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

 

[gaav87]

Well i still think rdna4 simds can access other simds beoynd the CU vs rdna3 (within cu) adroc already laughed at me for this ~1year ago.
This would be true dual issue for true 50Tflops of single precision = 4080s (has 50)

 

[CastleBravo]

It will be interesting to see how AMD's architecture stacks up against Nvidia's with them both being monolithic dies made on the same process.

 

[adroc_thurston]

Well i still think rdna4 simds can access other simds beoynd the CU vs rdna3 (within cu) adroc already laughed at me for this ~1year ago.
This would be true dual issue for true 50Tflops of single precision = 4080s (has 50)

shader oomph is whatever. look elsewhere.

 

[gaav87]

shader oomph is whatever. look elsewhere.

well there are few things pointing to this.
1. they changed the instructions
2. better prefetch / better identify and optimize dual issue opportunities
3. more flexible instruction scheduling

But this would req some global memory

 

[adroc_thurston]

well there are few things pointing to this.
1. they changed the instructions
2. better prefetch / better identify and optimize dual issue opportunities
3. more flexible instruction scheduling

Again, flops are whatever. look elsewhere.
They're getting a ton of perf outta 256b 20Gbps GDDR6 so try to imagine how and why.

 

[SolidQ]

Again, flops are whatever

Cerny!

 

[gaav87]

9070xt is not compatible with rdna2 cold plates.
If it was 220mm2 it should fit.

 

[adroc_thurston]

9070xt is not compatible with rdna2 cold plates.
If it was 220mm2 it should fit.

The package is just different for both dies.

 

[gaav87]

The package is just different for both dies.

Ye but why would the package be bigger than rdna2 if it was 220mm2 ?

 

[adroc_thurston]

Ye but why would the package be bigger than rdna2 if it was 220mm2 ?

Not bigger, just geometrically different.

 

[SolidQ]

I'm looking at this picture. I don't think it's perfomance chart. Maybe currently street prices. Because 7700XT lose to 4060ti doesn't make any sense

 

[adroc_thurston]

I'm looking at this picture. I don't think it's perfomance chart. Maybe currently street prices. Because 7700XT lose to 4060ti doesn't make any sense

It's the pricing stack, not the performance one.

 

[SolidQ]

It's the pricing stack, not the performance one.

Based on this chart 9070 nonXT is higher price, than 7800XT, that mean more perfomance, than 7800XT
Looks like 9070 non XT planning 500-530$

 

[gaav87]

Not bigger, just geometrically different.

Well anyway next week is annocement cant wait too see what they cooked.

 

[gaav87]

It's the pricing stack, not the performance one.

yes agree

 

[ToTTenTranz]

This has to be the strangest GPU (un)release ever. The 9070 cards are just there at the Gigabyte booth, with anything but the name: no chip specs, no memory specs, no release date, nothing.

Here are my pics. 3x8pin version confirmed.

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

Isn't this guy the FSR1/RSR creator?

 

[SolidQ]

yes agree

Looks like to me, based on amd slide Prices, 9600XT = 7800XT perf, 9600 is 7700XT

 

[ToTTenTranz]

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

I hope someone posts a video / comments about it

Went there right now.

It's running on an engineering sample GPU so I'm assuming it's the 9070/XT.

Frame rate seems to be around 30FPS (there's no frame counter) but for some reason it's set with VSync off nor freesync, so lots of tearing when you pan around.

As for how the demo looks, I really can't say it looks transformative. I guess it's more interesting to know what's happening behind it especially when you turn denoising off and then you definitely notice it.

 

[SolidQ]

Interesting statement from dev about AMD RT vs NV RT

As for how the demo looks, I really can't say it looks transformative

Does is look to similar C2077 indoors areas or worse?

Frame rate seems to be around 30FPS

Resolution? Native or upscaled?

 

[PJVol]

Went there right now.

Can you run a COD6 benchmark there?

 

[SolidQ]

Can you run a COD6 bencmark there?

Yeah, + need check it's real 9070 or XT

 

[iLLusiveMan]

Yeah, + need check it's real 9070 or XT

If the sticker on that GPU is the same as the engineering sample posted above good luck finding what GPU is in there.
I don't think anyone can walk in there and run CoD on that PC. If that was the case we would have seen all tech media posting CoD benchmarks today
That benchmark provided by IGN was most likely allowed by AMD. With was purpose ? IDK. Will it bite them in the back ? IDK.

 

[blckgrffn]

Either way I don't care. I want the official specs and third party reviews.

I'm tired of all the leaks and speculations.

So say we all.

 

[gdansk]

So say we all.

It's very sad. We were supposed to have dubious slides from two companies to compare by now. My disappointment is immeasurable and my whole week is ruined.

 

[poke01]

Here's to hoping. AMD and FSR is what gives me the option for Frame Gen in CP2077 on my 3080. Thanks Nvidia!

You actually want fake frames???

I mean Nvidia also gave improvements to 30 series with DLSS4 that are way more useful than fake frames especially for a 3080 user.