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

 

[Heartbreaker]

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

They didn't say forever. AMD strategy shifts all the time. Right now the strategy is to skip high end, but that could easily shift again.

 

[gdansk]

They didn't say forever. AMD strategy shifts all the time. Right now the strategy is to skip high end, but that could easily shift again.

I'm pretty sure it's a low priority going forward. Even if the ML market collapsed to 10% of its current sales it is bigger than gaming.

 

[marees]

Frank Azor​

The other thing I will tell you is that nobody has the final driver. So how can the data be accurate? That's why I can confidently say that the data is not accurate.​

So AMD working on drivers ??

 

[adroc_thurston]

So AMD working on drivers

No, it's just that AMD and NV always gimp pre-review window drivers in interesting ways to prevent leaks from happening.

You was saying FSR4 is only for mobile/handheld.

That's the main focus indeed.

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

They won't. Lisa won't ever sign it off.

So you see not leaving fully.

They exited the market.

AMD strategy shifts all the time.

There is only one strategy in client discrete graphics. You build to kill.
They're not building to kill anymore so they surrendered.

 

[poke01]

Yeah so FSR4 only works with RDNA4 card.

Good.

 

[adroc_thurston]

Yeah so FSR4 only works with RDNA4 card.

No but it's made for it so probably FP8 baseline model

 

[poke01]

No but it's made for it so probably FP8 baseline model

let’s phase it another way, it can work with RDNA3 but AMD chose RDNA4 as the base.

I’m not against that, just as long as all RDNA4 cards support FSR4

 

[marees]

let’s phase it another way, it can work with RDNA3 but AMD chose RDNA4 as the base.

I’m not against that, just as long as all RDNA4 cards support FSR4

Going by AMD's history they are probably testing for RDNA 4 first
But could enable for Navi 31 later

 

[adroc_thurston]

But could enable for Navi 31 later

gonna have 'orrible perf but it'll be funny.

 

[SolidQ]

So 5070 is 550, RDNA4 seems need to be 450 max

 

[gdansk]

So 5070 is 550, RDNA4 seems need to be 450 max

$500. Best I can do.

 

[SolidQ]

$500. Best I can do.

Lol.

https://twitter.com/x/status/1876464556714901603

 

[Heartbreaker]

I'm pretty sure it's a low priority going forward. Even if the ML market collapsed to 10% of its current sales it is bigger than gaming.

That doesn't really matter. That applies even more to NVidia, but they keep making awesome gaming GPUs.

 

[gdansk]

That doesn't really matter. That applies even more to NVidia, but they keep making awesome gaming GPUs.

They can be assured of 90% of the market so it's worthwhile to invest billions. Never going to work for Radeon even if they could get it mostly in order. Better to focus on the IP and only make parts they can assure volume.

 

[adroc_thurston]

That applies even more to NVidia, but they keep making awesome gaming GPUs.

they own the market. The game is always AMD's to lose.

 

[blckgrffn]

Wow, I unboxed the 7900XTX and its an impressive bit of kit.

Buuuuuuut... if I can get a 9070XT and it has horseshoes and hand grenades performance of the 7900XT for $500 and runs at ~300TBP instead of ~400W TBP I think I'll save the $500 and get a 9070 instead.

 

[SolidQ]

AMD just need merge FG + AFMF into AI and gonna be same as DLSS4 multi framegeneration.

 

[adroc_thurston]

AMD just need merge FG + AFMF into AI and gonna be same as DLSS4 multi framegeneration.

they do not have the matrix throughput for that.

 

[gdansk]

Wow, I unboxed the 7900XTX and its an impressive bit of kit.

Buuuuuuut... if I can get a 9070XT and it has horseshoes and hand grenades performance of the 7900XT for $500 and runs at ~300TBP instead of ~400W TBP I think I'll save the $500 and get a 9070 instead.

Which model did you buy?

 

[Keller_TT]

I think we must thank Jensen for doing an AMD at its best to AMD. Hope this forces AMD to drive the Strix Halo cost down for a 8060S and other stuff or loose all relevance and loss of morale. Saying this as someone long invested in AMD and wanting them to succeed, but of late they've started pretending to be King of the hill without the goods for it, and lying bluntly in charts.

If the Nvidia ARM SoC is priced competitively, NV will reinvent the Windows market and AMD can't pretend about their meagre uplifts as "groundbreaking".

 

[SolidQ]

they do not have the matrix throughput for that.

So only with UDNA?
Wonder how will be quality when enable FSR4 FG + AFMF

 

[adroc_thurston]

So only with UDNA?

They have to follow since they're on the cuck chair now, so probably?
Slap a CDNA4 MFMA core per SIMD, and there you go.

 

[blckgrffn]

Which model did you buy?

XFX Mercury Magnetic. Fancy thermal paste and supposedly really good resistance to hot spotting and all that. Was chilling at 360W (lol) and 65c running Helldivers at solid high settings on the new 34” ultrawide. Chill target of 95 fps and the “quiet” profile in the drivers to drop the voltage a solid amount.

Honestly, the drivers make that so simple. It’s great on that front.

 

[SolidQ]

So AMD users have already multigeneration, but ofc NV markteing is WIN

 

[adroc_thurston]

So AMD users have already multigeneration, but ofc NV markteing is WIN

no that's the bad one. chained frames, not one frame.