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

 

[PJVol]

N32 apparently shown by AMD in B-Roll video, without naming it:

Hmm... look very similar ))
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...a2-architectures-thread.2589999/post-40997944

 

[GodisanAtheist]

Sometimes, the problem is that people can't tell the good news from bad news, or good price for bad price.

From 6600 XT at $379 to 7600 at $269, that's a 30% price drop, while the inflation was 10% last year.

40% improvement in price performance over 2 years is a pretty good news, IMO. But there were some children throwing temper tantrum, as if they got socks under the Christmas Tree instead of a live pony they expected.

-I asked for a dead pony and all I got were some socks

 

[igor_kavinski]

-I asked for a dead pony and all I got were some socks

Well, hey, you can store your bitcoins in the socks

 

[Ranulf]

40% improvement in price performance over 2 years is a pretty good news, IMO. But there were some children throwing temper tantrum, as if they got socks under the Christmas Tree instead of a live pony they expected.

I wanted a ray traced pony...without the fake frames.

 

[MrTeal]

The 6600 XT was a crap value released right at the height of the mining GPU shortage, and was priced with that. It was functionally the same performance at 1080p as the 5700XT that launched at $400 and was slower at 4k so saying it's better is damning it with faint praise. The 7600 is a bit faster than the 5700XT at 1080p and basically tied with it at 4k, so it's really a 33% price drop over 4 years.

 

[TESKATLIPOKA]

If you think AMD is going to give you a great price on this GPU for its level of performance then you haven't been paying attention. These dies are using N5 wafers, which means every one of these is roughly 3 Zen4 chiplets they didn't make instead.

Keep in mind this is the same company that up until the day before the release wanted to charge $300 for the 7600. Even with the last minute $30 price cut, it's still too expensive by at least $20, but probably more like $50.

I'm expecting them to align their pricing to whatever NVidia is charging for a rough equivalent. If this squares up well enough against the 4070 Ti then expect $700 at the cheapest.

Did I say that AMD will sell N32 for a great price?
What I wrote is that N32 could be one of the better options this gen, but price is the key factor.

N32 looks like a RTX 4070 competitor at best, there is no way It can go against RTX 4070Ti.
If AMD wants to sell It for $599 like RTX 4070, then that's pretty much DOA in my eyes.
I would be quite happy with $499, but It will probably cost $549.
Then there is still the question of what would be considered as a great price for It. Is It $499, $449 or only $399?

 

[insertcarehere]

Sometimes, the problem is that people can't tell the good news from bad news, or good price for bad price.

From 6600 XT at $379 to 7600 at $269, that's a 30% price drop, while the inflation was 10% last year.

40% improvement in price performance over 2 years is a pretty good news, IMO. But there were some children throwing temper tantrum, as if they got socks under the Christmas Tree instead of a live pony they expected.

The $800 RTX 4070ti out performs the $1200 RTX 3080ti launched roughly 2 years earlier but there sure aren't many posts defending Nvidia's pricing.

Funny how when AMD does it then it is justifiable.

 

[Aapje]

I fully expect N32 GPUs to be worse value than N21/N22 cards at first. Then at best we can hope for prices to drop to more attractive levels over time.

 

[eek2121]

Geez, the RX 7800 XT (non mobile) better be > 6900XT or what is the point.

Not to be that guy, but you should be comparing the 7800XT to the 6800XT. Barring major improvements in other areas such as RT, I strongly suspect large gains are becoming a thing of the past.

 

[Ajay]

Not to be that guy, but you should be comparing the 7800XT to the 6800XT. Barring major improvements in other areas such as RT, I strongly suspect large gains are becoming a thing of the past.

