01:34 RSpliet: hmm, abi16 is a thing I have never understood... what's that codepath for?
03:03 karolherbst: yay, my gddr5 patch landed :)
03:15 karolherbst: mupuf_: isn't the FB area more like memory related? the issue I have is not related by the memory for sure
06:13 vita_cell: hi someone
06:18 vita_cell: hi
06:21 vita_cell: hi
06:45 vita_cell: hi
06:51 Wildefyr: buh bye~!
07:15 karolherbst: Wolf480pl: are you there?
07:16 karolherbst: Wolf480pl: I need to create a mmiotrace with you, because for some reasons my kernel does stupid things while I do it :/
07:42 vita_cell: hi, someone, how to add nouveau.pstate=1 for reclock
07:44 pmoreau: vita_cell: You need to put that on the kernel command line
07:45 vita_cell: GRUB_CMDLINE_LINUX="nouveau.pstate=1" ????
07:45 vita_cell: GRUB_CMDLINE_LINUX_DEFAULT="nouveau.pstate=1"???
07:45 vita_cell: I tryed this and nothing
07:45 pmoreau: What do you mean by nothing?
07:46 towo`: vita_cell, and you have done a reabuild of grub.cfg after the changes?
07:46 vita_cell: I tryed add this , updating grub, and does nothig
07:46 vita_cell: sudo update-drub
07:46 pmoreau: What do you mean by "does nothing"? What do you expect it to do?
07:47 vita_cell: I customized somethings in grub and all work
07:47 vita_cell: but pstate doesn't
07:47 vita_cell: pstate does reclock?
07:47 pmoreau: It makes a file visible for manual reclocking, that's all
07:48 vita_cell: so, waht to do for reclock?
07:48 pmoreau: With nouveau.pstate=1, you should see the following file /sys/class/drm/card0/device/pstate
07:48 pmoreau: If you cat it, you can see the various perflvls
07:48 vita_cell: adding nouveau-pstate=1 to coomand line of linux, but what to do later?
07:48 pmoreau: and by echo'ing to it a perflvl id, you can go to that perflvl
07:50 vita_cell: LoL, in /sys/class/drm/ I have too much cards, cuz I testing and benchmarking, so, how do I know what GPU I using now?
07:50 Wolf480pl: karolherbst, now I am
07:53 Wolf480pl: karolherbst, what should I mmiotrace? or do you want me to first try out that patch you mentioned?
07:56 karolherbst: Wolf480pl: well just mmiotrace the blob with Coolbits set to 8
07:56 karolherbst: then open nvidia-settings for the card
07:56 Wolf480pl: Coolbits? what's that?
07:56 karolherbst: xorg.conf option
07:56 karolherbst: wait
07:57 karolherbst: Wolf480pl: just add Option "Coolbits" "8" in /etc/bumblebee/xorg.conf.nvidia
07:57 karolherbst: inside Section "Device"
07:58 karolherbst: and while nvidia-settings is open you need to reclock the card a bit
07:58 karolherbst: I think you can set the core freq offset to -135
07:58 karolherbst: add a mark inside the trace
07:58 karolherbst: clock to 0
07:58 karolherbst: add another mark
07:58 karolherbst: and then stop the trace
07:59 pmoreau: vita_cell: Hum…
08:01 pmoreau: vita_cell: cards in /sys/class/drm are links to pci devices, so if you do "ls -l /sys/class/drm", you will see the PCI bus ID in the address of the linked file
08:01 pmoreau: And then, if you know which card is on which PCI bus ID, you're good to go
08:02 Wolf480pl: karolherbst, do I need to start mmiotrace before I start Xorg and bumblebeed, or is it ok if I start it just before nvidia-settings?
08:03 vita_cell: I only using PCI gtx650
08:04 karolherbst: Wolf480pl: you need to start the mmiotrade before the driver gets loaded
08:04 karolherbst: so
08:04 karolherbst: you could do something like that.
08:04 karolherbst: 1. start mmiotrace
08:04 karolherbst: 2. optirun -b none nvidia-settings -c :8
08:04 karolherbst: and then do the other stuff
08:04 vita_cell: vita@vita-desktop:~$ ls /sys/class/drm
08:04 vita_cell: card0 card0-DP-3 card0-HDMI-A-3 card1-DVI-D-1 controlD64 renderD129
08:04 vita_cell: card0-DP-1 card0-HDMI-A-1 card0-VGA-1 card1-HDMI-A-4 controlD65 ttm
08:04 vita_cell: card0-DP-2 card0-HDMI-A-2 card1 card1-VGA-2 renderD128
08:04 karolherbst: vita_cell: I see only two cards there
08:05 vita_cell: card1 and card0 I supposing
08:05 vita_cell: but I don't know what I am using right now
08:05 vita_cell: vita@vita-desktop:~$ lspci -vnn | grep VGA -A 12
08:05 vita_cell: 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GK107 [GeForce GTX 650] [10de:0fc6] (rev a1) (prog-if 00 [VGA controller])
08:05 vita_cell: Subsystem: ASUSTeK Computer Inc. Device [1043:848a]
08:05 vita_cell: Flags: bus master, fast devsel, latency 0, IRQ 35
08:05 vita_cell: Memory at f5000000 (32-bit, non-prefetchable) [size=16M]
08:05 vita_cell: Memory at d0000000 (64-bit, prefetchable) [size=256M]
08:05 vita_cell: Memory at e0000000 (64-bit, prefetchable) [size=32M]
08:05 vita_cell: I/O ports at e000 [size=128]
08:06 vita_cell: Expansion ROM at f6000000 [disabled] [size=512K]
08:06 vita_cell: Capabilities: <access denied>
08:06 vita_cell: Kernel driver in use: nouveau
08:06 vita_cell: 01:00.1 Audio device [0403]: NVIDIA Corporation GK107 HDMI Audio Controller [10de:0e1b] (rev a1)
08:06 vita_cell: Subsystem: ASUSTeK Computer Inc. Device [1043:848a]
08:08 vita_cell: I think that it is:
08:18 imirkin: please use pastebin for more than 3 lines
08:19 karolherbst: Wolf480pl: how are you doing so far?
08:34 Wolf480pl: karolherbst, sorry, haven't started yet, had a more important thing to do
08:37 Wolf480pl: hmm... it says problem with xorg config file
08:37 Wolf480pl: you sure the 8 shouldn't be quoted?
08:46 Wolf480pl: oh, actually, it was in your message, it's just me who can't read
08:48 karolherbst: :D
08:49 Wolf480pl: I set it to -135, maked, to 0, marked
08:49 Wolf480pl: but the performance level stayed at 0 all the time
08:50 Wolf480pl: isn this correct behavior?
08:51 Wolf480pl: also, the clock freq displayed didn't change
08:51 Wolf480pl: is this what was supposed to happen, or did it somehow not get set?
08:52 karolherbst: Wolf480pl: you need to fake the load
08:52 karolherbst: ohh wait
08:52 karolherbst: mhh
08:52 karolherbst: envytools installed?
08:52 Wolf480pl: yeah
08:52 karolherbst: nvapoke 0010a534 10380001
08:52 karolherbst: then it should clock to max
08:53 Wolf480pl: I already stopped the trace
08:53 karolherbst: ohh I see
08:53 Wolf480pl: so I should reboot, start the trace, nvapoke, open nvidia-settings, tweak the offset, mark, etc?
08:54 karolherbst: I just need the part where the driver reclocks after you changed the offset
08:54 karolherbst: and you need to press enter after setting the offset
08:54 Wolf480pl: yeah, I did press enter
08:54 karolherbst: you should see that the clock changes then
08:54 karolherbst: k
08:54 karolherbst: mhhh
08:54 karolherbst: well
08:54 Wolf480pl: it said on the status bar that it's setting the offset
08:54 karolherbst: you need to poke when the card is on obviously
08:54 Wolf480pl: maybe I should just repeat what I did with glxspheres running in the background?
08:54 karolherbst: no
08:54 karolherbst: the trace gets polluted then
08:54 Wolf480pl: oh, ok
08:54 karolherbst: I really odn't want to figure out what belongs to clocking
08:55 Wolf480pl: I see
08:55 Wolf480pl: so repeat the procedure but nvapoke before setting the offset?
08:56 Wolf480pl: also, do you need the part where driver is loaded?
08:57 Wolf480pl: karolherbst?
08:57 karolherbst: mhh
08:57 karolherbst: no, just the clocking thing
08:57 vita_cell: pmoreau
08:57 karolherbst: so
08:57 Wolf480pl: no need to reboot
08:57 karolherbst: start nvidia-drivers
08:58 karolherbst: nvidia-settings
08:58 karolherbst: then poke the value
08:58 karolherbst: check if the driver clocks to max
08:58 karolherbst: set offset to -135
08:58 karolherbst: add mark
08:58 karolherbst: set offset to 0
08:58 karolherbst: add mark
08:58 karolherbst: quit settings and stop the trace
08:59 vita_cell: how I edit /sys/class/drm/card0/device/pstate file?
08:59 pmoreau: vita_cell: echo 0f > /sys/class/drm/card0/device/pstate, if you want to switch to level 0f
08:59 pmoreau: (as root)
08:59 vita_cell: yes, how I swith to 07 or 0a level?
09:00 vita_cell: I tryed edit with Gedit as root pstate but I can't edit it
09:00 pmoreau: cf the command I just pasted
09:00 vita_cell: I have this in my pstate file:
09:01 Wolf480pl: karolherbst, maybe I also add a mark right before I set the offset?
09:01 vita_cell: 07: core 324 MHz memory 648 MHz
09:01 vita_cell: 0a: core 540 MHz memory 1620 MHz
09:01 vita_cell: 0f: core 1071 MHz memory 5000 MHz
09:01 vita_cell: AC: core 324 MHz memory 648 MHz
09:01 pmoreau: So you have three different levels: 07, 0a, 0f
09:01 vita_cell: 4 levels
09:02 vita_cell: and AC
09:02 pmoreau: To switch to one of them, you run (as root) "echo level_id > /sys/class/drm/card0/device/pstate"
09:02 pmoreau: No, AC is supposed to print the current level
09:02 vita_cell: ok
09:02 pmoreau: Which is 07 right now
09:03 vita_cell: permission denied
09:03 vita_cell: with root: sudo su
09:03 vita_cell: root@vita-desktop:/home/vita# echo 07 > /sys/class/drm/card0/device/pstate
09:03 vita_cell: bash: /sys/class/drm/card0/device/pstate: Permiso denegado
09:04 pmoreau: Maybe it is card1 for you
09:04 vita_cell: ahhhh I fool
09:04 vita_cell: lol
09:05 karolherbst: Wolf480pl: you can if you want
09:05 vita_cell: I tryed and it did something, some flash
09:06 pmoreau: Is your computer still running or did it crash?
09:06 vita_cell: promeau, this command switches between pstate levels (3 levels), this switches in realtime, without reboot?
09:06 pmoreau: Yep
09:06 Wolf480pl: karolherbst, too late, did the trace with just 2 marks
09:06 Wolf480pl: it's 127MB
09:06 pmoreau: And if you reboot, you'll go back to the boot level
09:06 Wolf480pl: xz-ing it now
09:06 vita_cell: not, only flickered, running well, so, I only tried switching to the first level
09:06 karolherbst: Wolf480pl: I just need the stuff between those marks
09:06 Wolf480pl: oh
09:07 vita_cell: so, this is realtime switch
09:07 pmoreau: vita_cell: you were already on the first level :D
09:07 pmoreau: Try to go to 0a
09:07 vita_cell: Can I damage my GPU if it not going hotter?
09:07 karolherbst: pmoreau: not quite right though
09:07 karolherbst: pmoreau: the init state is where the card is when it powers on
09:07 vita_cell: yes, first, lol, only trying
09:08 karolherbst: there is no "clock to lowest pstate" at init time
09:08 pmoreau: karolherbst: Except for him AC == 07
09:08 vita_cell: karolherbst, I testing only
09:08 karolherbst: so voltage and other stuff can change when going to 07
09:08 vita_cell: thanks
09:09 karolherbst: pmoreau: yes the clocks might be the same, but not everything is
09:09 pmoreau: Ah, voltage could, though not sure if you're not changing the clocks?
09:09 karolherbst: pmoreau: nouveua might decide to user different pll values and stuff
09:09 pmoreau: K
09:09 vita_cell: My GPU (gtx650) stays only at 30º, when playing it is on 33º
09:09 karolherbst: pmoreau: for me the voltage changes
09:09 karolherbst: Wolf480pl: the stuff shouldn't be more then 10.000 lines, more likely 1000
09:10 pmoreau: vita_cell: Nice cooling
09:10 Wolf480pl: karolherbst, it's 19k lines
09:10 karolherbst: Wolf480pl: ohh
09:10 pmoreau: vita_cell: Try to switch to 0a or 0f and see what happens
09:10 karolherbst: a bit much, but I will check what's going on there
09:11 vita_cell: cant I damage my GPU if I run it at 0a: core 540 MHz memory 1620 MHz?
09:11 Wolf480pl: the part between the marks is <800K
09:11 karolherbst: yeah
09:11 karolherbst: its not much
09:12 Wolf480pl: 58K after xz-ing
09:12 Wolf480pl: though I'll probably upload it uncompressed
09:13 pmoreau: vita_cell: Those values in pstate are not overclocking! Those are the values between which the card is supposed to work, to trade between performance and power consumption.
09:13 vita_cell: switched to 0a 0a: core 540 MHz memory 1620 MHz
09:13 pmoreau: What does AC: say?
09:13 vita_cell: flickered, when switching,
09:14 vita_cell: root@vita-desktop:/home/vita# cat /sys/class/drm/card1/device/pstate
09:14 vita_cell: 07: core 324 MHz memory 648 MHz
09:14 vita_cell: 0a: core 540 MHz memory 1620 MHz AC DC *
09:14 vita_cell: 0f: core 1071 MHz memory 5000 MHz
09:14 vita_cell: AC: core 324 MHz memory 1620 MHz
09:14 pmoreau: Can happen I guess
09:15 vita_cell: still on 324Mhz
09:15 pmoreau: Could you paste the output of `dmesg` somewhere on the internet?
09:15 vita_cell: So these values, is only for that model card?
09:15 vita_cell: what?
09:16 vita_cell: If I put the output, here, It comes very lagged
09:16 pmoreau: Run "dmesg > some_file", and then upload that file to whatever website you prefer, like pastebin, or some other
09:17 Wolf480pl: karolherbst, https://www.dropbox.com/s/ivu6xfqm8ow1wh8/nvidia-reclock-trace-cut.txt?dl=0
09:20 vita_cell: dmesg is too long, I can not see it completely
09:22 pmoreau: vita_cell: I need to go, check with karolherbst
09:23 vita_cell: ok, thaks you very much !!!
09:23 pmoreau: np
09:24 imirkin_: vita_cell: any use of unofficial drivers could, in theory, damage your hw. however we have no such reports that i can think of.
09:25 vita_cell: I use only free software OS
09:25 vita_cell: Trisquel 7 64
09:25 vita_cell: and I want to use only free software drivers
09:25 imirkin_: ok
09:25 imirkin_: you were asking about potential for hw damage, i answered.
09:25 vita_cell: Yes, it is now official binary blobs, and it can damage in theory
09:26 vita_cell: My GPU can be damaged only by termperature or something more?
09:26 imirkin_: a hammer often does the trick
09:26 karolherbst: imirkin_: my gddr5 patch landed in upstream nouveau now, so I think it will land for 4.4 then
09:27 karolherbst: Wolf480pl: ohh I think I will need everything :/
09:28 vita_cell: I have some question to do, what will perform better i5-5675C's GPU (Iris Pro 6200) or gtx770?
09:28 karolherbst: Wolf480pl: or at least the header of the trace
09:29 Wolf480pl: whole trace https://www.dropbox.com/s/e8nwpjcy36f143v/nvidia-reclock-trace.txt.xz?dl=0
09:29 Wolf480pl: 7MB xzipped
09:29 karolherbst: k
09:30 vita_cell: https://www.dropbox.com/s/bqafakq2utphu7p/3345?dl=0
09:30 vita_cell: my dmesg output, but I can't see it complete
09:31 karolherbst: Wolf480pl: ohh voltage
09:32 karolherbst: I meant
09:32 karolherbst: vita_cell:
09:32 karolherbst: vita_cell: do you have envytools installed?
09:32 karolherbst: if not, please install
09:32 vita_cell: I don't know what it is
09:32 vita_cell: wait
09:33 vita_cell: after reclock to second pstate, I improved performance, but +5º for GPU's temp
09:33 karolherbst: yeah
09:33 karolherbst: the memory is faster
09:33 karolherbst: should be 1620 then
09:33 vita_cell: and core?
09:33 karolherbst: well
09:33 karolherbst: you can't set the voltage
09:33 karolherbst: I want to find out why
09:34 karolherbst: but you may want to wait for linux 4.4 before trying out 0f
09:34 vita_cell: AC: core 324 MHz memory 1620 MHz
09:34 vita_cell: I am on 4.3 linux-libre
09:35 karolherbst: have to go for a bit
09:35 karolherbst: will be back later
09:35 karolherbst: vita_cell: after installing envytools
09:35 karolherbst: do: nvapeek 20344
09:37 vita_cell: wait, I installing depends, and later compile
09:42 vita_cell: Warning: nvapy won't be built because of un-met dependencies (python and cython)
09:51 vita_cell: vita@vita-desktop:~$ nvapeek 20344
09:51 vita_cell: WARN: Can't probe 0000:01:00.0
09:51 vita_cell: PCI init failure!
09:51 vita_cell: vita@vita-desktop:~$ sudo nvapeek 20344
09:51 vita_cell: [sudo] password for vita:
09:51 vita_cell: ...
09:51 vita_cell: vita@vita-desktop:~$
09:51 imirkin_: fyi, ... = 0
09:52 vita_cell: what is it?
09:52 vita_cell: I don't know what is nvapeek
09:52 imirkin_: 0? it's a number signifying "nothing". important advance between the roman and arabic numerical systems...
09:53 vita_cell: karolherbst told me to run it
09:53 vita_cell: but ... "0" output
09:59 imirkin_: karolherbst: btw, congrats on your first significant accepted contribution :) now time to try to upstream some more of it...
10:07 karolherbst: imirkin_: that just means he has GPIO based voltaging and not pwm based
10:07 karolherbst: mhhh
10:07 karolherbst: something really messy is going on then
10:07 karolherbst: vita_cell: in the directory wheere the pstate file was, there is a vbios.rom file
10:07 karolherbst: I need that
10:08 imirkin_: actually it's in a different dir
10:08 imirkin_: it's in debugfs
10:08 vita_cell: how I get Vbios.rom?
10:08 karolherbst: ohh right
10:09 karolherbst: in /sys/kernel/debug/dri/1
10:11 karolherbst: imirkin_: yeah, pcie stuff, voltage in sensors and pdaemon counters are targeted for 4.5 by me :)
10:11 karolherbst: maybe I get more in
10:12 vita_cell: can't copy vbios
10:12 vita_cell: permission denied
10:12 imirkin_: root
10:12 imirkin_: also try using 'cat', e.g. cat vbios.rom > /tmp/vbios.rom
10:13 vita_cell: did gksu thunar
10:14 karolherbst: vita_cell: you can't just open the file like something on your ext filesystem (or whatever you use)
10:14 karolherbst: these kernel files are a bit different then "ordinary" files
10:15 karolherbst: you won't be able to just copy the file
10:15 vita_cell: https://www.dropbox.com/s/d026wxe0056ims9/vbios.rom?dl=0
10:15 karolherbst: thanks
10:15 vita_cell: copyed with cat, thanks
10:16 karolherbst: vita_cell: and I need nvapeek 101000
10:17 karolherbst: mupuf_: -- Mode GPIO, Base voltage 0 µV, voltage step 0 µV, acceptable range [0, 0] µV -- :D
10:18 vita_cell: how I get it?
10:18 karolherbst: nvapeek 101000
10:19 vita_cell: 00101000: 8040989e
10:19 vita_cell: this?
10:19 karolherbst: thanks
10:19 karolherbst: I don't know what it is for, but mupuf_ always tells me to ask for it, so I just do that :p
10:20 vita_cell: I have more GPUs, I give my gtx460oc to my friend for gtx650, cuz it is Kepler
10:20 vita_cell: anyway 460 runs not very good
10:20 vita_cell: and I have 7300gt
10:21 karolherbst: kepler is well supported by nouveau if you care about performance, ...
10:21 karolherbst: there are just some cards not that good handled
10:21 karolherbst: like yours
10:21 pmoreau: karolherbst: strap_peek is used by nvbios iirc to correctly identify the card/something. I remember seeing it used once. think it was in nvbios but I could be wrong
10:22 vita_cell: I want to ask something, i5-5675C's GPU(Iris Pro 6200) will run better that gtx770?
10:22 karolherbst: pmoreau: yeah I saw that nvbios needs it for something
10:22 karolherbst: vita_cell: difficult question
10:22 karolherbst: vita_cell: wiht 4.4 there is a gddr5 reclokcing fix
10:22 vita_cell: Intel GPUs run very very well, but 60hz cap not enough for me
10:22 karolherbst: and with that the nvidia card should run faster with nouveau than the iris gpu
10:23 vita_cell: 4.4 what is it?
10:23 karolherbst: linux kernel
10:23 vita_cell: I am on 4.3
10:23 karolherbst: maybe the 0a pstate of the 770 is fast enough for outperforming the intel one
10:23 karolherbst: maybe
10:23 karolherbst: should be close though
10:24 karolherbst: my intel 4600 is as fast as my 770m at 0a pstate with nouveau
10:24 vita_cell: wow
10:24 karolherbst: the intel gpus are getting really fast now days
10:24 karolherbst: outperforming everything under x50
10:24 vita_cell: Iris Pro 6200 more powerful that HD series
10:24 karolherbst: like 640 or 740
10:25 vita_cell: I want to buy it, but it is only 60hz GPU
10:25 karolherbst: why is it only 60Hz?
10:25 vita_cell: Intel capped their GPUs
10:25 vita_cell: I think
10:25 vita_cell: all Intel's GPUs only run 60hz
10:25 karolherbst: why is that?
10:26 karolherbst: mhh
10:26 karolherbst: I am not sure though
10:26 karolherbst: but if the resolution is lower than max, it sould handle higher frequenciews
10:26 vita_cell: All Intel GPUs I tryed, are capped
10:26 karolherbst: ohh
10:26 vita_cell: and I using 144hz monitor
10:27 karolherbst: mhhh
10:27 karolherbst: I think you need to use hdmi
10:27 karolherbst: DVI is capped at 60hz
10:27 karolherbst: 1920x1080@120 hz should work
10:27 karolherbst: ohh 144
10:27 vita_cell: not, DVI-DL it is more powerful
10:28 karolherbst: not on intel
10:28 vita_cell: on HDMI only 1920x1080-120hz
10:28 karolherbst: right
10:28 vita_cell: Intel supports only 60hz
10:28 karolherbst: no
10:28 karolherbst: intel supports 1920x1080@120
10:28 karolherbst: on hdmi
10:28 karolherbst: just not DVI
10:28 karolherbst: use DP or HDMI
10:29 karolherbst: with DP you even get 1920x1080@144
10:29 vita_cell: I tryed HD4000, and only 60hz
10:29 karolherbst: I looked up 4600
10:29 imirkin_: yeah, intel DVI is only single-link. very weird.
