Question Speculation: RDNA3 + CDNA2 Architectures Thread

[uzzi38]

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 is much bigger than from RDNA1 to RDNA2. We should expect many big improvements in GFX11. 🤔" / Twitter

 

[Tup3x]

Is it only me, or others also feel as if this release is rushed, and AMD released an unfinished product?

You are definitely not alone...

 

[Timorous]

Is it only me, or others also feel as if this release is rushed, and AMD released an unfinished product?

No so sure about rushed. Maybe slightly due to finding and working around some of the hardware bugs.

Think the issue is clock speeds being low. With higher clocks and the current drivers you would have some games being between the 4080 and 4090 and other games being on par with the 4090 in raster performance so reviewers would be able to say stuff like 'comfortably beats the 4080 and sometimes matches the 4090' instead of 'generally matches and sometimes beats the 4080'. That extra 10% raster performance would lead to a huge shift in perception.

 

[Karnak]

Is it only me, or others also feel as if this release is rushed, and AMD released an unfinished product?

They told everyone 2022 so finished or not they had to launch RDNA3. But yeah to me it also does look like it's rushed though.

 

[Kocicak]

Is it possible that the high video playback power consumption is just a firmware/driver issue that can be fixed?

 

[eek2121]

I only read the Ars Technica review thus far, admittedly, but it looks like the 7900XTX is faster than the 4080 while consuming more power.

Will look at more reviews when I have time.

EDIT: Glancing through the GN video, it looks like AMD's reference model is garbage and the drivers also need work. Sad. Performance looks okay-ish. RT performance is a bit faster my 3090 FTW3 Ultra, Raster is ahead of the 4080.

 

[eek2121]

Is it possible that the high video playback power consumption is just a firmware/driver issue that can be fixed?

Yes, along with the fan curve issues. I now 100% believe now that this release was rushed, and we will get a refreshed model with all the problems fixed down the road.

 

[leoneazzurro]

As far as I understood the chip CAN achieve >3GHz but if gaming this makes the power consumption explode.
AMD Radeon RX 7900 XT 'Navi 31' GPU Overclocked To 3.7 GHz Front End Clock & 3.5 GHz Shader Clock (wccftech.com)
Wonder which are the hardware bugs leading to this and how will they fix them.
A part of the performance is surely lost on the software side, too. The move to a software scheduler is not without pain.

 

[TESKATLIPOKA]

N31 in this state is a FLOP only the price is at least good.
Power draw during Video and multimedia is embarrassingly bad.

 

[PJVol]

As far as I understood the chip CAN achieve >3GHz but if gaming this makes the power consumption explode.

Did you see what workload he used to measure the performance? Charts labeled as FP32.
And has anyone measure perfomance(fps) per clock compared to the N21?

upd:
Nvm, missed the posts below that (DeepL translate)

He: However, at least the power consumption and voltage of my 7900XT is not able to support it in this state for gaming, only for general computing work.

Dude: Maximum clocks to be able to use playing?

He: Approx. 3.2Ghz

Noticed also the "PPT limit sustained" was 330W and "PPT limit short" 400W
With a gaming frequency of 3.2 GHz, 400 watts peak and 330 sustained doesn't seem like an "explode" at all.

 

[Paul98]

Looks like it's what was expected, with some issues which was going to happen with a new design like this. Hopefully they can get it figured out and fix a lot of the issues in the drivers

 

[gdansk]

Is it possible that the high video playback power consumption is just a firmware/driver issue that can be fixed?

It's possible. But I wonder if it's MCD related.

 

[pj-]

Was hoping for this to be closer to the 4090, guess I'll be sitting out this gen. Thanks to nvidia and amd for making it easier to be slightly more fiscally responsible

 

[Stuka87]

N31 in this state is a FLOP only the price is at least good.
Power draw during Video and multimedia is embarrassingly bad.

RDNA2 had a similar issues at launch, but was fixed. It appears that RDNA3 also needs this fixed.

