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

 

[eek2121]

Fantastic article. This is why I love Chips and Cheese.

RDNA3 is not a bad architecture, but that dual issue is simply useless If It's left to the compiler.
Games will need heavy optimizations done by AMD's driver team to make this architecture "shine". This doesn't look optimistic.
I really have to wonder why AMD choose this path, when they knew how much work It needs to work correctly.

If the code is optimized correctly for VOPD instructions, then the improvement in TFLOPs is ~100%.
View attachment 74208


Naturally, this whole optimization is by no means an easy task, but It should result in significant improvements in games in my opinion.

This is an area where using AI to optimize could be a big help. Train a model to recognize a frame, then implement various tweaks and see if output = input while fps > old fps.

By default shaders are compiled in the driver while the assets are loading. Sometimes they are compiled on the fly (which is usually bad and causes stuttering), the compiled shaders are also often cached locally to speed up future loads.

By default, sure, but precompiled, distributed shader assets are a thing.

The amazing things AMD could do if they focused more on software. Imagine if they created a Radeon specific shader modification/replacement toolkit? They include a bunch of highly optimized shaders out of the box and let the community do the rest. They could distribute it via steam and use Steam workshop for user content.

God forbid they think outside the box.

Side note: Maybe the x50 refresh will be RDNA3+?

 

[Kepler_L2]

This is an area where using AI to optimize could be a big help. Train a model to recognize a frame, then implement various tweaks and see if output = input while fps > old fps.


By default, sure, but precompiled, distributed shader assets are a thing.

The amazing things AMD could do if they focused more on software. Imagine if they created a Radeon specific shader modification/replacement toolkit? They include a bunch of highly optimized shaders out of the box and let the community do the rest. They could distribute it via steam and use Steam workshop for user content.

God forbid they think outside the box.

Side note: Maybe the x50 refresh will be RDNA3+?

RDNA3+ is just for APUs.

 

[TESKATLIPOKA]

RDNA3+ is just for APUs.

It would be great If IGP was beefed up.
So next desktop stop is RDNA4. Hope they don't make the same mistakes as this gen.

When do we move to the next thread to start speculating about RDNA4? Will we wait for N32 or not?

 

[Heartbreaker]

RDNA3+ is just for APUs.

Where do you see any info about RDNA3+?

 

[PJVol]

Where do you see any info about RDNA3+?

GFX 1101 ?

 

[Kepler_L2]

Where do you see any info about RDNA3+?

Some gfx1150/gfx12 stuff leaked a while ago

 

[Kaluan]

GFX 1101 ?

GFX1150 for Strix Point's "RDNA3+", only 115x family code known (so far).

https://twitter.com/i/web/status/1535032958478069760

Edit: GFX1101 is N31 AFAIK

 

[PJVol]

Edit: GFX1101 is N31 AFAIK

Where did you get that? N31 is 1100, N32 - 1102, N33 - 1103.
Anyway, 1101 and 1104 are APUs:
https://elixir.bootlin.com/linux/v6.2-rc3/source/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c#L2213

switch (adev->ip_versions[GC_HWIP][0]) {
    case IP_VERSION(9, 1, 0):
         ...
    case IP_VERSION(10, 3, 7):
    case IP_VERSION(11, 0, 1):
    case IP_VERSION(11, 0, 4):
        adev->flags |= AMD_IS_APU;
        break;
    default:
        break;
}

 

[TESKATLIPOKA]

If RDNA3+ is only for APUs, then a good question is why?
Because It's not financially or based on performance worth It to make new GPUs based on RDNA3+ or RDNA4 would be too close to It to make any sense.

 

[moinmoin]

iGPUs a pretty distinct from dGPUs anyway, so I guess we could consider the iGPU in Phoenix as RDNA3- in the same way.

 

[Kaluan]

Where did you get that? N31 is 1100, N32 - 1102, N33 - 1103.
Anyway, 1101 and 1104 are APUs:
https://elixir.bootlin.com/linux/v6.2-rc3/source/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c#L2213

switch (adev->ip_versions[GC_HWIP][0]) {
    case IP_VERSION(9, 1, 0):
         ...
    case IP_VERSION(10, 3, 7):
    case IP_VERSION(11, 0, 1):
    case IP_VERSION(11, 0, 4):
        adev->flags |= AMD_IS_APU;
        break;
    default:
        break;
}

Ah, my mistake.

But do note that "IP version" and "GFX ID" are not the same from what we can see.

N31 may be both GFX1100 and IPver 11.0.0, but Phoenix is GFX1103 but IPver 11.0.1

もう 1つの RDNA 3 APU IP、GC 11.0.4 | Coelacanth's Dream

AMD の Yifan Zhang 氏により、Linux Kernel における AMDGPU ドライバーに GC 11.0.1 (gfx1103, Phoenix APU) に続く RDNA 3 APU IP、GC 11.0.4 のサポートを追加するパッチが投稿されている。 [PATCH 01/19] drm/amdgpu/discovery: enable soc21 common for GC 11.0.4 case IP_VERSION(11, 0, 1): + case

www.coelacanth-dream.com

 

[TESKATLIPOKA]

AMD RDNA3 ISA guide is now available.
Videocardz
Supposedly It covers only Shader ISA.