10:29 karolherbst: imirkin_: there is no native support for it
10:29 karolherbst: that's why
10:30 vita_cell: it is, only 60hz
10:30 imirkin_: ?
10:30 imirkin_: how is that an explanation...
10:30 karolherbst: vita_cell: just don't use DVI with intel for that kind of stuff
10:30 karolherbst: HDMI and DP are much better for that with intel gpus
10:31 imirkin_: it doesn't support dual-link becuase dual-link isn't supported? not very satisfactory for a reason why it got left out of the hw...
10:31 karolherbst: that's just how it is
10:31 karolherbst: imirkin_: right :D
10:31 karolherbst: there is just no support for it
10:31 imirkin_: i know... that was my initial statement
10:31 karolherbst: yeah, my statement was silly
10:31 karolherbst: I take it back
10:31 imirkin_: and it's very weird that there's no support for it.
10:31 imirkin_: not like dual-link is some new-fangled thing
10:32 vita_cell: Anyway, with VGA or HDMI, Intel's GPU will run only 60hz
10:32 karolherbst: especially if they do the stuff through DP
10:32 imirkin_: well, DP doesn't have enough pins
10:32 imirkin_: so DP -> DVI is always single-link (same with HDMI -> DVI)
10:32 hakzsam: imirkin_, do you want me to remove those ul suffixes or are you comfortable with them?
10:32 imirkin_: hakzsam: please remove them
10:32 imirkin_: assuming it doesn't cause problems
10:32 karolherbst: imirkin_: yeah, maybe there is no DVI support at all and it is just a DP -> DVI conversion on the gpu
10:32 karolherbst: who knows
10:33 pmoreau: imirkin_: They should only cause some implicit casts
10:33 hakzsam: imirkin_, fine by me. What about the patch 2?
10:33 karolherbst: vita_cell: I am pretty sure with HDMI you get 120hz
10:33 imirkin_: you'll get implicit casts anyways...
10:33 imirkin_: ul != uint64_t
10:34 vita_cell: I will try, but I think that not
10:34 imirkin_: dunno what the max pixclock on intel hdmi is
10:34 pmoreau: If long isn't 64bits
10:34 imirkin_: might still not be possible
10:34 imirkin_: but pretty sure it's more than 165mhz
10:36 karolherbst: imirkin_: yeah, but I somewhere read that 144hz won't work
10:37 karolherbst: imirkin_: weird though: 4096x2304@24Hz is stated as max for hdmi 1.4
10:37 imirkin_: karolherbst: in theory, yes
10:37 imirkin_: karolherbst: in practice, you can be hdmi-compliant without supporting the max speeds
10:37 karolherbst: 3840x2160@60Hz for my intel card
10:38 karolherbst: funny though
10:38 karolherbst: there is nothing mentioning DVI in the intel specs
10:38 karolherbst: so I really think there is just no DVI support on these intel gpus
10:39 karolherbst: 6200 has 883.2 GFLOPS :/ that's insane :D
10:40 karolherbst: vita_cell: well the 770 might have doubled performance at 0f pstate with nouveau compared to 6200
10:41 karolherbst: :D
10:42 vita_cell: and compared to gtx650?
10:42 karolherbst: intel should be 1.5x faster
10:42 karolherbst: compared to nouveau max clocks
10:42 imirkin_: 6200 is skylake right?
10:42 vita_cell: not
10:42 vita_cell: 5 generation
10:43 imirkin_: ah, broadwell
10:43 vita_cell: i5-5675C
10:43 karolherbst: 580 GT4e has 1152 GFLOPS :D
10:43 vita_cell: is here someway to get higher clocks with core?
10:44 vita_cell: AC: core 324 MHz memory 5040 MHz
10:44 vita_cell: too low
10:44 glennk: still a ways to go before it matches a 2009 era discrete gpu
10:45 karolherbst: vita_cell: voltaging doesn't work for your card for some reasons
10:45 vita_cell: :(
10:45 karolherbst: glennk: yeah, but before intel didn't even get close to those low end gpus
10:45 karolherbst: and now the just outperform half of the mid ends
10:46 glennk: oh yeah at least its within the same decade
10:46 karolherbst: yeah
10:46 karolherbst: well
10:46 vita_cell: 770 supports core reclocking?
10:46 karolherbst: same year you mean :p
10:46 karolherbst: vita_cell: all kepler do
10:46 glennk: though 2016 will be interesting for discrete gpus
10:46 karolherbst: just some cards are a bit more different than others
10:46 karolherbst: glennk: why?
10:46 glennk: 2nd gen HBM and 14nm finfet
10:46 imirkin_: vita_cell: there are loads of board-specific reasons why clocks might not be changed. in general it is supported though.
10:47 vita_cell: mine doesn't reclock the core :(
10:47 vita_cell: something I did wrong I think
10:47 imirkin_: doubtful
10:48 vita_cell: not latest drivers?
10:48 vita_cell: Gallium 0.4 on NVE7
10:48 imirkin_: that has nothing to do with anything
10:48 vita_cell: 4.1 (Core Profile) Mesa 11.0.4
10:48 imirkin_: the clock changes are done by kernel
10:49 vita_cell: only Kepler supports reclocking?
10:49 imirkin_: chances are in your kernel logs you see errors like "couldn't change voltage" or something
10:50 vita_cell: where kernel's logs stored?
10:50 vita_cell: I forget
10:51 imirkin_: dmesg
10:52 vita_cell: [ 3848.033539] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
10:52 vita_cell: [ 3848.033545] nouveau 0000:01:00.0: clk: error setting pstate 0: -22
10:52 vita_cell: :(
10:52 vita_cell: downgrading kernel can help? it is why I am using Linux-libre kernel?
10:53 vita_cell: i have some errors more
10:55 karolherbst: vita_cell: maybe 4.3 messed up and 4.2 works
10:55 karolherbst: but I doubt it somehow
11:10 m3n3chm0:nasZ
11:13 karolherbst: vita_cell: sorry, I won't be much help debugging this issue, I try to talk with mupuf_ bout that because he has more knowledge about the voltage situation in general
11:14 vita_cell: ok, no problem, anyway, thanks for your help
11:15 karolherbst: vita_cell: you could try out the 0f pstate if you want
11:15 karolherbst: but I am pretty sure your card will crash that way
11:15 vita_cell: what? how?
11:15 karolherbst: you need 4.4 so that it won't crash :p
11:15 karolherbst: or you could compile nouveua yourself and stuff
11:16 vita_cell: it works with of, but not for core
11:16 vita_cell: AC: core 324 MHz memory 5040 MHz
11:16 karolherbst: yeah
11:16 karolherbst: try to put some load on it
11:16 karolherbst: vblank_mode=1 glxgears or something
11:16 karolherbst: I am sure your screen will freeze or something
11:16 karolherbst: maybe not
11:16 karolherbst: who knows
11:17 vita_cell: I tryed playing Stunt Rally 2.5 and with 144hz
11:17 vita_cell: it goes to 40º temp
11:17 karolherbst: Wolf480pl: ohh right, there are multiple reclockings in the trace :/
11:17 vita_cell: works fine, but no more Mhz with core, stucked at 324
11:18 karolherbst: mhh
11:18 karolherbst: I see
11:18 karolherbst: maybe the memory doesn't get enough load to be unstable
11:18 vita_cell: don't know
11:33 karolherbst: Wolf480pl: can you run nouveau now?
