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

 

[igor_kavinski]

RTX Ada will arrive relatively soon after RDNA4

Blackwell

 

[maddie]

Everything with N44 & N48 screams cheap, but I guess the deaf can't hear anyhow and keep high pricing as inevitable. We''ll see soon enough.

 

[jpiniero]

Everything with N44 & N48 screams cheap, but I guess the deaf can't hear anyhow and keep high pricing as inevitable. We''ll see soon enough.

What would scream cheap is if they used Samsung.

 

[inquiss]

Everything with N44 & N48 screams cheap, but I guess the deaf can't hear anyhow and keep high pricing as inevitable. We''ll see soon enough.

It screams good margin and in particular good supply. That ram speed is extremely commonplace. They can use anything that would have gone to PlayStation's and ramp up the market share while ramping margins. It's like 480 but at a higher price and not margin cratering.

 

[marees]

What would scream cheap is if they used Samsung.

Samsung doesn't have a proven track record for GPUs right ?
Even Intel goes to TSMC

Ampere had the DLSS / RT moat otherwise the 6600 was handily beating the 3050 in all parameters including price & RT (& even power?)

On a side note, AMD is testing waters by giving future chromebook orders to samsung 4nm

Sonoma Valley (zen 5c — 4x zen 5c + RDNA 3.5?) is rumoured to be on samsung 4nm. Hopefully it is cheaper & faster compared to TSMC 6nm

 

[igor_kavinski]

Hopefully it is cheaper & faster compared to TSMC 6nm

Cheaper for AMD but not for Samsung

They will create a mountain of defective dies to fulfill the contract.

 

[Ghostsonplanets]

Samsung doesn't have a proven track record for GPUs right ?
Even Intel goes to TSMC

They do. They fabbed lots of GPUs for AMD and Nvidia before. They also fabbed the entire RTX 3000 Gaming Client line-up.

Sonoma Valley (zen 5c — 4x zen 5c + RDNA 3.5?) is rumoured to be on samsung 4nm. Hopefully it is cheaper & faster compared to TSMC 6nm

4LPP/SF4 and SF4X are fine. They're competitive against TSMC N5 family and yield well enough. I dunno how SF3 or SF2 will go, but SF4 series are fine enough and cheap.

 

[adroc_thurston]

4LPP/SF4

No but whatever e2400 used is.

SF4X are fine

yea.

 

[Ghostsonplanets]

e2400

e2400 used SF4/4LPP (renamed). e2200 was the one which used 4LPE.

 

[adroc_thurston]

e2400 used SF4/4LPP (renamed)

No, something newer.
Tensor G3 was 4LPP and it was, well, you've seen it.

 

[Ghostsonplanets]

No, something newer.
Tensor G3 was 4LPP and it was, well, you've seen it.

Oh, really? Had no idea then. Thanks for the info.

 

[Aapje]

Why would they?

AMD cancelled their RDNA4 high-end because chiplets are a huge success, and Navi 48 will reportedly be at the ~4070ti(s) level, so it's not like anyone is waiting for it for new performance levels. Even if you want to go with AMD, you can already buy a 7900GRE today for that kind of performance. Radeons are not getting much design wins in laptops, and this probably won't change with RDNA4. RTX Ada will arrive relatively soon after RDNA4, so people will probably wait and see where that will fall as well. Really, there's not much going for RDNA4.

I think that you are muddying the waters by treating cards with substantial performance differences as the same. To start with, Navi 48 is alleged to beat the 7900 XT, which means that it is not at 4070 Ti level, but above 4070 Ti Super.

The 7900 GRE is certainly not at that performance level. So if they price it at $550/600, it will be an improvement. Whether they will price it well or start off too high like they normally do, remains to be seen, of course.

 

[gdansk]

1.
It doesn't matter if Navi 48 and 44 are good. MSRP will still be higher than people want. They want a big cut so Nvidia is forced to lower prices. But Nvidia knows they won't have to: gamers agree that DLSS's quality more than makes up any performance and value deficit.

2.
Does AMD's MI300 sales projection explain why the big Navis were cancelled? That's likely more revenue than the entire Navi product stack could bring in a year.

 

[adroc_thurston]

It doesn't matter if Navi 48 and 44 are good. MSRP will still be higher than people want. They want a big cut so Nvidia is forced to lower prices. But Nvidia knows they won't have to: gamers agree that DLSS's quality more than makes up any performance and value deficit.

ehhh RDNA3 sold pretty well given how bad the actual parts are versus what they were ~supposed~ to be.

