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

HuesToo4 · Mar 22, 2025

SolidQ said:
So people claiming to me. only NVIDIA drives the industry forward, the rest companies just copy!

Nvidia seems to always be ahead of everyone else with introduction of gaming features like vrr, rtrt, ai upscaling, ai frame interpolation; regardless whether people care about these or how useful they are. There'll always be people shilling for Nvidia because of this, or until AMD tries to compete at the top.

moinmoin · Mar 22, 2025

SolidQ said:
So people claiming to me. only NVIDIA drives the industry forward, the rest companies just copy!

Well, this gets us back to the topic of Nvidia's whole FOMO spiel: The technical "advancement" Nvidia proposes are mostly of the nature of "close to unfeasible". Nvidia then sells its highest end cards which makes the tech "playable" like Nintendo 64 slideshows of lore. Nvidia then builds a line of fake solution tools to improve the bad performance, themselves relying on the latest tech.

This long hasn't been about improving tech anymore. So since Nvidia is perfectly fine with the current state and nobody else is interested in fighting the behemoth things simply stay as they are.

igor_kavinski · Mar 22, 2025

SolidQ said:
So people claiming to me. only NVIDIA drives the industry forward, the rest companies just copy!

I think Nvidia should simply quit the gaming market. I'm serious. We don't need them. Nothing they have done in the past 5 years to improve games interests me that much.

They actually have not improved gaming at all. They have only made it more expensive.

Abwx · Mar 22, 2025

Viral marketing and FUD in daylight :

Other leaks suggest that the RX 9060 XT will utilize an 8-pin power connector and consume at least 500 watts.

Asus Radeon RX 9060 XT RDNA 4 GPUs appear with 8GB and 16GB options as release date approaches

Asus is the latest GPU maker to reveal its selection of Radeon RX 9060 XT RDNA 4 cards. Gamers can choose from 8GB and 16GB versions, as well as six different models. The new report doesn't mention other specs, but buyers can expect a second-quarter RX 9060 XT release date.

www.notebookcheck.net

igor_kavinski · Mar 22, 2025

500W on a single 8-pin connector?

Had no choice but to look at the mugshot of the author of that bit.

Yup. He definitely LOOKS clueless

del42sa · Mar 22, 2025

adroc_thurston said:
Doesn't mean they're 3.5 either.

welcome RDNA 3.75

adroc_thurston · Mar 23, 2025

del42sa said:
welcome RDNA 3.75

Or 4.25.

itsmydamnation · Mar 24, 2025

we totally need a RDNA [3-4].unreal-number

scineram · Mar 24, 2025

adroc_thurston said:
a) little real world demand
b) just One More Codepath for devs to maintain is annoying

Windows 11 abandoned a bunch of old hardware, and some not so old. That should have been a clear hint to set up a new baseline on the graphics API front as well.

soresu · Mar 24, 2025

scineram said:
That should have been a clear hint to set up a new baseline on the graphics API front as well.

DX12U is kinda that I guess.

It's a shame that the Vulkan standardisation folks at Khronos haven't made moves on a new hw baseline since the initial basic one in 2016 - it's no wonder that Epic are dragging their feet so long on concrete support for the Vulkan back end when vital features aren't even in the mainline spec yet.

When various mobile GPU architectures now have RT HW it really does start to look ridiculous that Vulkan won't get their ducks in a row and make it standard already, let alone stuff like mesh shaders.

soresu · Mar 24, 2025

Gideon said:
I really wish they'd create a DX Next and a Vulkan 2.0. GPUs have come a looooong way since DX 12 / Mantle days and the APIs are still modelled based on 10 year old GPUs. Don't take my word for it, take sebbbi's

(Sebastian Aaltonen is a graphics developer with over 20 years of experience including being a principal engineer for Ubisoft and Unity):

https://twitter.com/x/status/1842987149182005387

I really hope the just do a new ground up API that only supports modern GPUs. Is a lot thinner and includes GPU-driven draw calls from the start. And among other things would allow to trace rays from anywhere (no arbitary stupid shader stages) and no pipelines (solving a respectable amount of "shader compilation stutter" issues)

This is his GitHub...

sebbbi - Overview

Building the HypeHype renderer. Previously at Unity and Ubisoft. - sebbbi

github.com

soresu · Mar 24, 2025

Did anyone catch this a couple of days ago?!