11:34 Wolf480pl: karolherbst, I can
11:34 Wolf480pl: it'll just take a reboot
11:34 karolherbst: though I have no clue what I search for currently
11:34 karolherbst: but when I want to try something out, it will be with nouveau
11:35 Wolf480pl: will it be the version I ran last time, or will there be some newer patches?
11:37 karolherbst: don't know yet
11:37 karolherbst: :D
11:38 karolherbst: I try to overclock my gpu until I reach 1 GHz :/
11:38 karolherbst: maybe I manage to get there maybe not
11:39 Wolf480pl: with `watch -n 0,5 sensors` in a second terminal window?
11:39 karolherbst: nah
11:39 karolherbst: my gpu doesn'T get hot
11:40 karolherbst: only 61° C
11:40 karolherbst: 35W power consumption
11:40 Wolf480pl: is it a desktop GPU?
11:40 karolherbst: my cooling stuff is laid out for 70W
11:40 karolherbst: nope
11:40 karolherbst: laptop
11:40 Wolf480pl: wow
11:40 karolherbst: 770M
11:40 karolherbst: just good cooling
11:40 karolherbst: seperated fan
11:40 karolherbst: no heatpipes shared with cpu
11:41 karolherbst: mhh
11:41 karolherbst: 914MHz was fine
11:41 karolherbst: 927MHz not
11:42 karolherbst: Wolf480pl: with heaven I reach like 80°C with the binary driver though :D
11:42 karolherbst: I don't know what is wrong though, this sounds a bit weird somehow
11:42 karolherbst: I mean the entire issue
11:43 karolherbst: there seems to be a magical barrier I don't understand
11:43 karolherbst: maybe I should check glxgears first, but well
11:45 karolherbst: mhhh
11:45 karolherbst: glxgears is stable indeed
11:46 karolherbst: okay, with glxgears 966MHz crashes the card
11:46 karlmag: How much is it supposed to be able to do?
11:47 karolherbst: karlmag: with blob I can drive most stuff at 997MHz overclocked
11:47 karolherbst: though
11:47 karolherbst: spec for this card is like 797 MHz :D
11:47 karolherbst: and vbios comes with 862MHz
11:47 karlmag: ah
11:48 karolherbst: trying pixmark piano
11:48 karolherbst: this should crash the card like immediatly
11:51 karolherbst: mhhhhhhhhh
11:51 karolherbst: I wonder about something
11:51 karolherbst: 979MHz stable with piano
11:51 karolherbst: or not
11:52 karolherbst: imirkin_: what does that mean? gr: GPC2/TPC0/MP trap: global 00000000 [] warp 3000d [GPR_OUT_OF_BOUNDS]
11:53 imirkin_: it means that i messed up
11:53 karolherbst: ohh
11:53 imirkin_: when running a shader we tell it what the max gpr is
11:53 karolherbst: really?
11:53 karolherbst: mhhh
11:53 karolherbst: I got this while running pixmark piano at 979MHz
11:53 imirkin_: i think GPR_OUT_OF_BOUNDS means that we used a gpr higher than what we said we'd use
11:54 imirkin_: could of course be some scrweup in tranferring the shader
11:54 karolherbst: mhh okay
11:55 karolherbst: imirkin_: then if it is only caused at higher core clocks, it most likely means the gpu itself screwed up?
11:55 karolherbst: would be nice to know what ROP0 80000000 80000001 means :/
11:55 imirkin_: it means "you screwed up" :)
11:55 karolherbst: :D
11:55 karolherbst: hey
11:55 karolherbst: I reclocked like the blob reclocks
11:56 karolherbst: going into divider mode, changing the pll, going into pll mode :p
11:56 karolherbst: :D
11:56 imirkin_: you can argue with the hw all you want
11:56 karolherbst: :/
11:56 imirkin_: no matter how convincing your argument, it still wins
11:56 vita_cell: why not run "glmemperf" and "glmark2"?
11:57 karolherbst: because I don't have it installed
11:57 karolherbst: and I don't kow what they benchmark
11:57 karolherbst: imirkin_: I am surprised that I can run my card 0.1V below what the blob does and still it runs
11:58 karolherbst: ohhh wait
11:58 karolherbst: I have to commit the voltage :O
11:58 karolherbst: nooooooooo, how can I forgot
11:58 karolherbst: :(
11:58 vita_cell: glmark2 is a benchmark for OpenGL (ES) 2.0. It uses only the subset of the OpenGL 2.0 API that is compatible with OpenGL ES 2.0.
11:59 karolherbst: now it all makes sense :/
11:59 karolherbst: head -> wall
11:59 karolherbst: damn it
11:59 karolherbst: Wolf480pl: I messed up
12:00 vita_cell: is Asus GTX770-DC2OC-4GD5 is a good one?
12:01 karolherbst: yay, running at 992MHz :)
12:01 karolherbst: Wolf480pl: okay
12:01 karolherbst: I know what I did wrong
12:02 karolherbst: Wolf480pl: you know that nvapoke 0x20344 44 part for your card?
12:02 karolherbst: where we wanted to increase the voltage?
12:02 Wolf480pl: yeah, it increased it to 1.02V, according to sensors
12:02 karolherbst: actually there needs to be a nvapoke 0x20344 80000044 after that
12:02 Wolf480pl: you sure about the address?
12:02 karolherbst: 8 means something like: commit the pwm value to the actual hardware
12:02 Wolf480pl: is it left-aligned?
12:03 Wolf480pl: or right-aligned?
12:03 karolherbst: nvapoke 20344 80000044
12:03 karolherbst: if you don't trust it :D
12:03 karolherbst: so clock to 0f
12:03 karolherbst: do nvapoke 0x20344 44
12:03 karolherbst: then nvapoke 0x20344 80000044
12:03 karolherbst: and then try out glxspheres
12:03 karolherbst: :/
12:03 karolherbst: I am so forgetfull sometimes
12:04 Wolf480pl: ok, sec
12:04 karolherbst: yep
12:04 karolherbst: nvapoke 0x20344 33 did nothing
12:04 karolherbst: 80000033 => crash
12:04 karolherbst: dues to undervolting
12:06 karolherbst: Wolf480pl: also
12:06 karolherbst: I would need the normal value for nvapeek 0x20344
12:06 karolherbst: then I can give you a patch where you just get higher voltages set through nouveau
12:07 vita_cell: Hi
12:08 vita_cell: 07: core 324 MHz memory 648 MHz
12:08 vita_cell: 0a: core 540 MHz memory 1620 MHz
12:08 vita_cell: 0f: core 1071 MHz memory 5000 MHz AC DC *
12:08 vita_cell: AC: core 324 MHz memory 5022 MHz
12:08 vita_cell: this means that I am on 324Mhz?
12:08 Wolf480pl: karolherbst, I accidentally poked 20334
12:08 karolherbst: mhhh
12:08 Wolf480pl: should I reboot?
12:08 karolherbst: don't know what 20334 does
12:08 karolherbst: mhhh
12:08 karolherbst: wait
12:08 karolherbst: no
12:09 karolherbst: nvapeek 20334
12:09 Wolf480pl: says '...'
12:09 karolherbst: yeah, then the card either didn't care or whatever
12:09 karolherbst: should be fine
12:10 karolherbst: but you need to poke that after switching to 0f
12:11 Wolf480pl: I know
12:11 Wolf480pl: it works
12:11 karolherbst: :)
12:11 karolherbst: okay, then I was the dense one :D
12:11 Wolf480pl: 76 *C
12:11 karolherbst: Wolf480pl: you would need a patch like this one: https://github.com/karolherbst/nouveau/commit/78f66e575de26f89ed46e513fa8de40476b9121a
12:11 karolherbst: I use +6 because it is stable enough for me
12:11 karolherbst: but still lower than the blob
12:11 RSpliet: vita_cell: yes...
