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

 

[adroc_thurston]

Do amd give up permanently with the multi-SED approach?

No. But you'll have to wait until the bubble pops and hatchets overcrowded GPGPU roadmaps into nothingness.

does all this fuzz mean there will be no rdna5 mcm gpu?

Yes. You get more CDNA instead.

 

[Mahboi]

Just 1 month ago: View attachment 108367

 

[PJVol]

No. But you'll have to wait until the bubble pops and hatchets overcrowded GPGPU roadmaps into nothingness.

Uh, this can be long way to go?

 

[ToTTenTranz]

Uh, this can be long way to go?

Yes, it's probably not happening until all these new fabs are up and running, which could take over 5 years.

Assuming China doesn't try to invade Taiwan in that timeframe.

 

[poke01]

Just 1 month ago: View attachment 108367

this dude said AMD laptops had more supply than macbooks. dude is a grifter. AMD will launch RDNA 4 when its ready. He knows nothing about volume on a global scale.
by dude i mean MLID

 

[SolidQ]

by dude i mean MLID

People sometimes confuse. he's just quote from other people. Same was with ps5pro and other

 

[marees]

Is this referring to RT capabilities ?

Sony Breaks Down the PS5 Pro's Tech and Reveals That It Was in the Works Before the PS5 Even Released

We talk to PlayStation designer Toshi Aoki about the PS5 Pro's tech, making the disc drive separate, and more.

in.ign.com

 

[marees]

Is this referring to RT capabilities ?

Sony Breaks Down the PS5 Pro's Tech and Reveals That It Was in the Works Before the PS5 Even Released

We talk to PlayStation designer Toshi Aoki about the PS5 Pro's tech, making the disc drive separate, and more.

in.ign.com

I am referring to this quote

PlayStation Senior Principal Product Manager Toshi Aoki says that

the PS5 Pro contains "dedicated custom hardware" co-developed with AMD that is in the GPU block,

and that games will be patched to use that part of the block.

Sony Breaks Down the PS5 Pro's Tech and Reveals That It Was in the Works Before the PS5 Even Released

We talk to PlayStation designer Toshi Aoki about the PS5 Pro's tech, making the disc drive separate, and more.

in.ign.com

 

[TESKATLIPOKA]

Now that it is clarified that launch will happen only in next year (after clearing navi 31 stocks), AMD could launch the cards for a reasonable price (rather than $600 if launched today)

N48xtx 16gb 64 CU — $550
N48xt 16gb 56 CU — $450
N48xl 12gb 48 CU — $350

N44xt 8gb 32 CU — $280
N44xl 8gb 28 CU — $240

N48XL with that price would make N44 unsellable.

 

[jpiniero]

I am referring to this quote

The Pro has a 300 TOPs NPU block, similar to what Strix has except bigger. I would have to think RDNA 4 dGPUs don't have this.

 

[marees]

N48XL with that price would make N44 unsellable.

True. I should add a sequence of launch to clarify this:

Q1 2025:
N48xtx 16gb 64 CU — $550
N48xt 16gb 56 CU — $480 (launch price)

Q2 2025
N44xt 8gb 32 CU — $280
N44xl 8gb 28 CU — $240

Q3 2025
N48xl 12gb 48 CU — $380 (launch price. If it is launched)

 

[Mopetar]

I don't think the prices will be that low. $600 is probably the lowest for the 64 CU part. They'll probably try to sell Navi 44 at $300 as well.

 

[SolidQ]

sell Navi 44 at $300 as well.

for 300$ with 8GB N44 is DOA. Reviewers gonna kill it

 

[adroc_thurston]

Reviewers gonna kill it

They're gonna kill it anyway.

 

[SolidQ]

They're gonna kill it anyway.

if 200-230 it's ok, but if price higher then yes, they're gonna kill it

 

[SolidQ]

Is this referring to RT capabilities ?

PS5

PS5pro

Also

Massive difference
So seems RDNA4 desktop will have good RT

 

[ToTTenTranz]

They're gonna kill it anyway.

If there's no nvidia-like cult following and reviewers aren't going to recommend it, then they're planning for failure already.

Releasing an 8GB part in 2025 had better be ultra cheap, or get full use of 16x PCIe 4/5 for decent memory allocation (though knowing AMD they're about to cheap out on the PCIe lanes again).

 

[Kepler_L2]

The Pro has a 300 TOPs NPU block, similar to what Strix has except bigger. I would have to think RDNA 4 dGPUs don't have this.

It's not an NPU why do people keep repeating this?

 

[Kepler_L2]

If there's no nvidia-like cult following and reviewers aren't going to recommend it, then they're planning for failure already.

Releasing an 8GB part in 2025 had better be ultra cheap, or get full use of 16x PCIe 4/5 for decent memory allocation (though knowing AMD they're about to cheap out on the PCIe lanes again).

N48 is Gen5 x16, N44 idk but probably Gen5 x8.

 

[jpiniero]

It's not an NPU why do people keep repeating this?

What else would it be? Same thing as Strix has, only bigger.

I still haven't seen any confirmation on the die size or the power consumption.

 

[RnR_au]

What else would it be? Same thing as Strix has, only bigger.

I still haven't seen any confirmation on the die size or the power consumption.

A Xilix FPGA? Purely dedicated to Sony's scaler/frame-gen? Just brain-farting here...

 

[maddie]

What else would it be? Same thing as Strix has, only bigger.

I still haven't seen any confirmation on the die size or the power consumption.

A dedicated 300 TOPS NPU would be around 90mm^2.

 

[poke01]

Whats AMDs equivalent of a tensor core in RDNA4?

 

[marees]

Whats AMDs equivalent of a tensor core in RDNA4?

I am thinking that comes only with UDNA or RDNA 5

 

[Kepler_L2]

What else would it be? Same thing as Strix has, only bigger.

I still haven't seen any confirmation on the die size or the power consumption.

It's Int8 WMMA running on the GPU. A 300 TOPs NPU would be almost 200mm² by itself.