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

 

[Thunder 57]

well, duh.
either AMD embraces the Ork warboss lifestyle and builds the biggest choppa' again and again, or it is stuck forever.

worked every bit as well as one could imagine in DYI DT CPUs so they just have to commit.

In what world are Zen cores bigger than Intel P cores?

 

[branch_suggestion]

In what world are Zen cores bigger than Intel P cores?

In what world can Intel stack cache?

 

[adroc_thurston]

In what world are Zen cores bigger than Intel P cores?

They aren't.
Fortunately, AMD ships 3 times the LLC.

 

[Thunder 57]

In what world can Intel stack cache?

You don't count LLC when measuring core size. Not even L2.

 

[Thunder 57]

They aren't.
Fortunately, AMD ships 3 times the LLC.

AMD is quite competitive without X3D with a smaller core.

 

[adroc_thurston]

AMD is quite competitive without X3D with a smaller core.

'competitive' is for suckers.
You gotta win.

 

[inquiss]

AMD is quite competitive without X3D with a smaller core.

You don't count LLC when measuring core size. Not even L2.

They have the biggest stick. Biggest stick doesn't mean it is the biggest in die size, consumers don't care about die size. They care about what's fast and what the price is. And if it's the fastest price is sorta what you get to set.

 

[SolidQ]

Best Selling GPU from Newegg
Poeple love Asrock?

 

[ToTTenTranz]

Best Selling GPU from Newegg

That the 3060 selling for $300, 5 years after release, is mind-boggling.
Using old and slow GDDR6 and a chip made on Samsung 8nm, Nvidia's (and OEMs') margins on this must be sky high.

Poeple love Asrock?

It's probably just the cheapest 9060 XT 16GB around.

 

[Vikv1918]

'competitive' is for suckers.
You gotta win.

Here's the problem, nvidia will then just release some new proprietary feature, like proprietary neural shaders or something and it will run like crap on AMD, and this new feature would suddenly become the most important thing ever and the narrative would be "sure 10090 XT beat 6090 in outdated games but in the new hotness 6090 is 20% faster" and we get back to square one.

 

[SolidQ]

nd this new feature would suddenly become the most important thing ever

and supported in few games only. Like Turing was long time without RT games

 

[ToTTenTranz]

Here's the problem, nvidia will then just release some new proprietary feature, like proprietary neural shaders or something and it will run like crap on AMD, and this new feature would suddenly become the most important thing ever and the narrative would be "sure 10090 XT beat 6090 in outdated games but in the new hotness 6090 is 20% faster" and we get back to square one.

It's exactly this.
If Nvidia doesn't win at the current batch of benchmarks, they'll just push all reviewers to focus exhaustively on this new, suddenly-super-important feature that their newest architecture excels on.

They have enough leverage over reviewers and developers to not let a GeForce FX situation happen ever again.

 

[Thunder 57]

That the 3060 selling for $300, 5 years after release, is mind-boggling.
Using old and slow GDDR6 and a chip made on Samsung 8nm, Nvidia's (and OEMs') margins on this must be sky high.



It's probably just the cheapest 9060 XT 16GB around.

And chumps say that "AMD needs to compete on price". Well you can get a 6600 for cheaper than a 3060. Hell a 600 and 3050 used to be priced the same. Guess which sold by the boatload?

 

[adroc_thurston]

Here's the problem, nvidia will then just release some new proprietary feature, like proprietary neural shaders or something and it will run like crap on AMD, and this new feature would suddenly become the most important thing ever and the narrative would be "sure 10090 XT beat 6090 in outdated games but in the new hotness 6090 is 20% faster" and we get back to square one.

None of that matters when AMD can ship 2x the shader core count of NV.

If Nvidia doesn't win at the current batch of benchmarks, they'll just push all reviewers to focus exhaustively on this new, suddenly-super-important feature that their newest architecture excels on.

This only works if AMD plays fair.
They don't have to!

 

[tsamolotoff]

gfx13 codenames are from Transformers

Alpha Trion, (Ultra) Magnus, Orion (Pax)

Where is the mighty Predaking, wink wink

But everybody wants a winner.

Pretty sure the media would be like "AMD did another space heater like Hawaii" (meme videos ensue)

 

[adroc_thurston]

Pretty sure the media would be like "AMD did another space heater like Hawaii" (meme videos ensue)

Would not matter one bit if it wins by chungus double digits.
All they need is the will.

 

[branch_suggestion]

MLID claims that the 2 month old internal AMD market share estimate for 'dedicated gaming' (DIY desktop) is 22% for RDNA4, up from 15% with RDNA3, not exactly sure if it is rev or unit share.

 

[SolidQ]

Nice diffrence Samsung vs Hynix
Need HUB retest it

https://videocardz.com/newz/xfx-says-its-radeon-rx-9060-xt-gpus-with-samsung-gddr6-memory-run-10c-cooler-than-hynix-based