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

 

[jpiniero]

AMD Slims Down Compute With Radeon Pro W7900 Dual Slot For AI Inference

www.anandtech.com

Turns out that W7900 AI Edition is just the same product as the W7900 except it is now dual slot instead of three. Oh and the cut the MSRP by $500 too.

 

[Mahboi]

New AMD RDNA 3.5 GPU Target "GFX1152" Appears In Open-Source Linux Driver - Phoronix

www.phoronix.com

MMMMMMMMMMMH.
New handheld somewhere?

 

[jpiniero]

New AMD RDNA 3.5 GPU Target "GFX1152" Appears In Open-Source Linux Driver - Phoronix

www.phoronix.com

MMMMMMMMMMMH.
New handheld somewhere?

Strix point refresh perhaps?

 

[Mahboi]

A refresh before a single laptop has gone to market?
I understand low expectations of Radeon but come on 🤣

 

[igor_kavinski]

A refresh before a single laptop has gone to market?

It would be more like a Ryzen Z2 Extreme with Zen5c, no?

 

[Mahboi]

Completely possible, although it's early, I mean, it's effectively the same gen as RDNA 3.
But I would absolutely expect some lowest tier STX as a handheld's chip. If possible as 2 Z5/4 Z5c with 12 RDNA 3.5 CUs, that would make a solid thing.

 

[Kepler_L2]

New AMD RDNA 3.5 GPU Target "GFX1152" Appears In Open-Source Linux Driver - Phoronix

www.phoronix.com

MMMMMMMMMMMH.
New handheld somewhere?

Kraken Point

 

[marees]

Completely possible, although it's early, I mean, it's effectively the same gen as RDNA 3.
But I would absolutely expect some lowest tier STX as a handheld's chip. If possible as 2 Z5/4 Z5c with 12 RDNA 3.5 CUs, that would make a solid thing.

Seems like it has the correct cpu for a handheld (equivalent zen 4 z1 non-extreme ). However GPU (& memory) need to be more for real gaming.

AMD has to make these as general purpose APUs that work for mobile/laptops also

 

[SteinFG]

Z1E is already 8 cores, they won't lower it.
4P+4C+16CU is the most likely confuguration for Z2E.
2P+4C+8CU (kraken chip, but some cores disabled to match Z1) is most likely configuration for Z2.

 

[Mahboi]

AMD Radeon 890M RDNA3.5 graphics are 39% faster than Radeon 780M in OpenCL test - VideoCardz.com

AMD Ryzen AI HX 170/370 in first leaked benchmarks AMD was rather reluctant to share details on the RDNA3.5 during the official keynote. What we know is that it uses an updated architecture and features more cores than last-gen products. Both can contribute to higher performance. Over at...

videocardz.com

It's looking very good.
Not "magical" obviously since supposedly RDNA 3 worked well in compute. But if we get 33% more CUs and an extra 6% from what I'm going to name "random improvements"(probably just the extra 200Mhz peak tbh), it should be quite stellar in games, we can expect a broad 40-45% improvement.
Git out there and into lappies.

 

[igor_kavinski]

it should be quite stellar in games, we can expect a broad 40-45% improvement.

If it's 20% improvement in raster and 20% in RT over 780M, it's still good enough progress for me. We shouldn't set AMD up for disappointment by expecting too much

In fact, let's establish a ground rule for AMD hype trains going forward. No more than 10% improvement no matter what any leaker or even AMD themselves say. Then we'll hopefully never be disappointed.

 

[Mahboi]

In fact, let's establish a ground rule for AMD hype trains going forward.

The vid might be deeper than I thought cause there's a lot of details people didn't pay attention to...

NOTHING STOPS THE TRAINS​

If it's 20% improvement in raster and 20% in RT over 780M, it's still good enough progress for me. We shouldn't set AMD up for disappointment by expecting too much

Not quite, it's 200Mhz more clocks and 33% more CUs, and should theoretically hold up to its boost clocks quite a bit better. 40-45% is no stretch for the 16CU top end.
If you mean WGP for WGP, oh yes, I'm only expecting about 10-12% overall better, nothing exceptional.

 

[jpiniero]

AMD announces Radeon RX 7800XT & 7700XT game bundle - VideoCardz.com

Finally new game bundle: AMD offers two games with Navi 32 cards Starting from today, gamers can pick two games out of for with select Radeon GPUs. The new promotion announced today is only for two graphics cards: Radeon RX 7800 XT and Radeon RX 7700 XT, both equipped with Navi 32 GPU. No other […]

videocardz.com

Looks like AMD is trying to get rid of N32... not so much N31 or N33 though.

 

[SteinFG]

Obviously, N48 replaces N32. and N44 replaces N33, but they can just lower the price of RX 7600 to RX 6600 level ($200)

 

[marees]

Obviously, N48 replaces N32. and N44 replaces N33, but they can just lower the price of RX 7600 to RX 6600 level ($200)

In terms of performance I expect n33 < n44 < n32 < n48 < n31

They can lower prices of all RDNA 3 cards (except 7900 gre, which has to go away) appropriately based on demand & stocks

I also expect 7600 to remain for a long time because it is on 6nm

 

[SteinFG]

They can lower prices of all RDNA 3 cards (except 7900 gre, which has to go away) appropriately based on demand & stocks

N48 seems to be just cheaper and faster than N32, lowering prices on those is kinda eh, better sell them off before all that.

 

[SolidQ]

Seems 3D mark heavily optimized for Intel, compared to real gaming

780m slower than ARC8 by 8%, but in reality difference story. Also think lunar lake IGPU, not going compete with 890m

 

[ToTTenTranz]

In terms of performance I expect n33 < n44 < n32 < n48 < n31

For rasterization, yes.

For heavy RT I'm expecting N31<N48.

 

[Tuna-Fish]

Seems 3D mark heavily optimized for Intel, compared to real gaming

It's rather that the Intel drivers are heavily optimized for 3D Mark and not so for gaming, but yes.

 

[Mopetar]

It's rather that the Intel drivers are heavily optimized for 3D Mark and not so for gaming, but yes.

I think it's probably more fair to say that their game drivers aren't nearly as well optimized as either AMD/NV.

If 3D Mark is an indicator of what they can do with a lot of work, it does speak well for the architecture.

I suppose it's possible to have the drivers cheat in ways that would not be possible in an actual game to give a better picture of what's possible than could normally be achieved, but both AMD/NV could do the same for this benchmark.

 

[SolidQ]

N36?

Seasonic Lists Down Several Unreleased AMD Radeon RX 7000 GPUs: 7990 XTX, 7950 XTX, 7950 XT

PSU manufacturer, Seasonic, has listed several unreleased AMD Radeon RX 7000 GPUs such as the 7990 XTX, 7950 XTX & 7950 XT.

wccftech.com

 

[jpiniero]

N36?

Seasonic Lists Down Several Unreleased AMD Radeon RX 7000 GPUs: 7990 XTX, 7950 XTX, 7950 XT

PSU manufacturer, Seasonic, has listed several unreleased AMD Radeon RX 7000 GPUs such as the 7990 XTX, 7950 XTX & 7950 XT.

wccftech.com

I think if they were going to do that, it would have been released by now.

 

[igor_kavinski]

I think if they were going to do that, it would have been released by now.

Possibly launching with 9800X3D.

 

[ToTTenTranz]

N36 was that 60 WGP + 384bit GDDR6 part, right?

Could AMD have decided to revive N36 after cancelling all the high-end RDNA4?

 

[soresu]

RDNA 3.5 slides from AMD Tech Day: