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

 

[Mahboi]

AV1 seems not getting any improvements

How the heck is AV1 worse?
That's a twist I didn't expect. Hopefully it's just a driver issue, I remember my AV1 encoder kept dropping frames at 4K on the 7900 XT, after some months it got entirely fixed and I went from constant stutters to absolutely fine, just 90% usage of the media encoder. Very happy with it now, and I get 4K video of games recording 1h45mins worth of video in just 8Go files. Honestly not changing off AV1 anytime soon.
And I'm not going to bother trying to stream on Twitch. They get to AV1 or they don't but I'm done with them.

 

[GTracing]

Hmmm, according to MC's website, there's a lot of MSRP 9070XTs still in stock at my local MC. By my count, there's at least 30.

Microcenter's online stock isn't accurate

 

[blckgrffn]

Microcenter's online stock isn't accurate

And you aren't in line right now, so....

 

[Hitman928]

For those that care, Linux performance seems to be very good. A little bit of pain getting the drivers installed right now though.

 

[Panino Manino]

I don't care. AMD said things that didn't come true. AMD is not a person, it doesn't have feelings. There's no point defending the outcome.

Consumers were promised something that didn't materialize. The end.

Doctor Su is flawless on the CPU side of AMD, if only Lisa was as good as her and kept her promises.
Can't we switch one for the other for a year?

 

[moonbogg]

I told you guys that cube wasn't solid.

 

[Hail The Brain Slug]

Doctor Su is flawless on the CPU side of AMD, if only Lisa was as good as her and kept her promises.
Can't we switch one for the other for a year?

I don't know, Zen 5 was pretty rough with the whole trying to sell to gamers and regular consumers a close to zero uplift for their use cases at a big price increase.

 

[gdansk]

I told you guys that cube wasn't solid.

A few thousand cards isn't enough to satisfy about 2 months of demand as graphics cards evaporated. But it is enough to build a cube. I don't know if the cube is metaphorical to you but it wasn't a lot of cards.

 

[itsmydamnation]

A few thousand cards isn't enough to satisfy about 2 months of demand as graphics cards evaporated. But it is enough to build a cube. I don't know if the cube is metaphorical to you but it wasn't a lot of cards.

I think it's called a joke. But I'm not sure , haven't seen one before

 

[Hitman928]

I told you guys that cube wasn't solid.

 

[marees]

I told you guys that cube wasn't solid.

More than 500 people queued outside each micro center

So yes the cube was solid

https://twitter.com/x/status/1897689641710723369

 

[coercitiv]

I think it's called a joke. But I'm not sure , haven't seen one before

The joke wasn't solid though, that's the problem. Illusory humor.

 

[Win2012R2]

Also these cards take many weeks to create from start to finish.

Months, many

I sure hope AMD is serious and will be pushing AIBs for MSRP pricing.

Why would they do that?

If the market value is much higher then AIBs will find ways to sell it for more money, why should AMD let them have those massive profits while taking all the heat for higher prices?

 

[coercitiv]

Frank is happy

 

[poke01]

Yep, it's why they're incremental. They've lived and breathed oversupply implications as a result of their GF relationship during the WSA days. They don't want to go back. Nor should they. If there were any prior green shoots that people would actually buy their GPUs, or had magic knowledge way ahead of time that Nvidia would stumble this hard, it might be different. They'd also have to convince the whole supply chain that this time, this time it might be different. What i don't get is the ire, or really the lack of comprehension here from some, that outside of some outlandish forecasting AMD would never be reasonably able to supply enough GPUs into the market in this particular window of buying opportunity

Just don’t.

 

[SolidQ]

Path Tracing is 4070S level, wonder if AMD gonna improve PT perf with driver further

 

[poke01]

Frank is happy
View attachment 119123

I’m sure he is. Tell him to visit Europe and Asia too. Not just MC, where they oversupplied the stock and to take PR photos

 

[blackangus]

Congratz, I guess...

LOL! That should be a <smack> for buying over MSRP! =)

 

[moonbogg]

More than 500 people queued outside each micro center

So yes the cube was solid

https://twitter.com/x/status/1897689641710723369

Hmm. Perhaps the cube was solid? It depends on how many of those people actually got a card though. Also, I'd bet 5 dollars that 80% of people in that line aren't trying to game, but instead are at work trying to resell a GPU.

 

[jpiniero]

but instead are at work trying to resell a GPU.

Yup

 

[SolidQ]

More news is coming

https://twitter.com/x/status/1897741520200962308

 

[poke01]

More news is coming

https://twitter.com/x/status/1897741520200962308

Very excited for PS6 and how FSR4 models will improve upscaling on PS.

 

[SolidQ]

Very excited for PS6 and how FSR4 models will improve upscaling on PS.

I think Sony will ditch PSSR. Seems PSSR exist because GPU frankenstein in PS5pro

 

[Mopetar]

Re: efficiency, yeah, 9070 is pretty efficient perf/watt, but 9070XT is pushed too far out of its efficiency sweet spot, its no better than 7900XT or XTX.

AMD has no card above the 7090 XT so no reason not to push it. If you want you can reduce the clocks or undervolt it. The reality is that no doing this would have made it review worse and they wouldn't charge any less for it.

People were so excited about the 5070 at $550 when Jensen unveiled it, and for my purposes the 9070 is in every way superior so... It is what it is.

That's because JHH was telling people it was raining by comparing the card to a 4090 at only $550. Who wouldn't be excited at the prospect of that?

 

[SolidQ]

AMD has no card above the 7090 XT so no reason not to push it.

Let's hope AMD is gonna announce GDRR7 version with +15% perf