Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

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

Darkmont · Mar 5, 2025

https://twitter.com/x/status/1897440681779716310

https://twitter.com/x/status/1897446317376594218

RDNA 4 stock doubters crying themselves to sleep

RnR_au · Mar 5, 2025

Sapphire... I'm not happy mate...

Hail The Brain Slug · Mar 5, 2025

RnR_au said:
Sapphire... I'm not happy mate...

Sapphire also
1) Placed a felt over the heatsink fins to protect them from the power cable which appears to block maybe 70% of the flow-through air

2) Designed dual bios into the PCB but left the switch and second bios chip off (Why? Is there a higher tier card above the Nitro+ coming they are trying to upsell?)

3) Forced a 90 degree bend immediately off the connector

4) The techpowerup review didn't mention this, but their teardown shots show the Nitro+ heatsink being NOT evenly mounted against the die. Either it's a design flaw or the heatsink screws were loose on one side, neither of which is good.

Nitro+ - Note the right side has a thick layer of PTM while the left is pressed against the coldplate

Compare to the TUF - Note the PTM melted enough to press out and show an even, firm mount

itsmydamnation · Mar 5, 2025

I need 7900XTX FSR4 ...... i then need optiscalar to update to implement it

inquiss · Mar 5, 2025

moonbogg said:
Many things can be said, but should not be believed. You have faith in the existence of this cube? Why? Because someone "explicitly" told you it exists? The existence of the cube is just like this release being an amazing deal; they are both likely an illusion. The cards are still too expensive and that cube ain't solid boys.

I can't listen to someone who doesn't know what DC means in the context of GPUs

Ranulf · Mar 5, 2025

RnR_au said:
Sapphire... I'm not happy mate...

Right out of the gate slinging arrows:

"Buildzoid here...taking a look Sapphire's implementation of the 12 volt high failure connector"

"They are not doing anything clever with it".

Bryo4321 · Mar 5, 2025

Hail The Brain Slug said:
Sapphire also
1) Placed a felt over the heatsink fins to protect them from the power cable which appears to block maybe 70% of the flow-through air
View attachment 119032

2) Designed dual bios into the PCB but left the switch and second bios chip off (Why? Is there a higher tier card above the Nitro+ coming they are trying to upsell?)
View attachment 119033
View attachment 119034

3) Forced a 90 degree bend immediately off the connector

4) The techpowerup review didn't mention this, but their teardown shots show the Nitro+ heatsink being NOT evenly mounted against the die. Either it's a design flaw or the heatsink screws were loose on one side, neither of which is good.

Nitro+ - Note the right side has a thick layer of PTM while the left is pressed against the coldplate
View attachment 119030

Compare to the TUF - Note the PTM melted enough to press out and show an even, firm mount
View attachment 119031

Yeah, a lot of weird design choices from sapphire for the nitro this go around.

NomanA · Mar 6, 2025

poke01 said:
I looked at reviews. Gotta say if you brought the RTX 40 super series or the RDNA3, there is zero reason to upgrade at for the money it costs to upgrade. I’m on a 4070 Super and the RX 9070XT loses to it in some non-NV titles as well and same for the 7900XT.

Adroc was right, what a boring generation. The uArch is nice but other that meh.

I rate it a 6/10 for RDNA4 and Blackwell is -1/10.

I have an RDNA3 in 7800XT and am quite looking forward to getting 9070XT which finally brings good raster and RT performance at $600 level. In TechPowerUp performance reviews, the Sapphire Nitro version is less than 5% below 4080 and 7900XTX at 4K. Technically, the new architecture is exciting too. Reviews are extremely positive across the board.

poke01 · Mar 6, 2025

NomanA said:
Reviews are extremely positive across the board.

Reviews are very positive because NV this gen is pathetic.

RnR_au · Mar 6, 2025

NomanA said:
Reviews are extremely positive across the board.

That means AMD priced them too cheaply.... /s

coercitiv · Mar 6, 2025

RnR_au said:
Sapphire... I'm not happy mate...

Looking forward to the first AMD cards with 12v-2x6 charred connectors. I was hoping they would isolate 3 pins per rail, alas that was not the case.