I don't think so, Nvidia did pretty well on performance gain this generation. Prices and RAM capacity are the problems for them, IMHO. Thinking that a current generation, 7800XT, should at least match a step up from the previous generation is pretty reasonable, IMHO. I really think the 7800XT should have about the same performance as the 6950XT from the last generation. AMD might be able to get $650 for that with 16GB of RAM. But, IDK, details are pretty sparse, so these are just musing on what would get my interest in buying a 7800XT. A straight up upgrade from my GTX 1070 to an RTX 4070 would have been a no brainer - but the 12GB of RAM really burns my butt. I tend to keep my cards for quite a while for budge reasons - so some future proofing is preferred on my part.

 

[MrTeal]

Not to be that guy, but you should be comparing the 7800XT to the 6800XT. Barring major improvements in other areas such as RT, I strongly suspect large gains are becoming a thing of the past.

The 6800 XT (and 3080 for that matter) are the absolute worst comparisons, at least if you're AMD or Nvidia. If it's tying the 6900 XT and $700 it's a lot nicer to say "Same performance as a 6900XT for only 70% of the price." rather than "5-10% faster than a 6800 XT 2 years later for only $50 more"

Be that guy, hopefully publications and users keep on calling out marketing bulletpoints meant to show the newest parts in the best light.

 

[igor_kavinski]

Barring major improvements in other areas such as RT, I strongly suspect large gains are becoming a thing of the past.

Stable diffusion performance is gonna leave the 6800 XT in the dust, if that has prevented someone from going to AMD before.

Stable Diffusion Benchmarks: 45 Nvidia, AMD, and Intel GPUs Compared

Which graphics card offers the fastest AI performance?

www.tomshardware.com

Should be somewhere in the 10K to 12K iterations per sec range.

 

[Heartbreaker]

Thinking that a current generation, 7800XT, should at least match a step up from the previous generation is pretty reasonable, IMHO. I really think the 7800XT should have about the same performance as the 6950XT from the last generation.

That would need to be another slight cut on the N31 then. 7900XT isn't that far ahead of 6950XT. I don't see N32 reaching that level.

 

[GodisanAtheist]

That would need to be another slight cut on the N31 then. 7900XT isn't that far ahead of 6950XT. I don't see N32 reaching that level.

- Yep, AMD literally released a 7800 Pro card that is N31 cut down to 70CUs.

AMD's stack is just a hot mess right now. One can only hope we'll one day get a behind the scenes answer on what the heck happened here.

 

[Heartbreaker]

- Yep, AMD literally released a 7800 Pro card that is N31 cut down to 70CUs.

AMD's stack is just a hot mess right now. One can only hope we'll one day get a behind the scenes answer on what the heck happened here.

Igor's lab tests it for gaming:

AMD Radeon RX 7800XT Review as Radeon Pro W7800 - The ideal gap filler between GeForce RTX 4070 and RTX 4070 Ti | igor´sLAB

I admit, it was tempting to test the current Radeon Pro W7800 as a gaming card, because AMD’s offerings between the already launched RX 7900 XT(X) and the RX 7600 are conspicuous by their noble…

www.igorslab.de

 

[jpiniero]

Igor's lab tests it for gaming:

AMD Radeon RX 7800XT Review as Radeon Pro W7800 - The ideal gap filler between GeForce RTX 4070 and RTX 4070 Ti | igor´sLAB

I admit, it was tempting to test the current Radeon Pro W7800 as a gaming card, because AMD’s offerings between the already launched RX 7900 XT(X) and the RX 7600 are conspicuous by their noble…

www.igorslab.de

7800 XT (if they do it), wouldn't be the W7800 cut. It would be N32, so 56 or 60 CUs.

The W7800 only really exists because N32 doesn't ATM.

 

[GodisanAtheist]

Igor's lab tests it for gaming:

AMD Radeon RX 7800XT Review as Radeon Pro W7800 - The ideal gap filler between GeForce RTX 4070 and RTX 4070 Ti | igor´sLAB

I admit, it was tempting to test the current Radeon Pro W7800 as a gaming card, because AMD’s offerings between the already launched RX 7900 XT(X) and the RX 7600 are conspicuous by their noble…

www.igorslab.de

-Very nice. Basically the gaming oriented 7800 would likely be anywhere between 5-12% faster than the 6800Xt from 1080-4K resolution.

