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

 

[SolidQ]

Your vid is showing a custom preset

You can disable even only DOF, and it will turn to "Custom"

 

[Gideon]

BTW Asus indeed did confirm RX 9070 non-XT is 16GB, thus the 12GB cut-down version might indeed be RX 9060 XT(X) which would be very good news:

https://videocardz.com/newz/asus-unveils-its-radeon-rx-9070-xt-tuf-and-prime-gpus-confirms-16gb-memory

Though It's essential AMD gets the pricing right. With a ~390mm die the SKU's manufacturing price is certainly closer to cut down Navi 31 (7900 XT / GRE) than to Navi 32.

BTW AMD also demoed their Real time path tracing + upscaling (the same the released a paper about last year). This should be an equivalent to Nvidia's DLSS 3 + Ray Reconstruction.

https://twitter.com/x/status/1876746324504179152

I hope someone posts a video / comments about it

 

[SolidQ]

This should be an equivalent to Nvidia's DLSS 3 + Ray Reconstruction.

As we see and know AMD going have RR and Neural Textures, but people already hooked by NV marketing

P.S Photo not very good quality, what is upper left screen? fps?

 

[gdansk]

RDNA4 halo would have beaten 5090 but at the cost of a very big die size.

It's easy to pontificate on parts not built. On paper RTX 5090 should be 1.8x perfomance of the RTX 5080. But big chips fall off in many ways.

 

[SolidQ]

God, i hate guys like this. Ofc AMD not gonna show games for now, until RDNA4 event

 

[insertcarehere]

You can disable even only DOF, and it will turn to "Custom"

Well seeing as I don't have a copy of COD6 in front of me I don't really know what changes between an 'Extreme' preset and actual 'max settings' . The point being is that the 9070xt benchmark run may not be as comparable as we think due to whatever changes people have been doing to their benchmark runs.

 

[SolidQ]

if that real, what is there new? RT?

https://videocardz.com/newz/amd-navi-48-rdna4-gpu-for-radeon-rx-9070-pictured-may-exceed-nvidia-ad103-size

 

[poke01]

As we see and know AMD going have RR and Neural Textures, but people already hooked by NV marketing

P.S Photo not very good quality, what is upper left screen? fps?

Maybe AMD should have invested some time at their event for RDNA4 to market Neural Textures or (RDNA4 ITSELF) like NV did.

 

[gdansk]

if that real, what is there new? RT?

It has wandered into the land of ML, RT and it became fat.

Maybe AMD should have invested some time at their event for RDNA4 to market Neural Textures or (RDNA4 ITSELF) like NV did.

Maybe. But the negativity around Radeon is so high that they should always shut up until stuff is shipping.

 

[poke01]

GB203 RTX 5080 is 377 mm². Why can't AMD match NV's PPA? Are NV's designs really dense?

 

[gdansk]

Why can't AMD match NV's PPA?

No idea why. But wouldn't you, as a product manager, cancel To Mega Navion if the reasonably sized part couldn't match the last gen die size peer - the RTX 4080?

 

[poke01]

No idea why. But wouldn't you, as a product manager, cancel To Mega Navion if the reasonably sized part couldn't even match the RTX 4080?

so it was a cost and engineering problem that made rdna4 halo not worth making?

 

[gdansk]

so it was a cost and engineering problem that made rdna4 halo not worth making?

We'll see soon. But it seems like a prudent decision to me.

 

[Thunder 57]

We'll see soon. But it seems like a prudent decision to me.

Maybe because people have been buying Nvidia without looking at the competition lately. The 3050 outselling the 6600 by a massive amount proved that and was just sad.

 

[SolidQ]

The 3050 outselling the 6600 by a massive amount proved that and was just sad.

That was heavy marketing RT/DLSS, but now if FSR4 good and RT also, what arguments for NV? aside specific Cuda for work

 

[poke01]

That was heavy marketing RT/DLSS, but now if FSR4 good and RT also, what arguments for NV? aside specific Cuda for work

DLSS4?? Most of its features are coming to all RTX cards. Well the ones that matter anyway.

3d rendering is another one, a huge one for designers and engineers. Blender improvements will be big because of huge bandwidth increases

 

[SolidQ]

3d rendering is another one, a huge one for designers and engineers.

I mean for gamers

DLSS4??

https://twitter.com/x/status/1876624801600532711

Maybe it's going to FSR4?

 

[Thunder 57]

That was heavy marketing RT/DLSS, but now if FSR4 good and RT also, what arguments for NV? aside specific Cuda for work

RT was DoA on the 3050. And who needed DLSS when a 6600 was far more powerful at the same price?

 

[SolidQ]

RT was DoA on the 3050. And who needed DLSS when a 6600 was far more powerful at the same price?

Yeah, but random people hearing marketing NV, NV, RT, RT, DLSS, DLSS etc

 

[Thunder 57]

Yeah, but random people hearing marketing NV, NV, RT, RT, DLSS, DLSS etc

And the halo effect. Their loss.

 

[SolidQ]

And the halo effect. Their loss.

to be fair not all people, i know some people who was only NV, but now they more rational.

Enabling Neural Rendering in DirectX: Cooperative Vector Support Coming Soon - DirectX Developer Blog

Neural Rendering: A New Paradigm in 3D Graphics Programming In the constantly advancing landscape of 3D graphics, neural rendering technology represents a significant evolution. Neural rendering broadly defines the suite of techniques that leverage AI/ML to dramatically transform traditional...

devblogs.microsoft.com

 

[poke01]

I mean for gamers

for games, DLSS4. That new model looks good, of course more testing is needed. I'm not upgrading to any GPU either from AMD or NV thanks to DLSS4 support on 40 series. This was a calculated move from Nvidia.

 

[poke01]

Enabling Neural Rendering in DirectX: Cooperative Vector Support Coming Soon - DirectX Developer Blog

Neural Rendering: A New Paradigm in 3D Graphics Programming In the constantly advancing landscape of 3D graphics, neural rendering technology represents a significant evolution. Neural rendering broadly defines the suite of techniques that leverage AI/ML to dramatically transform traditional...

devblogs.microsoft.com

oh look neural rendering from MS themselves

 

[insertcarehere]

RT was DoA on the 3050. And who needed DLSS when a 6600 was far more powerful at the same price?

Putting aside that 6600s were barely available outside the US for much of the past few years (think 2022/23). Nvidia is practically the only choice for CS Majors who need to learn programming with GPUs (and game a bit for R&R) . 6600 being 40% more powerful or whatever doesn't matter when RoCm is/was a tire fire.

 

[poke01]

RDNA4 is capable of Neural rendering. Lets see if they release support for it next month