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

 

[soresu]

Mind you, the underlying amdgcn ISA has been pretty stale bread ever since Tonga.

Does rapid packed math in GFX9 not count as significant development?

Or is that purely µArchitectural?

 

[adroc_thurston]

Does rapid packed math in GFX9 not count as significant development?

Well it added packed math instructions but that doesn't change the fundamental ISA.
It's very CPU-like in that way

 

[basix]

RDNA4 improved 1.35x is just not fact, you can make RTX 4090 work at 250W and say its 1.5x improvement also, same as 9070 at ~220w but then you have 9070 xt at 300+ w

Not really.
RDNA3 just had faaaaar higher clock targets and never hit them.

Did you read my disclaimer? And why introducing an undervolted 4090 or RDNA4 power targets for bandwidth efficiency topics?

Disclaimer:
- These numbers do not indicate, that these GPUs work in their bandwidth limit

Without specific tests you cannot derive absolutely correct bandwidth efficiency numbers. The real bandwidth efficiency increase might be lower than 1.35x, maybe 1.25x or whatever.
If you look at 7900GRE or 7700XT, you definitely see some bandwidth limitations. So I assume the numbers are not too far off from 1.35x. RDNA4 did improve bandwidth efficiency quite significantly.

But anyways, the point of the whole argument was not to prove any RDNA4 enhancements.
GDDR6 is good enough for RDNA4, indeed. But RDNA4 is not superior regarding bandwidth efficiency compared to Nvidia Lovelace or Blackwell.

 

[adroc_thurston]

And why introducing an undervolted 4090 or RDNA4 power targets for bandwidth efficiency topics?

I don't think anyone talks undervolt or anything here.

 

[basix]

I meant what reb0rn was saying. But if you have a better interpretation of his words, please enlighten me

 

[Thunder 57]

I don't think anyone talks undervolt or anything here.

Since at least Polaris undervolting has helped a good bit. My motto is "Undervolting is the new overclocking". It applies to CPU's as well.

 

[adroc_thurston]

Since at least Polaris undervolting has helped a good bit. My motto is "Undervolting is the new overclocking". It applies to CPU's as well.

It's fine but a lot undervolts ride on the edge of stability, and when the Si ages, they break. Fun!

 

[SolidQ]

Might be not release

That like 9070XT GDRR7 version with 15-45% perf

 

[marees]

Might be not release

That like 9070XT GDRR7 version with 15-45% perf

Gddr7 — $1000 ?
3gb — $1200 ?
3.7ghz — mega cooling requirements ?
N4x Arizona — $1500 ?
N3x — $1800 ??

 

[adroc_thurston]

Might be not release

That like 9070XT GDRR7 version with 15-45% perf

This is all ****

 

[soresu]

This is all ****

It seldom isn't.

 

[Mopetar]

It seldom isn't.

MLID is like a broken clock, only it's one that's still running only it cycles a bit more slowly than the actual day. Eventually he'll be correct, but I'll let you work out the math on how often that occurs.

Treat him like a horoscope - a good bit of fun that you shouldn't take too seriously - and you won't have any problems. Woe be any man who uses his videos as the basis for financial decisions though.

 

[eek2121]

MLID is like a broken clock, only it's one that's still running only it cycles a bit more slowly

Oh, his cycles run more slowly alright. 😉