Microsoft says DirectX Raytracing 1.2 will deliver up to 2.3x performance uplift

Well, in the best-case scenario.

www.tomshardware.com

soresu · Mar 24, 2025

Finally an update to DXR, still 1.x though so they don't want to make breaking changes clearly.

GTracing · Mar 24, 2025

I decided to try and put a number to RDNA4's gen-over-gen perf/area increase. I'm using Locuza's N31 die analysis from YouTube to estimate how big a theoretical monolithic N31 would be. For gaming performance, I'm using the PCGH's 20 game average 1440p raster from their 9070 XT review.

	equivalent 5nm die area	1440p raster gaming perf	perf/area
7900 XTX	463mm^2	65.60%	100%
9070 XT	356mm^2	60.20%	119%

Just using Locuza's estimate as-is, RDNA4 has 19% better perf/area. This is a very rough estimate with a lot of caveats.

In my opinion 463mm^2 is probably high. His calculation takes into account the big factors like removing the GCD-to-MCD links, but I'm sure there are some other small area savings.
This does not take into account node differences. From what I hear, N48 uses N4c and N31's GCD uses N5.
This does not take into account power consumption. The 7900 XTX has 16% higher power consumption and 9% higher gaming performance.
This does not take into account memory bandwidth. RDNA4 can achieve more performance with the same memory bandwidth.
This does not take into account RDNA4's better ray tracing and compute performance, or features like FSR4.

DisEnchantment · Mar 24, 2025

DisEnchantment said:
Not as good as yours but I got mine for 760EUR. I barely play games and my steam has quite a few items only with not much hours in. But I am excited nevertheless.
Still out on a business travel, so by the time I am home the card would have been there.
View attachment 119571

The price gouging is going down

Looks like I fooled myself thinking I got an RX 9070XT what I got was 9070. Seeing 760EUR I though it was jackpot with scalping gone. Alas.
Bought it really fast, when it arrived I thought I got scammed by Reichelt, then I check my order, it was indeed an RX 9070 I ordered.
Was so surreal, seems like I have been losing my sharpness since I got married probably. I bought it while I was on a trip, I guess I am also a moron

Was about to return it but it is OK, very silent compared to my 7800XT. I undervolt it and it can hit 3GHz consistently when I play around in Adrenalin Tuning tab unlike my 7800XT which is not stable at 3GHz

Glo. · Mar 24, 2025

DisEnchantment said:
View attachment 120603

Looks like I fooled myself thinking I got an RX 9070XT what I got was 9070. Seeing 760EUR I though it was jackpot with scalping gone. Alas.
Bought it really fast, when it arrived I thought I got scammed by Reichelt, then I check my order, it was indeed an RX 9070 I ordered.
Was so surreal, seems like I have been losing my sharpness since I got married probably. I bought it while I was on a trip, I guess I am also a moron

Was about to return it but it is OK, very silent compared to my 7800XT. I undervolt it and it can hit 3GHz consistently when I play around in Adrenalin Tuning tab unlike my 7800XT which is not stable at 3GHz

Usually when we are on a trip we are hallucinating.

Kronos1996 · Mar 24, 2025

It looks like the RDNA “.5” architectures were intended as mid-cycle refreshes for mobile since laptop is on an annual cadence. With RDNA 4.5 supposedly cancelled that may no longer be the plan. They’ve essentially been doing it since Vega APUs, multiple generations of modest improvements. Guess they decided to give it an official name though.

adroc_thurston · Mar 24, 2025

Kronos1996 said:
It looks like the RDNA “.5” architectures were intended as mid-cycle refreshes for mobile since laptop is on an annual cadence

Right on money.

moinmoin · Mar 24, 2025

adroc_thurston said:
Right on money.

So AMD is fine with slowing down its cadence more and more?

Josh128 · Mar 24, 2025

moinmoin said:
So AMD is fine with slowing down its cadence more and more?

Yes.

