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]

Ian has a writeup on it, but the short version is that AMD claims right now was just not the right time in the product lifecycle to do a major announcement. Still reading it, but seems fairly content-free.

"we're waiting for NV to define the product stack to plug our stuff into" yeah to surprise of no one.

from TPU
AMD's 3rd Gen Ray accelerator is expected to reduce the performance cost of ray tracing, by putting more of the ray tracing workload through dedicated hardware, offloading the SIMD engine. Lastly, AMD is expected to significantly upgrade the media acceleration and display I/O of its GPUs.

standard stuff.

looking how massive cards and 3x8 pins

they're useless premium AIC options. It's like that ROG Strix RTX 3050.

 

[Bryo4321]

They also say the performance leaks are wrong lol....

I take this to mean no 4070 Ti Super levels of raster or RT.

No, I strongly suspect they were talking about those leaked benchmarks being wrong. (7900 gre +5% or something like that). From their own press slides they slotted the 9070xt up to a 7900xt.

 

[SolidQ]

next time marketing team, should make announce for future GPU's events, not like this. That why a lot people dissapoint. AMD need fix their marketing team.

Could you maybe rank your priorities for RDNA 4, why you picked them, why you ranked them so high?

I would say our number one priority is focusing on improving performance in the areas that gamers care about most. In this generation, you'll see big ray tracing improvements, big MLOps improvements for things like FSR4 and ML Super Resolution.

 

[ToTTenTranz]

they, in fact, are not. lol

How are you claiming it's not good value if they're waiting for Nvidia's prices?

If they really didn't care about value, they'd have announced those already.

 

[adroc_thurston]

How are you claiming it's not good value if they're waiting for Nvidia's prices?

They're gonna go with the standard minus 50 bucks from NV comp. That's not value™.

If they really didn't care about value, they'd have announced those already.

They're literally waiting to see how hard NV hikes the price.
Really really standard GPU wars stuff.

 

[jpiniero]

They're gonna go with the standard minus 50 bucks from NV comp. That's not value™.

Depending on how high nVidia goes, they could do -100 and offer more VRAM and be happy with the margins.

Problem is, people want more and reviewers will slam AMD comparing the prices to RDNA 3 firesales.

 

[Meteor Late]

next time marketing team, should make announce for future GPU's events, not like this. That why a lot people dissapoint. AMD need fix their marketing team.

Could you maybe rank your priorities for RDNA 4, why you picked them, why you ranked them so high?

I would say our number one priority is focusing on improving performance in the areas that gamers care about most. In this generation, you'll see big ray tracing improvements, big MLOps improvements for things like FSR4 and ML Super Resolution.

Oh so gamers don't care about raster performance this time, right AMD? lol

 

[linkgoron]

Ian Cutress

Also to confirm, this isn't a case of wait and see where the competition sits with their new generation and align appropriately?​

Frank Azor​

​

There isn't any one simple thing. We didn't bring RDNA4 to the show because of the competition. It's a multitude of different things. Did that weigh into our decision? Yeah, of course it weighed into our decision. But it isn't any one thing. It's giving it the proper stage.

They're at least not completely lying about why they didn't release concrete numbers/prices (bold mine)

David McAfee
The performance data that's out there on RDNA4 is completely inaccurate.

Weird comment, as they've already said that performance is basically ~7900XT level which is where the latest "rumors" have been.

Oh so gamers don't care about raster performance this time, right AMD? lol

They have been relatively competitive in raster (price-wise).

 

[adroc_thurston]

they could do -100 and offer more VRAM and be happy with the margins.

No they won't be lol.

Oh so gamers don't care about raster performance this time, right AMD? lol

It's a shill point. Without the halo you have to adhere to the shill points of those who have the halo.

 

[jpiniero]

In any case, no RDNA 4 mobile is a total fail.

 

[gdansk]

Oh so gamers don't care about raster performance this time, right AMD? lol

They're fixing what people complain about. That you're complaining about them trying to fix the two hammers which has been used to smash Radeon for 4 years shows how they just can't win.