12:12 Wolf480pl: also, the voltage sensor doesn't seem to care about commits
12:12 RSpliet: which... if it works is rather hilarious in some ways O:-)
12:12 karolherbst: Wolf480pl: right
12:12 karolherbst: because the voltage thingy just reads 0x20344
12:12 Wolf480pl: which means it doesn't actually measure the voltage
12:12 karolherbst: Wolf480pl: yes, you are right
12:12 karolherbst: but it is better than nothing actually
12:13 karolherbst: maybe there are cards with dedicated voltage sensors which can read out the actualy voltage
12:13 karolherbst: but that's has to be found first
12:13 karolherbst: Wolf480pl: did you compile nouveau yourself?
12:13 Wolf480pl: I did a couple days ago
12:13 karolherbst: k
12:13 karolherbst: you could though decrease the value until it crashes
12:14 karolherbst: it's a hex value by the way
12:14 karolherbst: so 40-1 = 3f ;)
12:14 Wolf480pl: you want me to find the lowest working voltage?
12:14 karolherbst: nah, it's more for you
12:14 karolherbst: if you think the card is too hot
12:15 karolherbst: use a lower voltage
12:15 Wolf480pl: oh
12:15 karolherbst: it is good to know which value inside 20344 is stable for everything
12:15 Wolf480pl: the fan wasn't going full speed, so I don't thing the temperature is that much of an issue
12:15 Wolf480pl: I have a "force fan full speed" button, so yeah
12:15 t4nk433: Error loading nouveau driver on boot.
12:15 karolherbst: as long as it doesn't reach 100°C it's fine
12:16 karolherbst: Wolf480pl: yeah, I can do that with Fn+1
12:16 t4nk433: journalctl -b 0 | grep nouveau
12:16 Wolf480pl: karolherbst, I was told 80 is the borderline
12:16 karolherbst: Wolf480pl: depends on the gpu
12:16 karolherbst: for me the blob emergency downclock my gpu at around 97° C
12:16 karolherbst: until then, it doesn't care much
12:17 Wolf480pl: here, the sensors say 95 is high, 105 is crit, 135 is "run away!"
12:17 RSpliet: t4nk433: try not grepping and just dumping the entire log on a paste website... tht should give a lot more info
12:17 karolherbst: yeah
12:17 karolherbst: but these are static values I think
12:17 t4nk433: kernel: nouveau ![ DEVICE][0000:01:00.0] unknown Maxwell chipset
12:17 karolherbst: Wolf480pl: I have the same, they are pretty meaningless
12:17 t4nk433: kernel: nouveau E[ DEVICE][0000:01:00.0] unknown chipset, 0x118060a2
12:17 RSpliet: t4nk433: ah there you go, your GPU is too new for your kernel
12:18 Wolf480pl: karolherbst, are they tweakable? :P
12:18 karolherbst: Wolf480pl: ohhhh you have alos a power sensor
12:18 karolherbst: maybe you really want to use my branch
12:18 t4nk433: <Rspliet>Card-2: NVIDIA GM108M [GeForce 930M] bus-ID: 01:00.0
12:18 karolherbst: Wolf480pl: for me sensors also prints stuff like that: "power1: 57.06 W" for the gpu :D
12:18 RSpliet: t4nk433: in fairness, maxwell doesn't work too well with nouveau at all. You can try and get a picture by updating your kernel, but it's still unusable
12:18 Wolf480pl: karolherbst, the no-touchy one?
12:18 karolherbst: yes
12:18 karolherbst: it should be fine now
12:19 karolherbst: but it requires 4.3
12:19 karolherbst: there is also this pwm hack with the +6 offset
12:19 karolherbst: might be good for you too already
12:19 RSpliet:glares at gnurou
12:19 karolherbst: Wolf480pl: ohh als pcie patches are there
12:20 karolherbst: so you might get a _big_ speedup in glxgears and glxspheres
12:20 karolherbst: like I get 18000 fps with glxgears :)
12:20 t4nk433: may be on the kernel 4.3 suportted
12:20 imirkin_: t4nk433: search for GM108 in bugs.freedesktop.org
12:20 imirkin_: still unsupported
12:20 imirkin_: but pretty easy to hack it in yourself
12:20 karolherbst: Wolf480pl: and 1400fps with glxspheres
12:20 Yoshimo: how much from your repository is missing in 4.3 and 4.4 karol?
12:20 Wolf480pl: karolherbst, I get like 19k fps with glxgears and 1300 with glxspheres
12:20 karolherbst: Wolf480pl: ohhh
12:20 karolherbst: Wolf480pl: what branch are you usimng?
12:20 RSpliet: karolherbst: did your PCIe patches get merged? I don't see them *sob*
12:20 karolherbst: no, I think he uses my branch already
12:21 karolherbst: :D
12:21 Wolf480pl: karolherbst, karol_master_stable_4_3 + nvapoke
12:21 karolherbst: ahh k
12:21 karolherbst: I thought so
12:21 karolherbst: yeah
12:21 karolherbst: then you already have the pcie stuff
12:21 karolherbst: this gives you a plain +5% speed up overall
12:21 karolherbst: or more
12:21 t4nk433: my kernel it's 4.2.0-1-amd64 debian testing kde
12:22 Wolf480pl: karolherbst, is there much difference between 4.3rc7 and 4.3 ?
12:22 karolherbst: Wolf480pl: I don't know
12:22 karolherbst: but should only be bugfixes
12:22 karolherbst: if there would be something important
12:22 karolherbst: 4.3 would be 4.3rc8
12:22 Wolf480pl: mhm
12:23 karolherbst: after rc1 a kernel version only gets bugfixes usually
12:23 karolherbst: sometimes new stuff, but only if there is a good reasons
12:23 t4nk433: it's a kernel update.. the new kernel it's 4.3 i don't don't if this supported the chpiset
12:23 karolherbst: *reason
12:23 karolherbst: Yoshimo: which branch? :D
12:24 karolherbst: Yoshimo: everything above "drm: bump patchlevel to indicate availability of abi16/nvif interop" is missing
12:24 karolherbst: in 4.4
12:24 karolherbst: 4.3 a lot more
12:24 karolherbst: I just took the 4.4 nouveua branch, removed all the stuff making it incompatible with 4.3
12:24 karolherbst: and that's my master branch
12:24 karolherbst: and every other branch is on top of that, except the 4_1 and 4_2 branches
12:25 karolherbst: Wolf480pl: so I think we are done with your card, the issue that nouveau doesn't get the voltage exactly right is known
12:26 karolherbst: might tackle that now then, because it seems to be important now
12:26 karolherbst: :D
12:26 Wolf480pl: will it land in 4.4 ? :P
12:26 karolherbst: Wolf480pl: no, only my gddr5 fix is merged for 4.4
12:26 karolherbst: that's why voltage is more important now
12:27 karolherbst: because more people will complain about that now
12:27 karolherbst: because 0f will more likely work switching too
12:27 karolherbst: but crash for certain workloads
12:27 karolherbst: earlier it was like: 0f with gddr5 is broken because of the memory, now it is because of the voltage ;)
12:27 pmoreau: Progress! :-)
12:28 karolherbst: yay :D
12:28 pmoreau: Or just moving the problem around :D
12:28 karolherbst: nah it is better actually
12:28 pmoreau: ;-)
12:28 karolherbst: because increasing the voltage a bit is no big deal
12:28 karolherbst: I take stuff to serious sometimes
12:28 karolherbst: :D
12:28 Wolf480pl: also, imirkin_ said something about whitelisting my card's PCI ID so that War00C800_0=1 is enabled by default
12:29 karolherbst: yeah, I need to do that for my card too
12:29 karolherbst: :D
12:29 Wolf480pl: is this also waiting for 4.5 ?
12:29 karolherbst: mhhh
12:29 karolherbst: bugfix and then into 4.4 :p :D
12:29 karolherbst: don't know though
12:30 karolherbst: need to figure out how to add an entry there first
12:30 karolherbst: imirkin_: how do I do that?