Josh128 · Mar 24, 2025

DisEnchantment said:
View attachment 120603

Looks like I fooled myself thinking I got an RX 9070XT what I got was 9070. Seeing 760EUR I though it was jackpot with scalping gone. Alas.
Bought it really fast, when it arrived I thought I got scammed by Reichelt, then I check my order, it was indeed an RX 9070 I ordered.
Was so surreal, seems like I have been losing my sharpness since I got married probably. I bought it while I was on a trip, I guess I am also a moron

Was about to return it but it is OK, very silent compared to my 7800XT. I undervolt it and it can hit 3GHz consistently when I play around in Adrenalin Tuning tab unlike my 7800XT which is not stable at 3GHz

lol d*mn.

moonbogg · Mar 24, 2025

soresu said:
Finally an update to DXR, still 1.x though so they don't want to make breaking changes clearly.

I feel like Microsoft and everyone else would never bother with ray tracing if Nvidia didn't harass them into it. I agree with above comments that Nvidia should just get out of gaming. All they do is invent expensive crap that gives them an unfair advantage, force game devs to implement their garbage, then basically abandon the gaming market anyway leaving no one with the ability to buy a GPU to run this crap. No one wants to pay $600 for an 8gb POS card that can't run games at 1080p or $1500+ for a half-decent 1440p card. They need to surrender the market and move on completely to their AI thingy. I guarantee everyone will forget all about ray tracing in a single generation. No one will bother with that stupid crap anymore.

Saylick · Mar 24, 2025

moonbogg said:
I feel like Microsoft and everyone else would never bother with ray tracing if Nvidia didn't harass them into it. I agree with above comments that Nvidia should just get out of gaming. All they do is invent expensive crap that gives them an unfair advantage, force game devs to implement their garbage, then basically abandon the gaming market anyway leaving no one with the ability to buy a GPU to run this crap. No one wants to pay $600 for an 8gb POS card that can't run games at 1080p or $1500+ for a half-decent 1440p card. They need to surrender the market and move on completely to their AI thingy. I guarantee everyone will forget all about ray tracing in a single generation. No one will bother with that stupid crap anymore.

Yeah, when people say that Nvidia are the only innovators and they drive the industry forward, it's usually not done out of benefit for the industry but because Nvidia need a compelling marketing argument for customers to keep upgrading so that their gaming revenues are sustainable for the inevitable future. Now that AI has dwarfed gaming revenues for Nvidia, they don't even need to support the gaming market anymore, and based on Lackwell's lackluster reception in the market for various, well-documented reasons, it's obvious where their focus is.

soresu · Mar 24, 2025

moonbogg said:
I feel like Microsoft and everyone else would never bother with ray tracing if Nvidia didn't harass them into it

I think you aren't understanding the business level implications of the RT pivot.

It was basically nVidia going "oh dear, raster gfx isn't going to keep driving profits much longer and post 4K resolution scaling doesn't work because you need an insanely large TV/monitor to appreciate the change".

So in order to keep new generations of GFX cards relevant in the consumer side they had to go with ray tracing, though they definitely jumped the gun at least 5-10 years too early on both hardware and software readiness.

Without DLSS they would never have sold it, even now it is still extremely reliant on denoising to achieve a working framerate at low samples per pixel.

adroc_thurston · Mar 24, 2025

soresu said:
I think you aren't understanding the business level implications of the RT pivot.

It was basically nVidia going "oh dear, raster gfx isn't going to keep driving profits much longer and post 4K resolution scaling doesn't work because you need an insanely large TV/monitor to appreciate the change".

So in order to keep new generations of GFX cards relevant in the consumer side they had to go with ray tracing, though they definitely jumped the gun at least 5-10 years too early on both hardware and software readiness.

a) imgtec had RTRT h/w way before NV
b) it's a sorta-natural evolution of gfx.

The problems are that DXR sucks and it's gonna suck until I guess the nextgen consoles?

soresu said:
Without DLSS they would never have sold it, even now it is still extremely reliant on denoising to achieve a working framerate at low samples per pixel.

Well yeah that's the issue of doing RTRT to begin with.
As a workload, raytracing sucks.
It sucks extra hard for GPUs!

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Member

Diamond Member

Lifer

Lifer

Lifer

Member

Diamond Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Golden Member

Diamond Member

Member

Diamond Member

Diamond Member

Senior member

Senior member

Lifer

Diamond Member

Diamond Member

Diamond Member