspect there were roadmap changes. Have not heard/read anything yet, however.

In light of that lack of supporting information in general, I am classifying it as click bait until something more substantial hits to corroborate it.

 

[marees]

In light of that lack of supporting information in general, I am classifying it as click bait until something more substantial hits to corroborate it.

It is probably feedback from board partners in computex

 

[DaaQ]

What people do not seem to understand is silicon. A few months ago forum members believed TSMC 5nm silicon was all the same. It's not and they released newer versions of silicon based on the 5nm platform that have vastly different performance/power differences.

In October 2021, TSMC introduced a new member of its "5 nm" process family: N4P. Compared to N5, the node offered 11% higher performance (6% higher vs N4), 22% higher power efficiency, 6% higher transistor density and lower mask count. TSMC expected first tapeouts by the second half of 2022

There is no clock regression with RDNA3 vs RDNA4 because both processes are on 5nm silicon. If anything the newer 5nm process will clock better and not worse.

Now to the broken RDNA3 architecture where 20-30% of the GPU was non functioning. When RDNA4 was developed, are they basing the performance metrics on what RDNA3 was supposed to be or what the real world performance of the flawed silicon had. Again, not factoring in specific GPU's 7900xtx/7900xt/7800xt but questioning the uplift in gen vs gen performance gains in power and efficiency. Even if you halve the GPU's, you would still have the same ratio of performance based on the GPU metrics at a reduced level based on GPU die size.

The 7800xt uses 252-268 watts of power under load while gaming heavy. If you take 22% off that alone, you would have power consumption of 202.8w if you had a 7800xt on N4P. I am using the middle number between 252 and 268w to get 260w minus 22%. That is still a bit higher than the 4070 but it's in the ballpark for efficiency. Then you add 11% of free performance based on N4P according to TSMC performance metrics of N4P. If AMD had a refresh of 7800xt with N4P, above would be what you should expect. It's for comparison purposes only.

So My point is even with a stripped down RDNA4. Should we not expect the 8800xtx to be close to if not equal or better than the 7900xtx? I am not accounting for any performance uplift between RDNA3 and RDNA or power efficiency that AMD is advertising in their RDNA4 GPU slides.

Finally I will add. I am speculating and not saying that I am right or wrong about RDNA4. I am just trying to figure out the future of AMD GPU's. I will add there is speculation that RDNA4 GPU core clocks will boost up to 3.2ghz. Remember when RDNA3 was announced? They said the GPU would clock up to 3ghz and it didn't. The reason has been speculated that RDNA3 was a broken but functional architecture. That is why I am speculating that a RDNA4 GPU could/should perform much better than RDNA3 if the architecture is fully functional and sound.

My 7900xtx boosts over 3Ghz, what you talking about? I don't even have the 550w BIOS on it.

 

[Hans Gruber]

My 7900xtx boosts over 3Ghz, what you talking about? I don't even have the 550w BIOS on it.

We are going to need some screen shots.

 

[DaaQ]

here ya go, been running I think 3 or 4 days now

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

 

[del42sa]

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

problem solved, the chip being fixed, case closed

 

[DaaQ]

problem solved, the chip being fixed, case closed

Yea I bought it a month ago, Kryosheet and Kritical 1mm thermal pads on water block side.

 

[Hans Gruber]

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

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.

 

[soresu]

It's almost as if RDNA2 was worked on by a different company

AMD has leapfrogging µArch engineering teams, so effectively it was.

 

[soresu]

Because another 4 months of completely pointless delay for a low midrange product which already has its mesa updates + Valve compiler updates out is just ridiculous.

Release roadmap aside the Mesa/Valve movements have nothing to do with anything but the availability of specs/code to work on.

It's a good indicator for sure, but not definitive.

That code will still be worked on for months/years after the actual hardware release anyway.

 

[DisEnchantment]

I figured one of the new RDNA4 would launch in late November or December like N31.
It is weird to see so much activity in open source for something further out than that.
But discrete RDNA4 has to be one of AMD's lowest priorities now so who knows.

AMD literally scrambled to get RDNA4 patches inside LLVM18, submitting dozens of patches on the cutoff week. The dev literally mentioned that they need to get the patches in the LLVM18 release.
Would be odd that they would do that knowing the product is more than a year way.

Same with RadeonSI, Linux 6.10, why hurry to get them in.

 

[Tuna-Fish]

^ Yep. This is very weird to me, AMD is usually quite late for Linux driver support, now they hurried to get it done, and then crickets?

 

[jpiniero]

AMD literally scrambled to get RDNA4 patches inside LLVM18, submitting dozens of patches on the cutoff week. The dev literally mentioned that they need to get the patches in the LLVM18 release.
Would be odd that they would do that knowing the product is more than a year way.

Same with RadeonSI, Linux 6.10, why hurry to get them in.

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

 

[marees]

AMD literally scrambled to get RDNA4 patches inside LLVM18, submitting dozens of patches on the cutoff week. The dev literally mentioned that they need to get the patches in the LLVM18 release.
Would be odd that they would do that knowing the product is more than a year way.

Same with RadeonSI, Linux 6.10, why hurry to get them in.

^ Yep. This is very weird to me, AMD is usually quite late for Linux driver support, now they hurried to get it done, and then crickets?

I still have some hope. Wccftech would have asked around board partners. All of them could have said that they are not aware of launch plans this year.

Doesn't mean AMD wouldn't launch in sep/oct. Maybe Lisa is just running a tight ship 🤔

 

[SolidQ]

Wccftech would have asked around board partners

Unless wccftech talking about N44 next year