12:30 imirkin_: karolherbst: look at the 2 patches i sent
12:30 karolherbst: ... how am I not surprised by that answer?
12:30 imirkin_: Wolf480pl: are you sure that you need the War00C800 thing for your card to work?
12:30 pmoreau: Wolf480pl: I think I saw whitelisting for 4.4
12:30 imirkin_: karolherbst: http://cgit.freedesktop.org/~darktama/nouveau/commit/?id=10360ccf9bc14a1a0cfc3d2b14e40aca3d138c7f
12:31 imirkin_: Wolf480pl: i believe this should cover it for your case: http://cgit.freedesktop.org/~darktama/nouveau/commit/?id=29da420dfbe47fb7028b371e25175c631e4e14b8
12:31 karolherbst: what does this 0x0fcd value mean?
12:31 imirkin_: pci id...
12:32 karolherbst: there are like three different ones
12:32 karolherbst: so I assumed on of them isn't related to the card
12:32 imirkin_: karolherbst: lspci -vnn -d 10de::300
12:32 karolherbst: 10de:11e0
12:32 imirkin_: 10de:0fcd is the device id, 17aa:3801 is the subdevice id
12:33 karolherbst: ohhh okay
12:33 karolherbst: subdevice id is like id for _that_ specific model?
12:33 imirkin_: sort of... it's not 100% specific
12:33 imirkin_: but we can't really do better
12:33 karolherbst: k
12:33 imirkin_: ben saw that some hw had issues
12:33 Wolf480pl: imirkin_, I am sure, I tested 4.3rc7 without that option once, and IIRC it froze quickly after launching Xorg, or after setting offload provider
12:33 karolherbst: ohh okay
12:33 karolherbst: then I got it
12:34 imirkin_: so we can't just enable it everywhere
12:34 imirkin_: perhaps we can enable it on all the mobile ones... dunno
12:34 karolherbst: yeah, got it
12:34 imirkin_: iirc vbios even tells us if it's mobile or not
12:35 Wolf480pl: btw. do the nvidia guys know what that workaround actually does? (cause I saw someone somewhere say sth like "we need to ask nvidia guys why this workss")
12:36 imirkin_: like does everyone employed at nvidia know what it does? unlikely
12:36 imirkin_: does *someone* there know? seems eminently likely, since their driver is the one that does it.
12:41 vita_cell: does gt520 can be reclocked ?
12:41 imirkin_: not with nouveau
12:47 imirkin_: skeggsb: btw, dunno if you saw, but looks like it's the acpi fast method causing problems for the HP Zbook people
12:48 imirkin_: skeggsb: glancing at the before & after logic, i can't immediately tell what's changed before and after...
13:01 Wolf480pl: karolherbst, thank you for figuring out reclocking for my card
13:02 Wolf480pl:hugs karolherbst
13:15 karolherbst: Wolf480pl: no problem, I knew it was voltage related from the start, I just forgot that the pwm value needs to be commited :/
13:18 Niautanor: Can someone help me figure out why the maximum resolution of my new 21:9 monitor doesn't get detected propperly?
13:19 Niautanor: It works when I plug it into my Laptops intel graphics thingy but not on my GTX 560 ti
13:19 imirkin_: Niautanor: using hdmi?
13:19 Niautanor: jep
13:19 imirkin_: try DVI or DP
13:19 imirkin_: nouveau artificially limits HDMI to 165MHz pixel clock
13:20 imirkin_: there are some patches available if you're comfortable recompiling yoru kernel
13:21 Niautanor: DVI works thank you
13:21 Niautanor: why does it limit the pixel clock?
13:22 imirkin_: HDMI doesn't support dual-link, and we incorrectly limit the max to 165mhz/link
13:23 imirkin_: unfortunately we don't know how to retrieve the real max speed
13:25 Niautanor: Huh, I thought DVI would work because it showed up in the output of xrandr but now that I switched to it, my monitor just says "Out of Range"
13:25 imirkin_: heh
13:26 imirkin_: perhaps it's not a dual-link cable?
13:26 imirkin_: or the monitor doesn't actually support dual-link dvi?
13:26 imirkin_: that'd be surprising
13:26 Niautanor: It's actually just a HDMI to DVI cable because the monitor doesn't have a DVI port
13:27 imirkin_: oh
13:27 imirkin_: that's why
13:27 imirkin_: hdmi is single-link
13:27 imirkin_: i guess we're not smart enough to pick up on that
13:30 Niautanor: Hm :/
13:31 imirkin_: if you have a DP connector on the GPU and the monitor (and a DP cable) that should also use the higher freqs out of the box
13:32 Niautanor: My gpu only has 2xDvi and 1xHDMI
13:32 imirkin_: ok, well if you like i can point you at some patches
13:32 Niautanor: that would be nice
13:33 imirkin_: Niautanor: http://lists.freedesktop.org/archives/nouveau/2015-August/021841.html + http://lists.freedesktop.org/archives/nouveau/2015-August/021839.html
13:33 imirkin_: replace 225000 with your favourite large number. some GPUs apparently max out at 297mhz
13:33 vita_cell: hi, I am on https://launchpad.net/~xorg-edgers/+archive/ubuntu/ppa repository
13:33 vita_cell: can I install something more updated?
13:34 vita_cell: (drivers)
13:34 Niautanor: thank you imirkin_
13:46 TheSchaf: hey, i wonder how much effort it is to do some "hello world" on a gpu without any drivers (like rendering a textured triangle, or something)
13:47 TheSchaf: is it doable? or is there too many things required for that to work?
13:49 imirkin_: TheSchaf: without something like mesa, pretty easy. without something like the nouveau kernel driver... could be tricky
13:50 imirkin_: lots of things to set up
13:51 TheSchaf: mhh, okay
13:51 imirkin_: you could probably skip lots of stuff
13:51 imirkin_: you wouldn't need context switching for example
13:51 imirkin_: or memory management
13:51 imirkin_: and iirc there's a VM-less mode that the gpu can operate in
13:52 TheSchaf: i keep seeing all those "write your own OS" examples so i wondered how hard it would be to add some gpu rendering to them :)
14:06 karolherbst: TheSchaf: hard :p
14:07 karolherbst: TheSchaf: because first you are stuck with writing the PCIe driver stuff, which isn't trivial
14:08 TheSchaf: :(
14:13 karolherbst: imirkin_: like that? https://github.com/karolherbst/nouveau/commit/b9bd0577e112abd42306b87a1caaa3236fda3d9e
14:14 imirkin_: karolherbst: yep
14:14 mupuf_: karolherbst: i'm here
14:14 karolherbst: mupuf_: already solved the issue
14:14 mupuf_: ver good1
14:14 karolherbst: mupuf_: I forgot the pwm value has to be commited
14:15 karolherbst: ...
14:15 mupuf_: hehe :)
14:15 karolherbst: mupuf_: got my gpu overclocked by 135MHz without problems now under nouveua :)
14:15 karolherbst: so I guess everything does work indeed
14:15 karolherbst: and there is no magic with higher clocks
14:16 karolherbst: mupuf_: but the pwm voltage has to be higher indeed
14:16 karolherbst: much higher in fact
14:16 karolherbst: Wolf480pl: do you remember what was the stock value at 0f with nouveau in 20344?
14:17 mupuf_: higher than the blob or higher than computed?
14:17 mupuf_: I cannot work this week, I host a friend
14:18 karolherbst: higher than computed
14:19 karolherbst: mupuf_: like nouveau computes 0x31 for me, and blob uses 0x3d
14:19 mupuf_: yes, that I agree :)
14:19 karolherbst: 0x37 seems stable though
14:20 karolherbst: this is a difference of 0.7V :/
14:20 glennk: TheSchaf, you can probably find some cheap low end arm board which has a much simpler setup, look for ones that don't really have a full on gpu on them but just basic 2d hardware
14:22 glennk: karolherbst, sure you don't mean .07V? .7 sounds like "fry this hardware please" kind of difference...
