05:08 Fijxu: Since now turing and later support almost everything, there is a file where I can see the voltage, temperatures and freqs?
05:08 Fijxu: I can't find anything on the wiki about those stats
05:12 fdobridge_: <a​irlied> no
13:46 karolherbst: Fijxu: we support it through hwmon for some GPUs, but with Turing we need to work things out through GSP, but it's not a priority atm
17:29 fdobridge_: <t​om3026> is the gsp responsible for the "dynamic" boost some laptop dgpus comes with. essentially turning my 80w gpu into a 90w if there is room withing the power limits or similiar
17:29 fdobridge_: <t​om3026> with the blob i use the nvidia-powerd.service for that but the inner workings of it im unsure
17:35 fdobridge_: <k​arolherbst🐧🦀> mhhhhhhhhhh...
17:35 fdobridge_: <k​arolherbst🐧🦀> good question
17:35 fdobridge_: <k​arolherbst🐧🦀> glad you found something to figure out then :ferrisUpsideDown:
17:40 fdobridge_: <r​edsheep> Does that function properly when using the open Nvidia kernel module?
17:40 fdobridge_: <t​om3026> havent tried
17:41 fdobridge_: <r​edsheep> If not it might be something waiting on Nvidia to implement
17:51 fdobridge_: <S​id> that's not what powerd does, it's more for load balancing between the CPU and GPU
17:52 fdobridge_: <S​id> messes more with clock speeds when there's available headroom than power draw
17:53 fdobridge_: <S​id> I'd test it out but it's an Ampere+ feature only, and I have a Turing
18:01 fdobridge_: <t​om3026> well on this 3060ti the clocks are the same, all it does here is raise the 80w limit to 90
18:01 fdobridge_: <t​om3026> err -ti
18:02 fdobridge_: <S​id> huh
18:02 fdobridge_: <t​om3026> il show you :p
18:04 fdobridge_: <t​om3026> https://i.imgur.com/xaEjQ7h.png first is with a game running, and its 95w even. when its closed it drops down to 80w again, and if i simply dont use powerd at all it sticks to 80w
18:04 fdobridge_: <S​id> so the documentation is lying to me :o
18:04 fdobridge_: <t​om3026> but the clocks from the little i have checked earlier sticks to rather same
18:05 fdobridge_: <t​om3026> somewhat bit hard since it jumps up and down and i havent reliably measured it but
18:06 fdobridge_: <t​om3026> this gpu has like 5 different editions all ranging from 80w to 105 lol i bet the clocks are the "same" until it either TDP or temp "throttles" , powerd raises it to 95w lets it go a bit beyond?
18:06 fdobridge_: <t​om3026> *shrug* im no expert heh
18:17 fdobridge_: <r​edsheep> Nvidia has lots of really weird behavior that can mean wildly different performance with a different power limit, regardless of clocks. Also clocks change so fast that the software output isn't as useful as one would hope.
18:28 fdobridge_: <r​edsheep> Because the clocks change so fast and you can only poll clockspeeds so often you can also have sampling bias, like what happened a few years ago with the 3950x where everyone believed amd lied about it reaching 4.7 ghz. Turned out that the software tools to tell what clockspeeds it was hitting just weren't sufficient, and it was almost always done with its 4.7 ghz burst by the time it would get measured.
22:08 raket: Is it possible to make the nouveau kernel module only load it's driver to one of its card in the machine? and if so, how?
22:08 raket: I'm running multiple cards and want to use nvenc with the blob on one of the cards :)
22:49 anzel: Hi, just wondering if anyone working on nvk has done any preliminary work on VK_EXT_descriptor_buffer?
22:53 airlied: I'd expect gfxstrand might have some ideas
23:23 karolherbst: raket: not without patching the kernel afaik
23:23 karolherbst: well.. you could always unbind/bind devices to drivers...