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

 

[Glo.]

Why I said It's impossible was because N24 is a total lowend and neither for AMD or Its partners It was worth It to use 2x more memory chips in clamshell mode, but physically It is possible, true.

PULSE AMD Radeon™ RX 6500 XT 8GB

<p>Up to 2855 MHz, 8GB/64 bit DDR6. 18 Gbps Effective</p>

www.sapphiretech.com

Sapphire begs to differ.

 

[RnR_au]

That's a bit late for most of the world, isn't it?
Around midnight European time.
Shouldn't these things be early in the morning, to have it at a good time in most timezones?

Its 7 am in Australia (East coast)... not a bad way to start the day... a little saucy goodness from Lisa Su

Jumped on conclusion: they have nothing good to show and are trying to hide it from most of the world!

Nah... Its 4 pm since then they can start the party early

 

[TESKATLIPOKA]

PULSE AMD Radeon™ RX 6500 XT 8GB

<p>Up to 2855 MHz, 8GB/64 bit DDR6. 18 Gbps Effective</p>

www.sapphiretech.com

Sapphire begs to differ.

Wow, didn't know about that, thanks for posting.
I just checked and It can be bought in my country for 294,90€, which is 53€ more than the 4GB Sapphire pulse version. Sadly, I couldn't find a review for It to see If It's worth the extra cost or not.

Edit: Sapphire Pulse RX 6600 8GB costs only 324,38 €, so 6500XT is definitely not worth It. Doesn't matter If It's 4 or 8GB version.

 

[Saylick]

Not really new news if you've been following this thread, but more evidence that the 7900 series is all N31 with the 7900XT being the cut-down N31 w/ 20GB VRAM and the 7950XT being the full N31 w/ 24 GB VRAM.

Still no confirmation on performance targets.

AMD Radeon RX 7900XT rumored to feature 20GB GDDR6 memory - VideoCardz.com

AMD RX 7900XT rumored to feature 4GB more memory than RX 6900XT According to Wccftech sources, Radeon RX 7900XT RDNA3 GPU is said to feature 20GB of VRAM. This is reportedly not the flagship model, though. AMD Radeon RX 7900 series are to be announced in less than 2 weeks, which means that...

videocardz.com

 

[eek2121]

https://twitter.com/x/status/1583485036179095556

Another cryptic tweet by greymon. 6950xt performance in laptop power profile maybe?

 

[RnR_au]

https://twitter.com/x/status/1583485036179095556

Another cryptic tweet by greymon. 6950xt performance in laptop power profile maybe?

Isn't N33 supposed to be 6900XT like in performance at 1080p? And N33 is supposed to be the mobile part.

So I'm guessing is that they are hinting that the performance of N33 is somewhat faster than a 6900XT.

 

[jpiniero]

Isn't N33 supposed to be 6900XT like in performance at 1080p? And N33 is supposed to be the mobile part.

So I'm guessing is that they are hinting that the performance of N33 is somewhat faster than a 6900XT.

Mobile N31 variant perhaps?

 

[TESKATLIPOKA]

Mobile N31 variant perhaps?

I don't think we will see N31 in laptops.
N32 is the safe bet, but maybe N33 will surprise us with Its performance at 1080p.
If he meant with RT then It is N33, because RDNA2 is so bad in It.

 

[maddie]

Mobile N31 variant perhaps?

Is there a 6950XT mobile variant?

 

[jpiniero]

Is there a 6950XT mobile variant?

No. The top is N22 parts.

 

[Mopetar]

Is there a 6950XT mobile variant?

No, and there weren't any Navi 21 based mobile parts. Regardless of what AMD elects to call their mobile lineup, it seems unlikely that we'll see a Navi 31 part. Anyone who actually needs that kind of horsepower in a mobile workstation likely needs a CDNA card with the professional grade verified drivers that come with it.

I wouldn't be surprised if you could lower the clocks a bit to get Navi 31 to fit in a notebook's thermal limits, but it's a nice part and would be expensive. The 6850M XT is listed as a 165W TDP part. Why bother putting Navi 31 in a notebook when you have to cut the power by half or more when you have Navi 32 that will be just as capable in most situations and is a better fit for the thermal budget.

 

[Glo.]

I don't think we will see N31 in laptops.
N32 is the safe bet, but maybe N33 will surprise us with Its performance at 1080p.
If he meant with RT then It is N33, because RDNA2 is so bad in It.

The majority is N33. I don't know about N32, but based on the rumored scaling: 10754 RDNA3 ALUs are needed to tie, or beat RTX 4090 on desktop, then I'd assume that indeed N32 in laptop form is around 6950XT desktop performance.

Not bad.

P.S. Doesn't it make it 100% 4k@120 FPS capable...?

 

[TESKATLIPOKA]

https://twitter.com/x/status/1583485036179095556

Another cryptic tweet by greymon. 6950xt performance in laptop power profile maybe?

