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

 

[darkswordsman17]

I wanna say Kabini?

Not ideal power consumption wise for sure, but then neither is TX1 at full power.

Plus Kabini would have allowed them a significant level of code compatibility with PS4 and XB1.

IMHO Ninty simply went with TX1 because nVidia offered them a dirt cheap price

Definitely was not the case, as reports were that Nintendo was paying pretty close for the Tegra chip to what Sony was the PS4 APU.

It simply came down to Nvidia had a chip for Nintendo to test and AMD didn't. Nintendo tested multiple ARM chips, and they even had test units for the 3DS that used Tegra chips.

Kabini was too old and would have had terrible performance in the power envelope they needed.

Pretty sure it was physically too large especially when coupled with a heatsink. Nintendo was never going to use x86 based chips in a handheld, I don't even know why people are acting like it was at all a possibility. And that was in like the early 2010s, where x86 wasn't even attempting to actually compete with ARM.

 

[soresu]

Nintendo was never going to use x86 based chips in a handheld

The irony that there are probably more x86 game oriented handhelds these days than ARM ones since Steam Deck made it hip.

Worse that you can literally emulate the Switch on some of these handhelds 🤣

 

[igor_kavinski]

This is a testament for unified memory architectures.

Finally people can see the light .

Which would be a step towards unifying the CPU and GPU into a CGPU!

 

[soresu]

Which would be a step towards unifying the CPU and GPU into a CGPU!

SoC's are already as integrated as they will ever be.

 

[adroc_thurston]

This is a testament for unified memory architectures.

?
This isn't 300A being tested.

 

[Ghostsonplanets]

I do want to know what soured Nintendo away from AMD. Lisa is usually very accommodating to partnerships and existing partners.

Nintendo initial idea for Switch was to take the Wii U AMD Latte GPU (A Terascale 2 design based on Radeon HD 6450), cut it down in half (AMD Decaf GPU) and use it for a Portable first device, alongside 4x Cortex A53 for a release circa around 2014 - 2015. If you recall, Nintendo then CEO and President, Mr. Satoru Iwata, first words about the Wii U and 3DS successor was that it would adequately absorb the hardware characteristics of both.

That made sense because Nintendo would retain their 15 years R&D with ArtX/ATI/AMD GPUs and keep their mature and deeply optimized for AMD hardware toolchain.

What happened was that Nvidia had a much better solution than the SoC Nintendo was developing from both performance and performance per watt perspective, while also being much more modern silicon and middleware support wise.

Nvidia had to move heaven and earth to convince Nintendo to move on from their mature and built upon years of AMD GPU knowledge toolchain towards the new and radically different Nvidia one. For that to happen, they had a very attractive offer where Nvidia made toolchains and APIs, direct support from Nvidia to Nintendo teams and also the Tegra X1 SoC, which was eons beyond anything else in the market for a <10W part back in 2014.

Or an employee had a Tegra device which they rooted and somehow ported Mario over to it and then showed it to their bosses and they went "Mamma Mia!".

You're not too far off that Nvidia demoed Tegra X1 capabilities to Nintendo and is one of the reasons they won over the contract.

And, to be honest, Nintendo tacitly admitted they wanted to work with Nvidia for the longest time. The Nintendo 3DS (Known as DS successor ar the time) originally used to have a Tegra/Tegra precursor before the contract fell through and Nintendo had to choose DMP PICA solution for the GPU IP.

 

[soresu]

And, to be honest, Nintendo tacitly admitted they wanted to work with Nvidia for the longest time

How does that saying go.... be careful what you wish for?

Certainly both Sony and MS thought as much after PS3 and Xbox OG.

 

[Ghostsonplanets]

How does that saying go.... be careful what you wish for?

Certainly both Sony and MS thought as much after PS3 and Xbox OG.

Given Switch 2 is using a custom SoC from Nvidia once again, I'm sure their relationship is more than fine. It's been more than 10 years since the contract started.

And Nintendo, unlike MS or Sony, don't go penny pinching for price reduction nor made much fuss over the fuseé-geele TX1 security flaw. So it seems they're the ideal partner to Nvidia🤣

 

[adroc_thurston]

nor made much fuss over the fuseé-geele TX1 security flaw

They very much did.

 

[Ghostsonplanets]

They very much did.

Not really.

NTD was already aware of TX1 security flaws before the system launch and even reported them to Nvidia. Hence why the 20nm fixed revision was rolled out so quickly.

