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

 

[ToTTenTranz]

No fresh leaks as in he asked his sources / chatgpt & didn't get any further 'leaks'

I don't get the hate. He's been the first to leak a ton of AMD hardware, years before it's out. Especially when it comes to AMD GPUs and SoCs.

He was the first to leak AT0 (which is now regarded as pretty much certain) just days after our resident leakers were saying there was no high end Radeon in RDNA5. He was the first to leak the hardware specs for the PS6 handheld and PS6 home console. He was also the first to leak Magnus and the specs for Medusa Halo / Premium.


His sources for Intel GPUs suck or could be just trolls, but in what relates to AMD GPUs he's probably the safest bet out there.

 

[marees]

I don't get the hate. He's been the first to leak a ton of AMD hardware, years before it's out. Especially when it comes to AMD GPUs and SoCs.

He was the first to leak AT0 (which is now regarded as pretty much certain) just days after our resident leakers were saying there was no high end Radeon in RDNA5. He was the first to leak the hardware specs for the PS6 handheld and PS6 home console. He was also the first to leak Magnus and the specs for Medusa Halo / Premium.


His sources for Intel GPUs suck or could be just trolls, but in what relates to AMD GPUs he's probably the safest bet out there.

He leaked Magnus
But in same video he said he discussed it with chat gpt & concluded that it was for PlayStation (because PlayStation gets more clicks than xbox)

So watching his videos becomes a constant struggle to sift thru the sands to mine the grains of real info

Not to mention the 9070xt & 9060xt came out one half year after he said AMD was ready to release them

 

[marees]

8 CU RDNA 4 ??

https://twitter.com/x/status/1984778416089153761

 

[marees]

Specs

https://twitter.com/x/status/1984770798515863692

 

[SolidQ]

Specs

128TMU/32ROP for 8CU?

 

[marees]

Specs

https://twitter.com/x/status/1984770798515863692

Likely based on RDNA 4. Not confirmed yet

https://twitter.com/x/status/1948993107640344745

 

[maddie]

I don't get the hate. He's been the first to leak a ton of AMD hardware, years before it's out. Especially when it comes to AMD GPUs and SoCs.

He was the first to leak AT0 (which is now regarded as pretty much certain) just days after our resident leakers were saying there was no high end Radeon in RDNA5. He was the first to leak the hardware specs for the PS6 handheld and PS6 home console. He was also the first to leak Magnus and the specs for Medusa Halo / Premium.


His sources for Intel GPUs suck or could be just trolls, but in what relates to AMD GPUs he's probably the safest bet out there.

One has to feed one's animus I guess.

He does appear to field some high profile guest speakers and a wise person with doubts, would wonder "how is this possible?" instead of saying as some have here, " I can't understand why they would be there".

In the real world, theory follows data, not the reverse.

 

[Kepler_L2]

I don't get the hate. He's been the first to leak a ton of AMD hardware, years before it's out. Especially when it comes to AMD GPUs and SoCs.

He was the first to leak AT0 (which is now regarded as pretty much certain) just days after our resident leakers were saying there was no high end Radeon in RDNA5. He was the first to leak the hardware specs for the PS6 handheld and PS6 home console. He was also the first to leak Magnus and the specs for Medusa Halo / Premium.


His sources for Intel GPUs suck or could be just trolls, but in what relates to AMD GPUs he's probably the safest bet out there.

The "9090 XT" is something he completely pulled out of his ass though. He claimed AMD had already done a die shrink of N48 to N3P, added GDDR7 support and were running it at 3.7 GHz in labs while beating an RTX 5080.

Do you actually think that Lisa Su's TO-minmaxing AMD would actually waste money on such a prototype, and then never release it?

 

[ToTTenTranz]

The "9090 XT" is something he completely pulled out of his ass though. He claimed AMD had already done a die shrink of N48 to N3P, added GDDR7 support and were running it at 3.7 GHz in labs while beating an RTX 5080.

Do you actually think that Lisa Su's TO-minmaxing AMD would actually waste money on such a prototype, and then never release it?

IIRC he said it could be a shrink but wasn't told that much, the shrink was just his speculation. What he was told is of a refresh using faster memory and higher core clocks.
I don't think a N3P version of N48 exists. I have no reason to believe he's lying about a higher clocked version of N48 on a higher TDP and faster GDDR6 / GDDR7.

And it's not me hoping for this card to exist, either. I have a 5090. But it's pretty logical to expect AMD is planning on showing something when Nvidia releases the 50 Super line.

 

[coercitiv]

But it's pretty logical to expect AMD is planning on showing something when Nvidia releases the 50 Super line.

Logic would have given us big die RDNA4. That is why we differentiate between informed speculation and leaks, because leaks are supposed to be the real deal whether they're logical or not.

 

[ToTTenTranz]

Logic would have given us big die RDNA4.

Are you sure?

AMD was coming from RDNA3 that didn't go well at all, or at least far from the performance targets they had planned so trust in the graphics division was low, and then they heard Nvidia was coming up with a 512bit reticle size limit behemoth (that was also planned to clock much higher than it did) and they had dibs for GDDR7.


AMD was probably afraid they'd have their top end N4C trading blows with a RTX 5080 (if e.g. N4C couldn't its performance targets and GB203 could reach 3.4GHz or so), like the 7950XTX ended up trading blows with the RTX 4080 despite being much more expensive to produce.

 

[coercitiv]

Are you sure?

Yes. AMD knew even their conservative estimates were great, and they were set to release RDNA4 in early 2024. MLID himself chose to not divulge further information and hurt his audience numbers in order to protect RADEON's competitiveness. Imagine people telling MLID confidential product information and then asking him to refrain from reporting it in order to protect the company they were supposed to protect in the first place.



Jokes aside, IMHO the minor setback of RDNA3 did not justify the reduced RDNA4 lineup. If anything, RDNA3 would have been the more logical generation to limit risks considering the major change in architecture and packaging. Moreover, Nvidia was moving from Samsung to TSMC, this alone should have made AMD chicken out.

 

[adroc_thurston]

AMD was probably afraid they'd have their top end N4C trading blows with a RTX 5080

oh lmao no, it was a different class of a device altogether.