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

 

[Panino Manino]

RDNA 3.5 slides from AMD Tech Day:

Is the IODC on the Ryzen 9000 new?
The 2CU are RDNA3 or RDNA3.5?

 

[blckgrffn]

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

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

If they actually stepped all those big boy parts to RDNA 3.5 and re-released them now, wouldn't that be a good thing? They'd be clearly above the RDNA4 parts we are going to get in terms of pure performance and would carry until RDNA5.

And if you are AMD and are looking at the least work way to fill a full lineup while you pull in as much AI megabucks as possible, it seems pretty reasonable given that its all baked for mobile too.

I know that's 99.99% not what's happening but it doesn't seem like it would be the worst decision in the world for us as consumers. More choice would be better.

 

[soresu]

Is the IODC on the Ryzen 9000 new?
The 2CU are RDNA3 or RDNA3.5?

No it's still RDNA2 apparently.

Surely Zen6 should get an upgrade there.

 

[ToTTenTranz]

If they actually stepped all those big boy parts to RDNA 3.5 and re-released them now, wouldn't that be a good thing? They'd be clearly above the RDNA4 parts we are going to get in terms of pure performance and would carry until RDNA5.

I think if this 7990XTX card is actually releasing, it's more likely to be a revised N31 with a revised back-end that doesn't explode in power consumption above 2GHz, or just the N36 at lower clocks (now that N5 is cheaper), than it is of being an unexpected RDNA3.5 upgrade.

No it's still RDNA2 apparently.

Surely Zen6 should get an upgrade there.

It's mostly a matter of having that architecture ready for the process node at hand.

My guess is they'll only change iGPU architecture when/if they upgrade the process node for the IOD.
There's no RDNA3 for N7/N6, so as long as the IOD is on N6 they're not upgrading from RDNA2.

And to be honest, there's little reason to do so. What practical difference would it make to go from 2CUs RDNA2 to 2CUs RDNA3?

 

[soresu]

There's no RDNA3 for N7/N6

N33?

 

[soresu]

And to be honest, there's little reason to do so. What practical difference would it make to go from 2CUs RDNA2 to 2CUs RDNA3?

My first and only thought would be updating the VCN, as RDNA2's VCN lacks AV1 encoding.

 

[inquiss]

I think if this 7990XTX card is actually releasing, it's more likely to be a revised N31 with a revised back-end that doesn't explode in power consumption above 2GHz, or just the N36 at lower clocks (now that N5 is cheaper), than it is of being an unexpected RDNA3.5 upgrade.




It's mostly a matter of having that architecture ready for the process node at hand.

My guess is they'll only change iGPU architecture when/if they upgrade the process node for the IOD.
There's no RDNA3 for N7/N6, so as long as the IOD is on N6 they're not upgrading from RDNA2.

And to be honest, there's little reason to do so. What practical difference would it make to go from 2CUs RDNA2 to 2CUs RDNA3?

Future heavy APUs are RDNA5

 

[DaaQ]

I think if this 7990XTX card is actually releasing, it's more likely to be a revised N31 with a revised back-end that doesn't explode in power consumption above 2GHz, or just the N36 at lower clocks (now that N5 is cheaper), than it is of being an unexpected RDNA3.5 upgrade.




It's mostly a matter of having that architecture ready for the process node at hand.

My guess is they'll only change iGPU architecture when/if they upgrade the process node for the IOD.
There's no RDNA3 for N7/N6, so as long as the IOD is on N6 they're not upgrading from RDNA2.

And to be honest, there's little reason to do so. What practical difference would it make to go from 2CUs RDNA2 to 2CUs RDNA3?

Doesn't explode at 2Ghz? what about the Blackwell 500w TDP increase over 4090?

BTW power on N31 doesn't explode over 2 Ghz.

Waiting for launch day bar graphs from TPU.

 

[ToTTenTranz]

N33?

Oops you're right, N33 is N6.
Regardless, do we know if the IOD changed between Zen4 and Zen5?

My first and only thought would be updating the VCN, as RDNA2's VCN lacks AV1 encoding.

That's true, though I don't know how many people use AMD CPUsand want to do AV1 encoding and lack a graphics card capable of doing so.
Those who do seek that capability can still get the Ryzen 8x00G solution, though.

BTW power on N31 doesn't explode over 2 Ghz.

It was an exaggeration, but AMD was definitely counting on being able to boost the GPU above 3GHz in gaming workloads and while the shader processors do seem to have that capability, the backend does not.

 

[soresu]

Regardless, do we know if the IOD changed between Zen4 and Zen5?

AFAIK no, and it seemed that was telegraphed months before Zen5 was announced.

 

[marees]

AFAIK no, and it seemed that was telegraphed months before Zen5 was announced.

Makes sense for iod to remain at 6nm given that 6nm/7nm supply is free at TSMC whereas 5nm/4nm/3nm demand & price keeps on increasing

 

[DaaQ]

It was an exaggeration, but AMD was definitely counting on being able to boost the GPU above 3GHz in gaming workloads and while the shader processors do seem to have that capability, the backend does not.

My sarcasm meter is faulty atm sorry, IRL stuff.