Mariko (16nm fully patched TX1) was already in development by 2016.

None of the issues made Nintendo delay the release or wait for the fixed version.

 

[adroc_thurston]

Not really.

NTD was already aware of TX1 security flaws before the system launch and even reported them to Nvidia. Hence why the 20nm fixed revision was rolled out so quickly.

Mariko (16nm fully patched TX1) was already in development by 2016.

None of the issues made Nintendo delay the release or wait for the fixed version.

the point is that they complained. hard.

 

[Ghostsonplanets]

the point is that they complained. hard.

Yes. But still went ahead with the flawed SoC nonetheless.

I'm pretty sure had it been Sony or MS, such issue would have much greater consequences.

 

[soresu]

None of the issues made Nintendo delay the release or wait for the fixed version.

Given Ninty's issues with emulation and ROMs seem to far exceed any other console manufacturer in modern times that seems a bit odd.

You are saying they knew about and therefore basically invited the inevitability of a hack.

That seems to me the most boneheaded thing they could have possibly done.

 

[jpiniero]

Given Switch 2 is using a custom SoC from Nvidia once again, I'm sure their relationship is more than fine. It's been more than 10 years since the contract started.

There were rumors that Nintendo talked to Qualcomm and even Samsung about doing Switch 2's SoC. But nothing came of it.

And Nintendo, unlike MS or Sony, don't go penny pinching for price reduction

Oh but they penny pinch allright. I don't know if AMD could do an APU cheap enough for them while keeping power low.

 

[Ghostsonplanets]

Given Ninty's issues with emulation and ROMs seem to far exceed any other console manufacturer in modern times that seems a bit odd.

You are saying they knew about and therefore basically invited the inevitability of a hack.

That seems to me the most boneheaded thing they could have possibly done.

Yes. They needed a console on the market to replace the Wii U failure quickly enough and that overcome the security issues. But they knew about it and reported to Nvidia themselves before the issues were even public.

You can read about it in the emails and internal tickets files between Nintendo and Nvidia that were in the Gigaleak data. Not sure if that's still around the web nowadays though.

There were rumors that Nintendo talked to Qualcomm and even Samsung about doing Switch 2's SoC. But nothing came of it.

QCOM I know had said Nintendo and Sony had talked with them. But I'm pretty sure the way QCOM worded lacked context as wouldn't make sense for Nintendo to go from Nvidia GPU IP to QCOM IP. Would be an odd regression.

IMO there's only one GPU IP vendor that Nintendo could go instead of Nvidia and that's AMD.

I would have liked to see the AMD Decaf GPU into the Switch instead of the Maxwell TX1. But it was too outdated (TeraScale 2) and small (80 Shaders).

Oh but they penny pinch allright. I don't know if AMD could do an APU cheap enough for them while keeping power low.

They do as in they don't do extravagances like Sony or MS 300+mm² chips. But they still pay quite the money for their chips. The Wii U chips, for example, weren't cheap. It was a IBM PowerPC 3-core fabbed on IBM SOI 45nm + a custom Nintendo AMD GPU fabbed on Renesas 40nm + 32MB eDram. All within a MCM package done by Renesas. It lacked performance but it was certainly not an outdated thing from process and costs point of view.

 

[soresu]

Oh but they penny pinch allright. I don't know if AMD could do an APU cheap enough for them while keeping power low.

Aint that the truth.

Wii U basically uses an overclocked 3 core version of the Gamecube CPU, with a 4-5 year old GPU µArch.

If that aint penny pinching I don't know what is.

 

[reaperrr3]

Aint that the truth.

Wii U basically uses an overclocked 3 core version of the Gamecube CPU, with a 4-5 year old GPU µArch.

If that aint penny pinching I don't know what is.

To be fair, at least part of the reason was to maintain backwards compatibility with Wii games without the need for emulation, they were probably hoping a lot of Wii buyers would go and replace their Wii with a Wii U.

And they didn't only add 2 cores and higher clocks, the core L2 also went from 256KB to 2MB for the 'prime' core and 512KB for each of the 2 secondary cores, so it wasn't only more cores + higher clocks.

But yes, they penny pinched a tad too much on the CPU.
It would've needed at least 4 cores and notably higher clocks (at least for the prime core) for a lot of 3rd-party games of that time to be portable to the Wii U.
At least one reason that sank the U was that many 3rd-party games that publishers would've liked to release on the U simply wouldn't run well enough on that CPU.