Does AMD's MI300 sales projection explain why the big Navis were cancelled? That's likely more revenue than the entire Navi product stack could bring in a year.

This is Qmaga conspiracy looning when the actual answer to any AMD part getting canned is always "it impacted the execution cadence a bit too much".
RDNA4.5 APUs are also dead!

 

[gdansk]

This is Qmaga conspiracy looning when the actual answer to any AMD part getting canned is always "it impacted the execution cadence a bit too much".
RDNA4.5 APUs are also dead!

Execution machine jammed (twice in a row now for RTG)

 

[igor_kavinski]

Execution machine jammed (twice in a row now for RTG)

I wonder if the responsible people enjoyed Lisa's foot jammed up their you know what

 

[adroc_thurston]

Execution machine jammed (twice in a row now for RTG)

No?
4 is on track and lgtm.

 

[igor_kavinski]

4 is on track and lgtm.

Great. Then I guess I will need to own that piece of history, either brand new or when prices settle down, whichever looks better to me.

 

[adroc_thurston]

Great. Then I guess I will need to own that piece of history, either brand new or when prices settle down, whichever looks better to me.

The piece of history is anything RDNA3.5 really.

 

[igor_kavinski]

The piece of history is anything RDNA3.5 really.

Who knows? Maybe a fine wine driver in 2025 will turn all existing RDNA3 cards into RDNA3.5

 

[ToTTenTranz]

I am a bit confused here, as it seems you may be saying contradicting things, at least to me. You say that nothing will replace RDNA 3 at the high end, but then say that an 240mm die will be the high end. Assuming this new die is RDNA 4, either it is faster and it replaces the 7900XTX, or it is slower and it doesn't. It may be newer, but if it isn't faster, then it isn't the new top end card.

My perception is that anything above 6800XT / 7800XT is already a monster on rasterization performance, especially if you use temporal upscaling solutions. At this point, in modern games only raytracing is the bottleneck (which by itself is a bit nonsensical IMO but it is what it is).
So a graphics card that fails to meet the 7900XTX in rasterization but performs much better in raytracing will have a higher perceived value. All thanks to Nvidia's marketing and developer relations machine.

The thing is, I think the moment AMD matches Nvidia in raytracing performance, the latter will pivot its marketing to something else. They actually already started doing this with their AI PCs stuff. Nvidia probably knows RDNA4 already catches up in raytracing performance (per price bracket), so they're changing the tune to "our GPUs run local LLMs".

Blackwell is going to smoke AMD and Intel.

If true, that's usually when Nvidia also gets less careful and overestimates how much people are willing to pay for their GPUs. Just look at Fermi and Turing.

What would scream cheap is if they used Samsung.

I see we're now bringing the Switch 2 into the mix.

 

[SolidQ]

ehhh RDNA3 sold pretty well given how bad the actual parts are versus what they were ~supposed~ to be.

There is some "experts', steam stats only matters

Spoiler

 

[adroc_thurston]

There is some "experts', steam stats only matters

Spoiler

They'll pull out of dedicated GPUs

how many times do people have to say it's not happening

 

[inquiss]

1.
It doesn't matter if Navi 48 and 44 are good. MSRP will still be higher than people want. They want a big cut so Nvidia is forced to lower prices. But Nvidia knows they won't have to: gamers agree that DLSS's quality more than makes up any performance and value deficit.

2.
Does AMD's MI300 sales projection explain why the big Navis were cancelled? That's likely more revenue than the entire Navi product stack could bring in a year.

I don't think it's DLSS that makes any difference. It's that DLSS is done by Nvidia. Whatever feature they have will be praised as _the_ reason why Radeon isn't viable. If it wasn't DLSS it would be one of their other features. Remember when AMD was too power hungry with their cards and Nvidia was cooler and put out less heat. Remember when Nvidia switched the following generation and advertised as a positive how much juice you could get into your card because that's better. It doesn't matter what AMD do until they can beat Nvidia into submission, even then it's uphill for the mind share. Not impossible, but you need a really big stick.

 

[SolidQ]

how many times do people have to say it's not happening

Because they think, if AMD can't compete with NV, they've out(there enought posts about it), also trusting in Intel to compete vs NV

I don't think it's DLSS that makes any difference.

If you read difference gaming forums, there always praise DLSS to heaven. called DLSS "Black Magic"