 

[eek2121]

As far as I understood the chip CAN achieve >3GHz but if gaming this makes the power consumption explode.
AMD Radeon RX 7900 XT 'Navi 31' GPU Overclocked To 3.7 GHz Front End Clock & 3.5 GHz Shader Clock (wccftech.com)
Wonder which are the hardware bugs leading to this and how will they fix them.
A part of the performance is surely lost on the software side, too. The move to a software scheduler is not without pain.

No benchmarks at those clocks except FP32? No comparison to the 4090? Note that the screenshot showed 400W, which isn't bad if the performance saw a 20+% uplift.

 

[eek2121]

Was hoping for this to be closer to the 4090, guess I'll be sitting out this gen. Thanks to nvidia and amd for making it easier to be slightly more fiscally responsible

I won't be buying the reference card, but will keep an eye out for partner cards. I need something the size of the reference card or smaller, and I want to see some reviews.

It is unfortunate that AMD dropped the ball with their reference cards, and it is also unfortunate that gaming power consumption is so high. The card performs as expected, however.

 

[Geddagod]

I won't be buying the reference card, but will keep an eye out for partner cards. I need something the size of the reference card or smaller, and I want to see some reviews.

It is unfortunate that AMD dropped the ball with their reference cards, and it is also unfortunate that gaming power consumption is so high. The card performs as expected, however.

The card doesn't perform as expected, not as expected as the leaks, not as expected as to the general discourse in the public, and certainly not as expected against AMD's own performance claims....

 

[SPBHM]

seems to follow the trend of super pricey and power hungry, with no midranged/low end priced alternative so far other than old cards (thankfully there are plenty of those now),
now... the failed 4080 12GB branded as 4070 ti, with the same price as the XT, can probably beat the XTX in RT?? hmmm.

if only AMD was a little more aggressive with pricing...

 

[Kepler_L2]

Is it only me, or others also feel as if this release is rushed, and AMD released an unfinished product?

No question about it.

 

[biostud]

If N32 doesn’t improve something, is there any reason to release it?

N31 has close to same price/performance to current 6800/6900 cards, so N32 has to be an improvement in this regard.

 

[gdansk]

On the plus side, for those of us in the USA: the reference board is made in Taiwan unlike the RTX 4090/4080 FE.
So shouldn't impacted by pending tariffs

 

[biostud]

If we get 60% better than 6900XT @ 45% better performance/watt, I don't think you will see many complaining.

Unfortunately it seems more the lines of 50% better @ 40% 4k vs 6900XT

 

[Kaluan]

Not great, not terrible.
At least better than those suspicious 3DMark "leaks" VideoCardz put out (and later double downed on, without a source... again)

Power usage for 7900XT is around reference 6900XT's and for 7900XTX slightly less than reference 3090. Closer to 4090/4080 efficiency than closer to last gen's best (RX 6800). Overall big efficiency gains, tho not as huge as I hoped. There seems to be some bugs with power usage in non-gaming workload tho. Hope they're driver level fixable.

IMO the architecture shows a lot promise, (look at the huge percentile lows advantage it has over 4080 and even 4090 in a lot of the reviews), too bad a lot if it got neutered by this need to rush it before they're properly adressed.

Hopefully AMD's eventual second go at it will properly flesh out RDNA3.

 

[Rekluse]

I'm curious. Since RDNA3 is broken, we can expect marked improvement next year with a RDNA3+ refresh.

But how much does nVidia have in the tank in comparison? Is the RTX 4090 a cut down chip? do they have room for an RTX 4090ti to compete with a RX 7970

 

[leoneazzurro]

No benchmarks at those clocks except FP32? No comparison to the 4090? Note that the screenshot showed 400W, which isn't bad if the performance saw a 20+% uplift.

Yup, but the problem it seems there is no such +20% in performance in actual overclocking. Best so far I've seen is a +5% @2700 MHz and some undervolting. Simply raising the power limit seems to do not much.

 

[Mopetar]

Even if AMD were to refresh Navi 31 with all of the issues fixed, it's not going to overtake a 4090. AD-102 is a big chip with a lot of cores.