The only way they can win is by offering feature parity, bigger chip, with better raster and ray tracing. Sounds good! Except of course it'll have be at a lower cost than Nvidia too or no one will buy it and they have to do it on a lower R&D budget. I hope they stop making graphics cards for gamers. They deserve to be shafted by Huang for eternity.

 

[adroc_thurston]

In any case, no RDNA 4 mobile is a total fail.

I'm pretty sure both it and mobile Blackwell slipped into Q2. In any case, RDNA4 mobile will have like no slots due to obvious reasons.

 

[SolidQ]

r no one will buy

I think once they fix AI software, they can make again Halo cards, and sell to AI enthusiasts

 

[RnR_au]

they're useless premium AIC options. It's like that ROG Strix RTX 3050.

Nah... I prefer my toast buttered... AMD has a nice clocker on their hands.

 

[Meteor Late]

They're fixing what people complain about. That you're complaining about them trying to fix the two hammers which has been used to smash Radeon for 4 years shows how they just can't win.

The only way they can win is by offering feature parity, bigger chip, with better raster and ray tracing. Sounds good! Except of course it'll have be at a lower cost than Nvidia too or no one will buy it and they have to do it on a lower R&D budget. I hope they stop making graphics cards for gamers. They deserve to be shafted by Huang for eternity.

What people expect from AMD is much cheaper cards than Nvidia, so if an Nvidia card costs 500$, I would want an AMD equivalent card for 300$. That's why we want AMD around, so we can game for very cheap.

 

[jpiniero]

I'm pretty sure both it and mobile Blackwell slipped into Q2.

Videocardz has PRs from both Razer and MSI. The MSI one doesn't mention a release date, but the Razer one says Q1. Course they probably mean end of March so maybe.

 

[adroc_thurston]

I think once they fix AI software, they can make again Halo cards, and sell to AI enthusiasts

The bubble will be long over by then and they're not making halo GPUs anymore. Ever.

What people expect from AMD is much cheaper cards than Nvidia, so if an Nvidia card costs 500$, I would want an AMD equivalent card for 300$. That's why we want AMD around, so we can game for very cheap.

There's no 'we' and you're not getting that.

Videocardz has PRs from both Razer and MSI. The MSI one doesn't mention a release date, but the Razer one says Q1. Course they probably mean end of March so maybe.

Yeah stuff got pushed out.

 

[gdansk]

What people expect from AMD is much cheaper cards than Nvidia, so if an Nvidia card costs 500$, I would want an AMD equivalent card for 300$. That's why we want AMD around, so we can game for very cheap.

And so they pursue other markets as soon as they present themselves.

Gamers will go buy Nvidia cards. It's better for everyone if the 6070 is $1000. Radeon can finally do Nvidia -$200 like you want. And maybe they can make some money at even 10% market share.

 

[SolidQ]

Ever.

never say ever. People out from companies and come to companies. There no official statement from AMD about not making HALO anymore

 

[adroc_thurston]

There no official statement from AMD about not making HALO anymore

They literally said that. Lmao. In an official PR blurb more or less.

 

[SolidQ]

They literally said that. Lmao. In an official PR blurb more or less.

You was saying FSR4 is only for mobile/handheld.

Here Jack answer
JH: One day, we may. But my priority right now is to build scale for AMD

So you see not leaving fully. Just need RoCM and AI software

 

[marees]

If the leaks are wrong on performance, AMD, then how about publishing your own data ??

 

[gdansk]

If the leaks are wrong on performance AMD, then how about publishing your own data ??

If they deny something it is true. But which leaks were they talking about?

 

[gdansk]

Just need RoCM and AI software

You think that will ever pay off for consumer parts?
I have some AI stocks to sell you.

 

[SolidQ]

But which leaks were they talking about?

There was 2 leaks GRE and 4080 perf

You think that will ever pay off for consumer parts?

don't know, but people buying 4090 as AI card. Same AMD can do, but they don't have CUDA/software. Rocm and software still in development