DownTheSky · Mar 6, 2025

Anyone has BM:Wukong 1440 RT Medium results. Because in Path Tracing AMD sucks ass it seems.

fastandfurious6 · Mar 6, 2025

just like old times where top-end model was around 600 bucks. they made the decision to only release 1 card on rdna4 many years ago right? such foresight

9700XT is like 800 whole bucks cheaper than RTX5080 and only around -20% slower max...

with ~1500 bucks you build the ultimate game pc: 9800X3D, 9700XT and the rest

absolute bang for buck go AMD

uzzi38 · Mar 6, 2025

inquiss said:
"Nudge*

C&C got their sample running at 3.9GHz in vkpeak FP64, and ~3.8GHz in vkpeak FP16.

Timorous · Mar 6, 2025

uzzi38 said:
C&C got their sample running at 3.9GHz in vkpeak FP64, and ~3.8GHz in vkpeak FP16.

4Ghz Edition refresh here we come!

inquiss · Mar 6, 2025

DownTheSky said:
Anyone has BM:Wukong 1440 RT Medium results. Because in Path Tracing AMD sucks ass it seems.

Who cares about path tracing? No one should

inquiss · Mar 6, 2025

uzzi38 said:
C&C got their sample running at 3.9GHz in vkpeak FP64, and ~3.8GHz in vkpeak FP16.

So it can clock like a monster. What limits that in games. What's pushing up power in games that means it can't hit these clocks? That it runs out of memory bandwidth so can't use these higher clocks on games? What's the takeaway?

leoneazzurro · Mar 6, 2025

inquiss said:
Who cares about path tracing? No one should

Also, what is around in this moment is "Path tracing Nvidia edition". Zero optimization for AMD's architecture.

uzzi38 · Mar 6, 2025

inquiss said:
So it can clock like a monster. What limits that in games. What's pushing up power in games that means it can't hit these clocks? That it runs out of memory bandwidth so can't use these higher clocks on games? What's the takeaway?

In vkpeak you don't need your ROPs and other gfx related fixed function hardware doing anything.

In games, you do, and those all need significant amounts of power too. So the power budget becomes the limiting factor preventing such high clocks

exquisitechar · Mar 6, 2025

uzzi38 said:
In vkpeak you don't need your ROPs and other gfx related fixed function hardware doing anything.

In games, you do, and those all need significant amounts of power too. So the power budget becomes the limiting factor preventing such high clocks

RDNA3 had similar behavior. It's still something to see a GPU hit nearly 4GHz in anything, though.

Meteor Late · Mar 6, 2025

I think the main reason AMD didn't try with the top die is because of the RT deficit against Nvidia. Think about it, what's the point in "raster parity" or even slightly winning, if then there are like 3 examples (that I have seen: Alan Wake, Black Myth Wukong, Indiana Jones) where in RT AMD is massively behind, the difference is just too big. At those price points (>1000$), almost no one would buy AMD, when you spend so much money for a GPU, you want the best experience without compromises.
Ultimately, knowing what we know now, AMD did the right choice. I know that's maybe not the only reason AMD canned the top die, they maybe thought Blackwell would be better than it is, but still.
You can be moaning about how path tracing is irrelevant, but many people care about it, especially when you spend so much on a GPU, you want to try the latest technologies.

maddie · Mar 6, 2025

Meteor Late said:
You can be moaning about how path tracing is irrelevant, but many people care about it, especially when you spend so much on a GPU, you want to try the latest technologies.

Ah yes, the GPU purchase as a technology demonstrator. Completely forgot about that market segment.

Meteor Late · Mar 6, 2025

maddie said:
Ah yes, the GPU purchase as a technology demonstrator. Completely forgot about that market segment.

It also looks much better, this is a reality. One can argue that the enormous performance impact makes it not worth it, but that it makes a visual difference is evident.

Win2012R2 · Mar 6, 2025

uzzi38 said:
So the power budget becomes the limiting factor preventing such high clocks

4 ghz at 500w then edition?

maddie · Mar 6, 2025

Meteor Late said:
It also looks much better, this is a reality. One can argue that the enormous performance impact makes it not worth it, but that it makes a visual difference is evident.

Try a Hi-def photograph or movie.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Junior Member

Platinum Member

Diamond Member

Diamond Member

Senior member

Platinum Member

Member

Member

Diamond Member

Platinum Member

Diamond Member

Senior member

Senior member

Platinum Member

Golden Member

Senior member

Senior member

Golden Member

Platinum Member

Senior member

Senior member

Diamond Member

Senior member

Senior member

Diamond Member