13:10FLHerne: I have a dual-GPU Optimus laptop, without doing anything special I seem to be using the Intel GPU
13:10FLHerne: which tbh works fine for my normal workloads
13:10FLHerne: Nvidia GPU is GK107M [GeForce GT 640M]
13:11FLHerne: Is there anything I should do to make sure it's powered down fully while not being used?
13:11FLHerne: And does this generation perform well enough with Nouveau that it would be worth using if I do want to play games?
14:05karolherbst: FLHerne: check the power consumption
14:06karolherbst: also, we do support reclocking on the gk107, but because it's quite a low end model, it might not be always faster than the intel one
14:15RSpliet: FLHerne: the easiest way to see if your NVIDIA GPU is powered down when unused is by checking powertop to see if your CPU drops into PC6 or PC7. If it doesn't, something is keeping it awake - often that thing is the GPU on the PCIe bus (but sometimes something on the SATA bus)
14:16RSpliet: When nouveau is loaded this should be the case. When it isn't, I'm not 100% sure how the stars should align.
14:17RSpliet: Note that at some point there used to be a bug in some rare cases (i.e. my laptop) where snd_intel_hda prevented the NVIDIA GPU from powering down on idle. That has been fixed in kernel... erm.... can't remember.
14:18RSpliet: Just run a new kernel, you should be fine :-P
14:54karolherbst: RSpliet: I wouldn't rely on that honestly
14:54karolherbst: there are buggy CPUs where you dont' drop because of stata power states
14:54karolherbst: it's a mess
14:54karolherbst: sensors is probably the best idea
14:54karolherbst: we report N/A as the temp once nouveau runtime suspends the GPU
14:55RSpliet: karolherbst: I did say that ;-) but you're right, if sensors reports N/A rather than waking up the GPU that's more reliable indeed
14:55karolherbst: yeah.. at some point I fixed the hwmon interface stuff
20:57karolherbst: imirkin: any past plans you had at some point about reworking our fence handling? I might have to rework the entire thing in the end :/
20:59karolherbst: although... I think each fence object _has_ to have a screen attached, but some of the aPI makes it soo annoying to extract the screen
20:59karolherbst: and nv04_resource objects without fence attached also suck a bit :/
21:06karolherbst: just wondering if you had anything you disliked about it and I can just fix it while at it as well
21:18imirkin: karolherbst: i wanted to make a kernel syscall
21:18imirkin: so that we didn't have to spin in the cpu
21:18imirkin: that's the main thing
21:18imirkin: separately
21:18imirkin: we have the fences which can be attached to buffers, and then the weird resource status, with gpu reading/writing, which is like ... totally bogus
21:19imirkin: but not 100% bogus, so it still sorta works
21:54isapgswell: hi everyone
21:54imirkin: karolherbst: check logs
21:54isapgswell: i experienced bad external monitor support nouveau/nvidia drivers
21:55isapgswell: i my case hdmi external monitor
21:55isapgswell: any idea?
21:55imirkin: maybe provide details?
21:56isapgswell: i connected my external monitor to hdmi both(nvidia/nouveau) drivers and it lags
21:56isapgswell: video lag
21:56karolherbst: imirkin: ahh yeah :/
21:56karolherbst: I think I might just start thinking about how to deal with the entire thing
21:57imirkin: karolherbst: and the "manual" fences are basically only needed for resource sub-division, and knowing when we can nuke BO's, so we don't do it too early
21:57karolherbst: I think skeggsb will probably take care of the syscall stuff.. dunno
21:57isapgswell: imirkin my notebook has hdmi and displayport
21:58karolherbst: the main issue I have with the current "fence" API is, that it's not as self contained as the mm stuff
21:58imirkin: isapgswell: let's try this ... what GPU?
21:58karolherbst: so.. making it thread safe isn't quite as obvious
21:58karolherbst: and callers have to lock the fence list before operating on fences
21:58imirkin: karolherbst: yeah, esp if you don't want to spin on cpu when you hold a mutex :)
21:58isapgswell: imirkin nvidia gtx 1060
21:58karolherbst: well.. that's why we have simple_mtx_t for
21:58karolherbst: s/for//
21:58imirkin: isapgswell: and what monitor?
21:59imirkin: (what monitor resolution)
21:59karolherbst: but yeah
21:59isapgswell: imirkin aoc 1920x1080
21:59karolherbst: imirkin: anyway, would be good if you could review the mm threading patches so we can land this at least. And in the meantime I try to properly fix the fence stuff
22:00karolherbst: I got the emulator to run somehow.. but this fencing is slowly getting to become very very annoying
22:00imirkin: isapgswell: xorg or wayland?
22:00isapgswell: imirkin xorg
22:00karolherbst: but at least I know what the painful bits are :)
22:01imirkin: isapgswell: and what's the precise problem?
22:01isapgswell: imirkin when connected to hdmi port the video lags, but the mouse pointer dont
22:01isapgswell: imirkin opengl apps too
22:02imirkin: define 'lats'
22:02imirkin: lags
22:02isapgswell: imirkin unusable
22:03imirkin: ok
22:03imirkin: well
22:03imirkin: given this level of detail, not sure what i can do to help
22:04isapgswell: imirkin nvidia and nouveau drivers same problem
22:04imirkin: your setup, as described, should largely work
22:04isapgswell: imirkin windows works great
22:04imirkin: ok, so then your problem isn't really with nouveau or nvidia
22:04isapgswell: imirkin yep
22:05isapgswell: imirkin i ran xrandr --auto and almost usable, but no opengl apps
22:05imirkin: the way a "remote" display works with xorg
22:05imirkin: is that the primary gpu renders the image
22:06imirkin: and then it is copied by the secondary gpu
22:06isapgswell: imirkin i ran xrandr native
22:06imirkin: and scanned out
22:06imirkin: some lag is natural
22:06imirkin: but we're talking like 1 frame at most
22:06isapgswell: imirkin yes
22:06isapgswell: imirkin with xrandr command, is better than nothing
22:07isapgswell: imirkin maybe 20fps
22:07imirkin: if you're complaining about that like 1-frame lag, there's nothing that can be done about it
22:07imirkin: oh, you're not talking about lag
22:07imirkin: you're talking about rendering performance?
22:07isapgswell: imirkinyes
22:07isapgswell: imirkin yes
22:07imirkin: well, unless you're doing something special, you're using your iGPU to render
22:08isapgswell: imirkin i frame max
22:08isapgswell: imirkin 1 frame max
22:08isapgswell: imirkin askubuntu has some questions about that
22:09imirkin: sorry, i don't think i'll be able to help you. good luck in your search!
22:09isapgswell: imirkin ok thanks
22:10isapgswell: gdm-disable-wayland is buggy
22:46FLHerne: RSpliet, karolherbst: Thanks, according to both of those my GPU is off
22:46FLHerne: (thus, any problems I might have aren't yours :p)
23:43Lyude: (jfyi, just changed the threshold for "too many receipients" on nouveau's mailing list to 40 cause i'm tired of it holding all my patches for approval
23:48ccr: :)