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

 

[marees]

The PS5 Pro GPU is based on which RDNA Architecture?
Today a "new and improved PSSR" comes to PS5 Pro games and Sony says it's based on 6 month of refinement over FSR 4.

Upgraded PSSR upscaler is coming to PS5 Pro

Resident Evil Requiem, launching today, is the first title to use this more advanced PSSR.

blog.playstation.com

AMD is burning itself for no reason by gate keeping FSR4 for RDNA 4 only.

raster = RDNA 2
upscaling = RDNA 3
RT = RDNA 4

Technically PSSR like upscaling should be possible for all strix point / hawk point APUs

 

[Panino Manino]

RDNA2 (mostly).

Gives people a reason to buy RDNA4.

No, gives a reação for people to buy Nvidia when they see new versions of DLSS working on older models officially.

 

[adroc_thurston]

No, gives a reação for people to buy Nvidia when they see new versions of DLSS working on older models officially.

Good.
People should buy Nvidia.

 

[SolidQ]

AMD is burning itself for no reason by gate keeping FSR4 for RDNA 4 only.

Soon FSR 4.1

Hogwarts Legacy FSR test - Imgsli

Hogwarts Legacy FSR test - Imgsli

1080p UP - Imgsli

I assume PSSR2 = FSR 4.1, but int8 version

 

[MrMPFR]

Soon FSR 4.1

Hogwarts Legacy FSR test - Imgsli

Hogwarts Legacy FSR test - Imgsli

1080p UP - Imgsli

I assume PSSR2 = FSR 4.1, but int8 version

Very different architectures, but possible.

 

[coercitiv]

Good.
People should buy Nvidia.

Sigh, they got to him already.

Spoiler

 

[ToTTenTranz]

raster = RDNA 2
upscaling = RDNA 3
RT = RDNA 4

Technically PSSR like upscaling should be possible for all strix point / hawk point APUs

If by "upscaling" you mean tensor throughput to run PSSR then it's probably closer to RDNA4's as well. Early leaks point to 300 TOPs which would only fit into 60 CUs RDNA4 @ ~2.35GHz (289 TOPs FP8).

 

[SolidQ]

https://twitter.com/x/status/2027369641488592898

Stay tuned for next FSR update.

Ofc Sony have int8 PSSR2, but AMD seems not going update for RX 7xxx users

 

[adroc_thurston]

but AMD seems not going update for RX 7xxx users

Nope.
They gotta buy RDNA4 instead.

 

[Panino Manino]

Good.
People should buy Nvidia.

Or, surely people will buy Nvidia. Now if only Jensen wanted people to buy GeForce...

 

[MrMPFR]

Finally do any of you know how NVIDIA supports it? Aren't they known for basically deferring almost everything to the driver since Kepler?

Does this have anything to with CUDA? HyperQ, Dynamic Parallelism, Cooperative Groups, CUDA Graphs...

Very confused by this discrepancy between NVIDIA and AMD. AMD artificially locking out gfx10 Work Graphs support thus limiting to gfx11 (RDNA 3) and newer or Work Graphs still off the table for older stuff?

 

[adroc_thurston]

AMD artificially locking gfx10 Work Graphs support to gfx11 (RDNA 3) and newer or Work Graphs still off the table for older stuff?

RDNA2 support is EOL, yes.