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

 

[deasd]

Time Spy Extreme 14558 is literally at 7900xtx/4080s level

edit: the driver was still immature.

 

[Win2012R2]

not releasing a bigger RDNA4 will end up as yet another blunder from the graphics leadership. Not believing in their brand, engineering - and something about missing 100% of the shots you don't take.

Whoever made that decision should be fired, unless it's Lisa, then all is forgiven.

 

[jpiniero]

If those numbers were accurate, that should make it close to the 5070 Ti in RT and way faster in raster.

The new ~350mm^2 numbers vs the old ~240mm^2 have really changed perceptions. Looks like everyone here is boarding the hype train for RDNA4, even though AMD just essentially cancelled the launch in the last minute, which should be a red flag.

I think they expected the 50 Series to be a lot better so they were unsure about pricing.

If AMD ends up competitive with Nvidia (relative to size, power etc.), not releasing a bigger RDNA4 will end up as yet another blunder from the graphics leadership. Not believing in their brand, engineering - and something about missing 100% of the shots you don't take.

If this ends up being $700, that might be about as high as they can go and get people to buy AMD GPUs, regardless of performance.

 

[SolidQ]

Did he mention if this score was of stock or OCed gpu

Based on Gpu Z screen it's just AIB card.

LMAO

Price rumor is too good to be true

 

[PJVol]

Looks like everyone here is boarding the hype train for RDNA4

Yep, Choo... choo... Jump on!

 

[misuspita]

I think they expected the 50 Series to be a lot better so they were unsure about pricing.

Yeah, I think this is the crux of the matter. They expected 5090 and 80 to be much more powerfull and not reliant on AI trickery and probably in the last hours lound out about real performance of Nvidia lineup and had chose to wait it out even though the board makers were already with them on display.

I'm actually marginally optimistic they have a good product on their hands by fooling aveeryone with their small die strategy, whici is not small anymore and has to have higher than expected perf to back it up

 

[coercitiv]

If AMD ends up competitive with Nvidia (relative to size, power etc.), not releasing a bigger RDNA4 will end up as yet another blunder from the graphics leadership. Not believing in their brand, engineering - and something about missing 100% of the shots you don't take.

This is fun part: they either screwed up the entire lineup or they screwed up by abandoning the big chip. The gray area where both are untrue is kinda' unlikely (unless the arch has issues with scaling)

 

[PJVol]

Based on Gpu Z screen it's just AIB card.

Oh, I missed that... based on this info the gpu is at stock settings (if we're to believe OD telemetry)

 

[SolidQ]

based on info the gpu is at stock settings.

and price rumor looks insane, i don't believe in this price

 

[JustViewing]

Whoever made that decision should be fired, unless it's Lisa, then all is forgiven.

Maybe power is the limiting factor? If these cards uses 330w, big Navi might have used 500+W.

 

[PJVol]

big Navi might have used 500+W.

Who cares if they could deliver?

 

[SolidQ]

big Navi might have used 500+W.

NV 575W, who cares?

 

[Keller_TT]

If this ends up being $700, that might be about as high as they can go and get people to buy AMD GPUs, regardless of performance.

Just a mere glimpse of promising performance, and folks loose perception of reality and purpose. I still think the reference XT will be ~7900 XT, like 7800 XT vs 6800 XT and offering optimum PPW at 260W, and should be priced at $500 max for any gen on gen value and upgrade. AIBs will likely go up to $600.
If you're stagnant in value for 2 gens when the real 7900XT was a $700-750 launch card overpriced to upsell the XTX, NV will take the cake with 5070. AMD have done enough to drive down costs for RDNA 4, and should transfer that value to the customer.

 

[leoneazzurro]

https://twitter.com/x/status/1877312568051499194

Seems they only had to define price

 

[PJVol]

Just a mere glimpse of promising performance, and folks loose perception of reality and purpose

You think it is unrealistic to expect 7900xt and 7900xtx levels from ref. and aibs resp. ?
Get off our train!

 

[JustViewing]

NV 575W, who cares?

Once upon a time 300w was a dual card territory

 

[lucasworais]

Damn, they could have beaten NV so easy by these benchmarks. Too bad they canned the big stick.

 

[linkgoron]

Maybe power is the limiting factor? If these cards uses 330w, big Navi might have used 500+W.

Nvidia have basically removed power use norms a few gens ago, especially with the 30 gen.

I think they expected the 50 Series to be a lot better so they were unsure about pricing.

AMD rebranding these as 9080XT as we speak.

 

[coercitiv]

Seems they only had to define price

This takes me back 6 months ago:

Combined with this would be Nvidia waiting until next year to introduce 5070 and under. How would AMD know how to price their products without guidance from professionals?!

 

[Keller_TT]

You think it is unrealistic to expect 7900xt and 7900xtx levels from ref. and aibs resp. ?
Get off our train!

Can you read?

 

[PJVol]

Can you read?

Yes

 

[SolidQ]

Seems they only had to define price

Next week shoud be RDNA4 event, then reviews 22+
Mlid hinted 15th january. Funny, but MLID from leakers was closest with 9070XT perf

 

[leoneazzurro]

This takes me back 6 months ago:

Yes, but this is what happens when your competitor has 85+% of the market.

 

[soresu]

Funny, but MLID from leakers was closest with 9070XT perf

If AMD is telling the truth and no one has the final drivers then it's basically just guess work on his part.

 

[SolidQ]

If AMD is telling the truth and no one has the final drivers then it's basically just guess work on his part.

Yeah, let's just wait final drivers.
That leak from August