10:38pmoreau: RSpliet: I need to read the whole conversation, but which convention are you using here? CUDA or OpenCL?
10:44pmoreau: RSpliet: Ok, you were talking about local/shared, not local memory in CUDA. got it. I’ll have a look on Monday, I haven’t looked that much at the SASS of my CUDA kernels.
12:51karolherbst: mupuf: do you think it might make sense to limit the voltage on battery mode? because I get 0.85V max
12:52karolherbst: didn't verify it yet
12:52mupuf: well, that's one way of limiting the performance
12:52karolherbst: let me verify first
12:52mupuf: well, it is possible that the bios allows you to limit: the voltage, the frequency and the power ;D
12:52karolherbst: I know it clocks to 600MHz
12:52karolherbst: but sadly, temperature doesn't affect it
12:53mupuf: power is probably what everyone wants, but not every card has a power sensor and the bandwidth of the power sensor is quite low
12:54karolherbst: huh, the hell
12:54karolherbst: now it clocked down to 405MHz
12:55karolherbst: after I changed the votlage entry for 600MHz
12:55karolherbst: uhh wait
12:57karolherbst: it didn't read my modded vbios...
12:57karolherbst: silly reg
13:00karolautumn: my battery is so broken :(
13:00karolherbst: hey, silly irc
13:00karolherbst: wrong nick
13:05karolherbst: ohh wait, maybe I can just set the kernel power mode without the need to unplug the AC
13:10mupuf: that could be appreciated by your power plug
13:10karolherbst: there is a sysfs file somewhere
13:17karolherbst: maybe it was this
13:18mupuf: that would be very weird ":D
13:18karolherbst: yeah, mhh doesn't work anyway
13:19karolherbst: but there should be a way to debug things like that easlily...
13:23karolherbst: noooo, totally forgot I have sl installed
13:26mupuf: the return of the attack of the steam locomotive!
13:27RSpliet: pmoreau: I'm talking about OpenCL local memory, but that should correspond with shared mem in Cuda and there must be a third term for this in (compute?) shaders :-P
13:28karolherbst: well I could also test with glxgears and not furmark
13:28karolherbst: (16W vs 45W)
13:28pmoreau: RSpliet: I would guess that compute shaders use the same terminology as OpenCL
13:29karolherbst: mupuf: my battery is broken for real, charge_full_design: 5200000, charge_full: 831000
13:31mupuf: oh, wowo
13:31mupuf: that is clearly broken, indeed
13:35karolherbst: so it indeed the 600MHz seem to be fix
13:35karolherbst: 0.86875V at 1°C and 0.85V at 95°C
13:35karolherbst: on battery
13:35karolherbst: maybe the battery got an entry referenced though
13:35karolherbst: but I don't think so
13:36karolherbst: yep, no voltage limiting on battery
13:36karolherbst: the 600MHz are also power unrelated
13:40karolherbst: ohh please...
13:40karolherbst: mupuf: the cstate has the index 16... in decimal
13:40karolherbst: guess how annoying will this be to find
13:41karolherbst: maybe I found it
13:42karolherbst: yeah, this might take a while
14:09karolherbst: nvidia, please error check your driver :p
15:02karolherbst: I don't find anything
15:26karolherbst: mupuf: could you imagine nvidia being so weird to hardcode such values?
17:45mupuf: karolherbst: unlikely
17:45mupuf: maybe you should look for the frequency
17:45mupuf: instead of a cstate entry
17:56karolherbst: I did
17:56karolherbst: maybe it isn't within the P tables though
18:20mupuf: maybe another table?
18:25karolherbst: mupuf: I printed out every P table here
18:25karolherbst: I will take another look but meh :/
18:26karolherbst: maybe external vpstate references
18:26karolherbst: maybe nvidia-smi can help here
20:17RSpliet: damn... the unknown SEQ opcodes are a dead-end
20:45sevaho: For a long time I have direclty installed nvidia and blacklisted nouveau, now I want not to do this and use nouveau. However I have some questions
20:45sevaho: Is there a command like optirun/primusrun for nouveau?
20:46sevaho: Is it out of the box with F25 or do I need to configure things to maximize the performance for nouveau?
20:46sevaho: How I know for sure nouveau uses my GPU?
20:47pmoreau: You can use the environment variable DRI_PRIME=0 or 1, with 0 usually being the Intel card, and 1 the NVIDIA one. So, for example `DRI_PRIME=1 glxgears`
20:48sevaho: They are exactly the same
20:48sevaho: I mean the output of those two commands
20:49pmoreau: There is no dynamic reclocking, you need to do it manually: for example, `echo 07 > /sys/class/drm/card0/device/pstate` to switch to level 07. You can cat that file to see the available ones. Only works on Tesla and Kepler cards.
20:49pmoreau: Try with `glxinfo | grep OpenGL` to check which card is used in which case
20:49sevaho: shall I give you more info about my setup?
20:50pmoreau: Kernel version and GPU could be useful
20:50sevaho: 4.8.10-300.fc25.x86_64 and GTX 860M
20:51sevaho: I used to install bumbelbee because I have optimus card
20:51pmoreau: 860M, is that a GM107? (it should be printed in the output of dmesg, or when using lspci)
20:51bernie: Hello, I managed to get modesetting to work on the GTX 970 with 4MB, but I still don't get 3D acceleration in X. Is it expected to work with latest mesa, libdrm and xf86-video-nouveau?
20:52sevaho: 3D controller: NVIDIA Corporation GM107M [GeForce GTX 860M] (rev a2)
20:52pmoreau: bernie: 4GB I assume, cause 4MB isn’t much at all :-D
20:52bernie: yeah right :)
20:53bernie: The 4GB needs a hackpatch in order to work, because the memory layout is funny
20:53pmoreau: Do you have the firmwares installed?
20:53pmoreau: sevaho: I don’t remember whether there is reclocking support for Maxwell v1
20:54pmoreau: But you would want Linux 4.10 anyway, as it has some reclocking fixes for Kepler+
20:54bernie: I get a nice 3840x2160 mode (however it's only 30Hz, not 60Hz), but Xorg.0.log says:
20:54bernie: [ 2962.382] (EE) NOUVEAU(0): Failed to initialise context object: 2D_NVC0 (0)
20:54bernie: [ 2962.382] (EE) NOUVEAU(0): Error initialising acceleration. Falling back to NoAccel
20:55pmoreau: bernie: Could you link the output of dmesg and the whole Xorg.0.log please?
20:55bernie: pmoreau: I attached both to https://bugs.freedesktop.org/show_bug.cgi?id=94990
20:56bernie: (the last 2 attachments a few minutes ago)
20:56bernie: I'm afraid dmesg is truncated, but I could reboot to get the top part
20:56sevaho: aha ok, ps. I see there is a project nouveau/optimis is it any stable, it looks very bleeding edge and tricky
20:58pmoreau: sevaho: Why do you want a separate nouveau/optimus project? The optimus functionalities found in the kernel you are currently running should be enough
20:58pmoreau: bernie: Just a tiny bit truncated :-D Let's have a look at the Xorg one.
20:59sevaho: I just looked it up so idk what it was/is.
21:00bernie: pmoreau: amfern's patch is spammy, I commented out the printk
23:38orbisvicis: does the external firmware ever need to be updated ?
23:47orbisvicis: is this a problem: Unknown PGRAPH archive order in this version.