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]

Mindshare if formed by constant propaganda, viral marketing and lies by the truckload, otherwise a firm that sold dozen millions faulty chips that busted in as much laptops wouldnt be leader these days.

It's formed by the biggest chungus.

 

[RnR_au]

AMD posted this

Interesting that they benched their AI performance using Vulkan rather than ROCM. The 7000 series is said to be supported under ROCM.

And they don't show prompt processng scores.

 

[blackangus]

GeRadeon

RForce!

 

[gaav87]

RForce!

GradeON

 

[gaav87]

Well now that we know 5070 will be slower than 4070S and 5070ti than 4070tiS.
Lisa will price 9070xt at 800$ and frank azor saying they wont price it at 899$ will be true

 

[Meteor Late]

Well now that we know 5070 will be slower than 4070S and 5070ti than 4070tiS.
Lisa will price 9070xt at 800$ and frank azor saying they wont price it at 899$ will be true

They can't, because the name is 9070 XT, so a 70 class card by name. So as Nvidia has the marketshare, people will only buy it (well some people) if it's cheaper than Nvidia's 70 class. So if this thing performs better than 5070 ti, doesn't matter, it will have to be preferably at least 100$ cheaper. So 650$ it is.
Do we really think AMD is not making a huge profit selling this card at 650$ with a ~350 mm2 die with GDDR6 memory?

Also, 5070 ti will be faster than 4070 ti super because it has like 5% more cuda cores, it will be around 10% faster or so, because it is a similar delta than the one between 5080 and 4080 super. To me it looks like 5070 ti will be around 5% slower than 4080 non super. So 9070 XT if it is around 7900 XTX performance on average, then it should definitely be a little faster than 5070 ti at least.

 

[adroc_thurston]

Do we really think AMD is not making a huge profit selling this card at 650$ with a ~350 mm2 die with GDDR6 memory?

they can make more.

 

[Meteor Late]

they can make more.

Sure, buyers can also buy less.

 

[Meteor Late]

Ultimately, I think what AMD will do is price 9070 XT at 699$, just enough so that they can say they are pricing it lower than 5070 ti while beating it even if by 7-8% or so.
In the end, the vast majority of buyers won't care about 7-8% more raster performance (assuming that's the delta, we will see) if DLSS is still much better with the new model (we will see). AMD still needs to be cheaper even if ever so slightly.

 

[IEC]

$700 would be a lame price. $500 would be a welcome change in a stagnant and overpriced GPU market where the most interesting launch in recent memory is Intel's B580.

Anyone who bought a cheap 6800/XT or RTX 3080 is feeling pretty good about their bang for the buck compared to the gimped 8GB 128-bit $300+ cards on offer.

 

[adroc_thurston]

Sure, buyers can also buy less.

No one buys Radeon anyway.

$500 would be a welcome change in a stagnant and overpriced GPU market where the most interesting launch in recent memory is Intel's B580.

It'll force Nvidia pricecuts at $500. Then it would die.

 

[gdansk]

AMD does have to price very carefully.
GB205 is smaller, they can't win if Nvidia makes price cuts.
But I'm not sure if $500 would cause that. Nvidia seems pretty happy with its MFG and DLSS4 moats.

 

[adroc_thurston]

GB205 is smaller, they can't win if Nvidia makes price cuts.

it's also way slower.

Nvidia seems pretty happy with its MFG and DLSS4 moats.

No one cares about fakeframes and FSR4 is good nuff.

 

[gdansk]

it's also way slower

Doesn't matter. 3050 proved it. Nvidia didn't have to cut prices because of inferior performance.

No one cares about fakeframes and FSR4 is good nuff.

Good enough to cause Nvidia to cut prices? I'm not so sure.

 

[adroc_thurston]

Nvidia didn't have to cut prices because of inferior performance.

Yeah. Kinda.
I guess AMD has to cart a wagon of dollar to alex cumraglia.

Good enough to cause Nvidia to cut prices? I'm not so sure.

Fish is fish.

 

[Thunder 57]

No one buys Radeon anyway.

It'll force Nvidia pricecuts at $500. Then it would die.

You are annoying. Stop speaking in absoultes.

 

[adroc_thurston]

You are annoying. Stop speaking in absoultes.

But it's the truth. It has been the truth for like a million years.

 

[Shmee]

I for, one, have also bought Radeon before, as well as Geforce. But I have had many cards over the years. I even had a bunch of R9 290s xD Great cards at the time, and that was when crossfire was at its peak. And yeah I had 5870 cards as well, first high end Radeon card I got. Coming from a GTX 260. Next I went for 6950 CF and unlocked the shaders. Good times.

Now that is just me, but when you say "no one buys Radeon", I would say that is borderline flame bait/trolling.

 

[adroc_thurston]

Now that is just me

Yes but we're both tech nerds.
As far as mainstream perception goes, Radeon doesn't exist. When it does, it exists to force NV pricecuts.

 

[HurleyBird]

No one cares about fakeframes and FSR4 is good nuff.

More than good enough. But the issue, less so for the 9070 series and more so for lower performance tiers, is that FSR4 isn't in anything yet. There are plenty of titles it will never be in, some of which still make sense to use upscalers, and for those games DLSS >>> FSR. In a sense, it doesn't matter much good FSR4 is until it has some meaningful market penetration. If you're AMD and your launch price consideration assumes FSR ~= DLSS, then that's not a smart consideration. Once FSR4 has become more ubiquitous, then sure.

 

[adroc_thurston]

In a sense, it doesn't matter much how good FSR4 is until it gets some actual market penetration

Driver will hijack any FSR3.1 .dll for FSR4 so at least some baseline is covered.
The problem for RDNA4 parts is down to the brand perception and no way to fix it.

 

[Mopetar]

AMD posted this

Just great. AMD might release a decent card at a reasonable price and now they've ensured that smelly hobos will line up in droves so that they can flip them for $2,000 to tech startups with more venture capital than sense.

As much fun is it might be to have a GPU use cutting edge learning models to create pictures of Taylor Swift pegging the Easter Bunny, this is turning out worse than the crypto boom. At least those people got imaginary money they could spend on NFTs of pictures of Taylor Swift pegging the Easter Bunny.

 

[SK10H]

AMD posted this

I test it on 7900xtx but with 24.12.1 driver on LM studio 0.3.8 on Win10, using Deepseek R1 distill owen 14B/32B. It outputs Gibberish no matter how simple the question.
Gave up on it and use Ollama, docker, WSL2 with Open WebUI, work fine on both.
At least the speed is comparable and setup is easier than doing image generation with Zluda ComfyUI.

 

[SolidQ]

MLID add part to previous AMD slides leak

 

[Hans Gruber]

MLID add part to previous AMD slides leak

MLID is an idiot. He knows nothing. He just reads google news articles to people on youtube who either cannot read or are too lazy to read articles for themselves.