Is there a sensor on HWinfo64 for the backend? If so what is its, label?

 

[marees]

RDNA 3.5 slides from AMD Tech Day:

Rdna 3.5 is future of handheld igpu gaming
Incorporates lessons from samsung's failed mobile experiment

So what we have is both cpu & gpu (rdna 3.5) is tuned for low power

What you are missing is RT goodness of RDNA 4. Dissapointing if you wanted to enable nvidia crippleware on a handheld or games made on nv ue5 such as bmw

 

[marees]

Oops you're right, N33 is N6.
Regardless, do we know if the IOD changed between Zen4 and Zen5?




That's true, though I don't know how many people use AMD CPUsand want to do AV1 encoding and lack a graphics card capable of doing so.
Those who do seek that capability can still get the Ryzen 8x00G solution, though.



It was an exaggeration, but AMD was definitely counting on being able to boost the GPU above 3GHz in gaming workloads and while the shader processors do seem to have that capability, the backend does not.

Afaik both front & back of rdna3 can independently hit 3.75 ghz

In real gaming scenarios (data transfer) the freq is 1ghz lower

Monolithic rdna 4 (navi 48) should fix power draw issue

Mlid gave a power target of 7700xt for top rdna 4 card (7900xt raster, 4070 ti super RT, 4070 super price)

 

[marees]

Afaik both front & back of rdna3 can independently hit 3.75 ghz

In real gaming scenarios (data transfer) the freq is 1ghz lower

Monolithic rdna 4 (navi 48) should fix power draw issue

Mlid gave a power target of 7700xt for top rdna 4 card (7900xt raster, 4070 ti super RT, 4070 super price)

It is my speculation but what is broken in rdna 3 remains broken in rdna 4 (which is why AMD redesigned chiplet based navi 43 as Monolithic navi 48)

The rdna 5 successor of this chip (6700xt of RDNA 5) on 3nm should match or better navi 31. Which is why we don't seem to get a 7950xt or 7950xtx

 

[jpiniero]

https://videocardz.com/newz/amd-to-launch-radeon-rx-7650-gre-at-ces-2025

Speculation that AMD is going to launch the 7650 GRE at CES. I wonder if this means that AMD is pushing RDNA4's launch past CES to give more time for higher end RDNA3 to sell out.

 

[beginner99]

I'm not a business specialist but if they have a working chip in hand, isn't it nonsensical to wait this long like till the competition can easily counter you? Either AMD has an internal resources issue or something went pretty wrong, again. The RDNA3 stock makes little sense, if that would really be the case, why not make a sale?

 

[marees]

The RDNA3 stock makes little sense

It makes no sense. AMD knew the shitshow with RDNA 2 inventories. Navi 32 launch was delayed by atleast 6 months. Then why overproduce RDNA 3 chips ???

 

[SolidQ]

Speculation that AMD is going to launch the 7650 GRE at CES

Seems for china, for rest is RDNA4

 

[coercitiv]

I'm not a business specialist but if they have a working chip in hand, isn't it nonsensical to wait this long like till the competition can easily counter you? Either AMD has an internal resources issue or something went pretty wrong, again. The RDNA3 stock makes little sense, if that would really be the case, why not make a sale?

It only makes sense if the competition is behind, such as Nvidia delaying 4000 series to allow 3000 inventory normalization while AMD could not do much about it anyway.

 

[SteinFG]

I'm not a business specialist but if they have a working chip in hand, isn't it nonsensical to wait this long like till the competition can easily counter you? Either AMD has an internal resources issue or something went pretty wrong, again. The RDNA3 stock makes little sense, if that would really be the case, why not make a sale?

Only 5090/5080 launch in January, and those are not competition.

 

[jpiniero]

Only 5090/5080 launch in January, and those are not competition.

Videocardz claimed today that the 5070 will launch at CES as well, although the release would be much later. Top Navi 48 model could be very competitive with it in raster at least... and it might work out for AMD if they can manage to get people to compare the price to the 5070 rather than RDNA3 firesales.

 

[GodisanAtheist]

It makes no sense. AMD knew the shitshow with RDNA 2 inventories. Navi 32 launch was delayed by atleast 6 months. Then why overproduce RDNA 3 chips ???

-Who says overproduced?

Maybe they undersold even AMD's tepid sales expectations...

 

[SolidQ]

Top Navi 48 model could be very competitive with it in raster at least.

i think if rumor true around 4070ti super RT, than gonna be same with 5070, except path tracing
That from June TPU. I think N48 can be easy 4080 with those double RT engine

 

[linkgoron]

https://videocardz.com/newz/amd-to-launch-radeon-rx-7650-gre-at-ces-2025

Speculation that AMD is going to launch the 7650 GRE at CES. I wonder if this means that AMD is pushing RDNA4's launch past CES to give more time for higher end RDNA3 to sell out.

They should just cancel RDNA4 and wait for RDNA5 maybe by then the old inventory will clear.

If I had to bet, they're just waiting for Nvidia to see if they can bump the price to maximize their pricing. I'm extremely skeptical that RDNA4 is not done or in a finalized state yet.