AMDVLK 2023.Q1.3 Released With Various Fixes, TMZ For RDNA2 & Newer

[phoronix]

Phoronix: AMDVLK 2023.Q1.3 Released With Various Fixes, TMZ For RDNA2 & Newer

It's been nearly one month since AMDVLK 2023.Q1.2 as AMD's latest open-source Vulkan driver code drop while today it was replaced by AMDVLK 2023.Q1.3 as what will likely be their last Vulkan driver update for the quarter...

https://www.phoronix.com/news/AMDVLK-2023.Q1.3

[Mahboi]

I really wish I knew what is making the 7000 cards draw so much power at idle on multi monitor. If some Red Team genius wouldn't mind either explaining or linking to something explaining why or how a GFX can overdraw like this, or even just what is it that decides how much power should the card draw, I'd love to learn.

[smitty3268]

I really wish I knew what is making the 7000 cards draw so much power at idle on multi monitor. If some Red Team genius wouldn't mind either explaining or linking to something explaining why or how a GFX can overdraw like this, or even just what is it that decides how much power should the card draw, I'd love to learn.

Check the GPU clock and memory speeds, would be my guess. Historically I think the issue has been that they can't go down to the minimum and fully draw both screens, and the next step up is set way too high.

[qlum]

I really wish I knew what is making the 7000 cards draw so much power at idle on multi monitor. If some Red Team genius wouldn't mind either explaining or linking to something explaining why or how a GFX can overdraw like this, or even just what is it that decides how much power should the card draw, I'd love to learn.

I personally don't know about that being a specific issue, most my gpu's have drawn quite a bit of power on idle, be it my current rx7900xt my previous gtx1080 or the hd7950 before that. I will add here that I use 4 monitors so that may also factor in here.

[agd5f]

I really wish I knew what is making the 7000 cards draw so much power at idle on multi monitor. If some Red Team genius wouldn't mind either explaining or linking to something explaining why or how a GFX can overdraw like this, or even just what is it that decides how much power should the card draw, I'd love to learn.

Memory clock switching has a certain latency. Additionally, memory clock switching has to happen during a blanking period to avoid corruption on the display. There are various ways to handle it, but depending on the monitor and the refresh rate there's not always a way to handle it. Finally there are some limitations in X that we are still trying to find a good workaround for. See:

https://gitlab.freedesktop.org/drm/amd/-/issues/2315

[Mahboi]

Memory clock switching has a certain latency. Additionally, memory clock switching has to happen during a blanking period to avoid corruption on the display. There are various ways to handle it, but depending on the monitor and the refresh rate there's not always a way to handle it. Finally there are some limitations in X that we are still trying to find a good workaround for. See:

https://gitlab.freedesktop.org/drm/amd/-/issues/2315

So if I'm understanding you correctly, it's not so much a high power usage "by default" but a difficulty at finding the right window to clock down without corruption?
And I expect that whatever standard draw was present with say, 1 1080p monitor, is now completely pointless/insufficient if you have multiple 2K/4K monitors?

It seems that AMD is slowly fixing the issue though:

https://www.techpowerup.com/review/sapphire-radeon-rx-7900-xt-pulse/37.html

On the original XT review, multi-mono cost 85W. There is clear progress.
I wonder why the 4080 manages to be fine with 20W though, but with Ryzen or Radeon, it seems that AMD isn't aiming for the lowest idle consumption possible anyway. There's always quite a bit more leverage on their products (8W idle for Ryzen vs 4 for Intel or something around).