 

[moinmoin]

AMD RDNA3 ISA guide is now available.
Videocardz
Supposedly It covers only Shader ISA.

Direct links for those who prefer them (like me):
Announcement
AMD RDNA 3 ISA reference guide (pdf)

 

[Joe NYC]

I would even consider 7950X3D If It's not too expensive for you, this way you don't have to worry about CPU for a pretty long time.

N32 should be faster than RTX 3080, RT will be likely worse, but 16GB Vram is very tempting.
Is this level of performance enough for your monster monitor?

I would consider longevity for components differently. The last thing I would want to replace is mobo, 2nd from last is CPU (and no problem replacing is GPU.)

So regarding 7800x3d vs. 7950x3d: I think there is a good case for saving money on the difference between 7800x3d and 7950x3d and apply it to one CPU more upgrade to AM5 motherboard, Zen 5 or something after it

For now, I have no big need for 16 cores, so I will likely also be getting 7800x3d. Unless, there is a measurable performance increase for 7950x3d in gaming...

The way I could see there could be is if there was some way to improve gaming performance in 7950 would be to vacate the CCD with V-Cache of all the threads not related to the game being played, to have all of the background / system threads on the non-V-Cache CCD, which would leave the V-Cache CCD and all of its cache data for the game.

But this would need some sort of customized work on the scheduler...

 

[coercitiv]

Direct links for those who prefer them (like me):
Announcement
AMD RDNA 3 ISA reference guide (pdf)

Finally, the "poor VALU dual issues"

 

[TESKATLIPOKA]

Finally, the "poor VALU dual issues"

Is It Karma for Poor Volta?

 

[PJVol]

But do note that "IP version" and "GFX ID" are not the same from what we can see.

I'm not sure I understand what "GFX ID" is, but driver's code usually follows a certain naming rules, and GC IP version (major, minor, rev.) refers to a certain SKU (or SKUs, if cut-down version exist),
which in itself comprise the 'gpu family', e.g. dgpu - 11.0.0 or apu - 11.0.1.
So, for example the string "1102" just means Graphics Engine maj. ver. 11, min. ver. 0, and rev. 2, which devs seem to have adopted instead of those horrible hard-to-spelll fish names.

 

[TESKATLIPOKA]

ComputerBase tested Adrenaline 22.12.2
Playing youtube reduced consumption:
7900XT: 71W -> 46W
7900XTX: 80 -> 54W
But It's still higher than N21, If you also enable HDR, then It's comparable.

Power consumption didn't change while playing, but limiting performance to 144 FPS decreases power draw much more than before.
Maybe some of you still remember that debate, where N31 ended up less efficient than N21 by limiting FPS to 144.

 

[coercitiv]

Maybe some of you still remember that debate, where N31 ended up less efficient than N21 by limiting FPS to 144.

Now it's N21 that's behaving a bit odd in the ComputerBase tests: all things being equal, the 6800XT and 6900XT should use the same or even lower power than the 6800.

 

[TESKATLIPOKA]

Now it's N21 that's behaving a bit odd in the ComputerBase tests: all things being equal, the 6800XT and 6900XT should use the same or even lower power than the 6800.

You are right. For the same performance, RX6800 should need higher clocks and voltage than RX6900XT, which will increase the power consumption.

 

[PJVol]

ComputerBase tested Adrenaline 22.12.2

I don't get, what's the point of such tests if neither the test scene used nor the in-game graphics settings are specified and there's no way to reproduce the results?
This approach strikes me either overconfident or simply careless.

 

[Glo.]

Paul, from RGT suggests in his recent video that Strix Point has 24 CUs/12 WGPs, not 16 CUs/8 WGPs.

1536 ALUs/3072 vALUs.

This thing HAS to have Infinity cache. Otherwise it doesn't make any sense to put such big amount of sheer horsepower into an APU that would be fed just by memory controller.

 

[moinmoin]

Even with IC that's way too many CUs. On APUs up to now the limited bandwidth also helps limit power usage of the iGPU. Those CUs being optimally used (which is what IC is for) would make the chip a desktop grade one power consumption wise.

 

[Kepler_L2]

Paul, from RGT suggests in his recent video that Strix Point has 24 CUs/12 WGPs, not 16 CUs/8 WGPs.

1536 ALUs/3072 vALUs.

This thing HAS to have Infinity cache. Otherwise it doesn't make any sense to put such big amount of sheer horsepower into an APU that would be fed just by memory controller.

If only there was a way to double memory bandwidth...

 

[leoneazzurro]

If only there was a way to double memory bandwidth...

- Double the number of DDR5 memory channels? (Expensive)
- (Stacked) Infinity Cache?
- HBM on board as L4?
- Other exotic solution?