He also said this:

reference's TBP is amazing.

https://twitter.com/x/status/1582599964567314432

 

[TESKATLIPOKA]

The majority is N33. I don't know about N32, but based on the rumored scaling: 10754 RDNA3 ALUs are needed to tie, or beat RTX 4090 on desktop, then I'd assume that indeed N32 in laptop form is around 6950XT desktop performance.

Not bad.

P.S. Doesn't it make it 100% 4k@120 FPS capable...?

If It's the full N32 at 150W then It's not so great considering He also claimed N33 was close to RX 6900XT, even If only at 1080p and desktop N33 supposedly consumes <170W.

 

[DisEnchantment]

Still no confirmation on performance targets.

Well Paul from RGT's has been confirming on a regular basis all the performance targets.
Not his fault that AMD is changing specs every day till day before launch, but otherwise good ole Paul has all the details.

 

[TESKATLIPOKA]

Bondrewd just confirmed that It's N32.

 

[Timorous]

Bondrewd just confirmed that It's N32.

He also said 150W +/- so that is more than 2x the perf/watt if true. 150W makes me think low clocked cut down version as well.

 

[biostud]

I can’t find the tweet, but one of the leakers had a link up to an RV770 article from AT 2008.
I read it as RDNA3 is going to be similar

 

[GodisanAtheist]

IF AMD is dropping super efficient / low power cards on us then it means that the performance has to be competitive with what NV is posting up.

Otherwise we'd be hearing "falling short of their efficiency gain/power targets" as AMD crushes the curve to squeeze out every last drop of performance and close the gap.

 

[maddie]

He also said this:

https://twitter.com/x/status/1582599964567314432

reference's TBP is amazing.

Rearchitectured CU is a key advantage. I/2 the size says ~ 1/2 the transistors needed. How? Not a clue. Switching power follows.

 

[TESKATLIPOKA]

He also said 150W +/- so that is more than 2x the perf/watt if true. 150W makes me think low clocked cut down version as well.

We still compare a Laptop model vs Desktop model.
For example RX 6800M has ~45-50% better perf/W compared to desktop model.

	Gaming Frequency	TBP	Specification
RX 6700 XT	2424 MHz	230 W	40CU; 192bit 16gbps
RX 6800M	2300 MHz	145+ W	40CU; 192bit 16gbps

6700XT is only 5% faster in Cyberpunk 2077 than 6800M and RX 6900XT is 76% faster.
So this N32 will provide ~80% higher performance at ISO power(150W) than N22. If It's actually a cutdown model that would be even better, but I wouldn't bet on It.

 

[TESKATLIPOKA]

reference's TBP is amazing.

Rearchitectured CU is a key advantage. I/2 the size says ~ 1/2 the transistors needed. How? Not a clue. Switching power follows.

There is no longer a CU just WGP as far as I read and that is the same size at the same process.

 

[Timorous]

We still compare a Laptop model vs Desktop model.
For example RX 6800M has ~45-50% better perf/W compared to desktop model.

	Gaming Frequency	TBP	Specification
RX 6700 XT	2424 MHz	230 W	40CU; 192bit 16gbps
RX 6800M	2300 MHz	145+ W	40CU; 192bit 16gbps

View attachment 69623
6700XT is only 5% faster in Cyberpunk 2077 than 6800M and RX 6900XT is 76% faster.
So this N32 will provide ~80% higher performance at ISO power(150W) than N22. If It's actually a cutdown model that would be even better, but I wouldn't bet on It.

The 6700XT is a bit of an outlier because AMD just pushed the desktop part to its limits.

Since 6800M was a lower clocked 6700XT it is not far fetched to think 7800M will be a lower clocked 7700XT.

 

[TESKATLIPOKA]

The 6700XT is a bit of an outlier because AMD just pushed the desktop part to its limits.

Since 6800M was a lower clocked 6700XT it is not far fetched to think 7800M will be a lower clocked 7700XT.

Difference in gaming frequency is only 5% so 6800M is also pushed very high.
This just shows how much better a binned chip can be.

Now let's go back to greymon55's leak.
I have to wonder how high is this N32 actually clocked, because 80% higher performance at ISO power is not so much If I think about N32's specs.
N32 has 50% more WGPs than N22(30 vs 20) and each WGP should be significantly better.

Expected performance per ISO power(~150W)
RX 6800M : 100%
N32 mobile : +80%

N32 has +50% WGPs and If we say RDNA 3 WGP is 50% better than RDNA2 WGP then ideally that's 100*1.5*1.5=225% or +125% higher performance, but It's actually only +80%. I have to lower clockspeed by 20% from 2300MHz to 1840MHz to have the expected performance.

This leak from Greymom55 could actually be a cutdown version of N32, because I don't expect a mobile N32 to be downclocked from >3GHz down to <2GHz just to be within 145-165W, when N22 was downclocked by only a few %.

 

[Kepler_L2]

There is no longer a CU just WGP as far as I read and that is the same size at the same process.

CUs still exist and they are smaller than RDNA2 CUs @ 7nm, though N33 CUs @ 6nm and N31/32 CUs @ 5nm are slightly different.