Based on those numbers I'm still not seeing a universe where the N32 can really provide a functional N21 replacement.

 

[Heartbreaker]

7800 XT (if they do it), wouldn't be the W7800 cut. It would be N32, so 56 or 60 CUs.

The W7800 only really exists because N32 doesn't ATM.

It seems, at best, N32 could be used for a 7800 (non XT) card. Which is a step up on where N22 ended.

 

[Thunder 57]

Stable diffusion performance is gonna leave the 6800 XT in the dust, if that has prevented someone from going to AMD before.

Stable Diffusion Benchmarks: 45 Nvidia, AMD, and Intel GPUs Compared

Which graphics card offers the fastest AI performance?

www.tomshardware.com

View attachment 81718

Should be somewhere in the 10K to 12K iterations per sec range.

I think we need more recent data, as AMD's latest driver claims they boosted performance by an average of 2x.

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-23-5-2

• Performance optimizations for Microsoft Olive DirectML pipeline for Stable Diffusion 1.5 on AMD Radeon RX 7900 series graphics
  • Boost your performance by an average of 2x in Microsoft Olive Optimized DirectML Stable Diffusion 1.5 using AMD Software: Adrenalin Edition™ 23.5.2 on the AMD Radeon™️ RX 7900 XTX graphics card, versus the previous software driver version 23.5.1. RS-579

It is interesting to see the 7000 series doing much better here than one might expect compared to RDNA2.

 

[igor_kavinski]

I think we need more recent data, as AMD's latest driver claims they boosted performance by an average of 2x.

Really hope so but TomH's RX 7000 results are based on SD 2.1 whereas AMD driver optimizations are for SD 1.5. Still, let's be cautiously optimistic

 

[Mopetar]

I think we need more recent data, as AMD's latest driver claims they boosted performance by an average of 2x.

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-23-5-2

It is interesting to see the 7000 series doing much better here than one might expect compared to RDNA2.

Well it has double the FP32 resources (2x per shader) compared to RDNA2, so the theoretical performance was always there, but they didn't have the software in place for a lot of applications to actually use any of it. Good to see that the magic drivers are actually materializing, unlike Vega.

 

[Ajay]

Well it has double the FP32 resources (2x per shader) compared to RDNA2, so the theoretical performance was always there, but they didn't have the software in place for a lot of applications to actually use any of it. Good to see that the magic drivers are actually materializing, unlike Vega.

Still, only on Stable Diffusion. Not much help for gamers.

 

[Timorous]

-Very nice. Basically the gaming oriented 7800 would likely be anywhere between 5-12% faster than the 6800Xt from 1080-4K resolution.

Based on those numbers I'm still not seeing a universe where the N32 can really provide a functional N21 replacement.

N32 would need to clock far higher at a given voltage.

We know N31 can manage pretty high clocks when given enough juice so the architecture can handle high clocks. Just needs to manage it without blowing the power budget.

 

[GodisanAtheist]

N32 would need to clock far higher at a given voltage.

We know N31 can manage pretty high clocks when given enough juice so the architecture can handle high clocks. Just needs to manage it without blowing the power budget.

-Therein lies the problem. I don't think AMD is getting 60 CUs to 3Ghz without blowing the power budget.

N31 will do it at 450W, thing is N32 needs to do it at 200-250W and I don't think that's happening. It will likely need 300W or more and that's just not acceptable for second tier silicon, expecially when Nvidia will be offering that performance for 200W.

 

[Mopetar]

Still, only on Stable Diffusion. Not much help for gamers.

No, but at least they're getting it working somewhere. Hopefully they get it figured out before RDNA4 and don't decide to not enable whatever improvements they can get from it for RDNA3 cards.

AMD has a pretty good track record for things like this and RDNA3 sales being on the low end wouldn't hurt future generation sales prospects. On the other hand they certainly haven't minded acting like NVidia on the pricing front.

Edit: Fixed RDNA version being incorrect.