14:23 karolherbst: glennk: yes 0.7
14:23 karolherbst: ohh wait
14:23 karolherbst: you are right :D
14:23 karolherbst: 0.07
14:23 karolherbst: :D
14:23 mupuf_: ah ah
14:23 mupuf_: .7V is more than the range of pwm values ;p
14:23 karolherbst: mupuf_: ahh right, I'Ve uploaded a nice vbios for ya
14:23 glennk: .7 is slightly higher than the silicon diode junction voltage
14:24 karolherbst: mupuf_: what do you think if you see something like that: "-- Mode GPIO, Base voltage 0 µV, voltage step 0 µV, acceptable range [0, 0] µV --"
14:25 mupuf_: are there gpios to control the voltage?
14:26 karolherbst: this is the header of hte voltage table
14:26 karolherbst: ohhh
14:26 karolherbst: missread your sentence
14:26 karolherbst: yes, 5
14:27 karolherbst: and the voltage tables has 27 entries
14:27 karolherbst: *table
14:27 karolherbst: starting with -- Vid 0, voltage 0 µV/837500 µV (unk = 2288) --
14:27 karolherbst: mupuf_: nve7/vita_cell
14:27 karolherbst: he can't reclock his gpu because setting voltage just fails
14:28 vita_cell: yes, I want to reclock my gtx650's core
14:28 vita_cell: I tryed downgrading kernels and drivers
14:29 karolherbst: vita_cell: and older kernels also didn't work I assume?
14:29 vita_cell: 4.1, 4.1.x, 4.2.5
14:29 vita_cell: doesn't
14:30 vita_cell: no idea where it is the problem
14:30 karolherbst: I have an idea and I think mupuf_ also has :D
14:30 mupuf_: hehe, sure
14:30 karolherbst: strange voltage table header
14:30 vita_cell: mupuf_ trying to get higher core clock?
14:31 karolherbst: maybe this header doesn't do much on the blob?
14:31 vita_cell: he could reclock memory?
14:31 mupuf_: vita_cell: the only think I am trying is to understand how power management works on nvidia gpus
14:32 mupuf_: let me have a quick look
14:32 vita_cell: yes, no body know, only blobs maker do
14:32 vita_cell: no problem
14:32 mupuf_: we know a lot, but the edge cases are the problem
14:33 vita_cell: the problem is, that I thinking to buy 770 and I don't know if it is a good idea
14:33 imirkin_: reclocking is very much a work in progress
14:33 imirkin_: if you want something that works, use intel, or amd
14:33 mlankhorst: just like 4 years ago :)
14:34 vita_cell: AMD, just joke lol
14:34 mupuf_: mlankhorst: right :D But we got knowledge about kepler since then :D
14:34 mupuf_: well, part of it
14:34 vita_cell: AMD works only with non free firmware loaded in the kernel
14:34 imirkin_: amd has a full-time paid team that is armed with internal and external documentation
14:34 imirkin_: nouveau can't operate an nvidia board without the VBIOS either...
14:34 vita_cell: you can not run AMD fully free, like Nvidia with Nouveau
14:35 mlankhorst: mupuf_: then maxwell and pascall came along
14:35 mupuf_: karolherbst: hmm, this header is not really extremelly necessary if we are in GPIO mode
14:35 imirkin_: not sure i follow your logic
14:35 mupuf_: mlankhorst: we are always playing ketchup, yes
14:35 mupuf_: :D
14:35 imirkin_: are you saying that if they made the cards $1 more expensive and stuck a ROM with the firmware on them, it'd be fully free?
14:35 karolherbst: mupuf_: :D
14:35 mupuf_: ketchup is funnier than catch up
14:35 vita_cell: you rigght, but with Nvidia I have no blobs in the kernel, the last GPU from AMD I have was 7970 and no more AMD
14:36 karolherbst: vita_cell: well the blob isn't in the kernel but on your filesystem usually :p
14:36 imirkin_: and it's a blob of data, not a blob of code (at least executable by your system's cpu)
14:37 vita_cell: the problem is, that you can't run fully free OS with AMD, not with good resolution, 3d support acceleration, compiz...
14:37 karolherbst: vita_cell: well the code inside those blobs are usually executed on the gpu
14:37 karolherbst: not the cpu
14:38 vita_cell: with 7970 I had my room and PC like furnance
14:38 vita_cell: very very hot
14:38 karolherbst: vita_cell: could you boot with nouveau.debug=volt=debug ?
14:38 karolherbst: wait
14:39 karolherbst: is that right? :D
14:39 mupuf_: karolherbst: yes
14:39 vita_cell: in here? GRUB_CMDLINE_LINUX=""
14:39 mupuf_: yep
14:39 vita_cell: or here'? GRUB_CMDLINE_LINUX_DEFAULT="nouveau.pstate=1"
14:39 karolherbst: strange, doesn't work here :/
14:39 karolherbst: ohh
14:39 karlmag: ok, this might be a really stupid question, but regarding vbioses.. Is there a place where one can know which are available, and perhaps more important; which are needed?
14:40 karolherbst: karlmag: you use the vbios on your gpu
14:40 karolherbst: at least for nouveau that is
14:40 vita_cell: GRUB_CMDLINE_LINUX_DEFAULT="nouveau.pstate=1 nouveau.debug=volt=debug" ????
14:41 karlmag: karolherbst: well.. you wanted a copy of one of mine for whatever reason.
14:41 karolherbst: karlmag: yes, debuging
14:41 karlmag: I guess I meant for development reasons (primarily)
14:41 karolherbst: not primarily, but only for that
14:41 karlmag: 'k
14:42 vita_cell: Why AMD doesn't release their code like Intel with GPUs, but they working on open-source drivers? WTF?
14:42 karlmag: assuming they are more or less different for different cards (are they?), is there a "centralized repo" of sorts for it?
14:43 karlmag: I might have access to fairly many different ones, so I could just try to get hold of them if it's desired.
14:44 karolherbst: karlmag: well
14:44 karolherbst: a vbios only helps if there is something not working for the card which depends on stuff in that vbios
14:45 karlmag: seems reasonable
14:45 karlmag: I don't know.. I guess it's a somewhat naive try to possibly be helpful.
14:45 imirkin_: Niautanor: fyi i just sent a slightly more robust version of the second patch: http://lists.freedesktop.org/archives/nouveau/2015-November/022992.html
14:46 vita_cell: how to boot with GRUB_CMDLINE_LINUX_DEFAULT="nouveau.pstate=1 nouveau.debug=volt=debug" ?
14:50 gryffus: imirkin_: sorry for the delay, i fell asleep :x The patch you sent me is working :)
14:50 imirkin_: you fell asleep like 24 ago :p
14:50 imirkin_: that's quite a nap
14:50 imirkin_: i think my personal best is like 17h
14:50 imirkin_: (in one sitting, so to speak)
14:50 sarnex: hi guys, my nouveau card came in. i dont see it in xrandr --listproviders but i see it in dmesg | grep nouveau, any ideas
14:51 gryffus: imirkin_: i wasn't sleeping all the time :D but i was able to get to the PC just now
14:51 imirkin_: sarnex: your life will be much simpler if you use DRI3
14:51 imirkin_: sarnex: in which case it'll Just Work (tm) with DRI_PRIME=1
14:51 sarnex: ok im already using it for radeon
14:51 gryffus: hi sarnex
14:51 imirkin_: sarnex: do DRI_PRIME=1 glxinfo
14:51 sarnex: so ill just use the xorg.conf section
14:51 sarnex: imirkin_: i have alot of cards so it should be DRI_PRIME=2
14:51 sarnex: but ill try it
14:51 imirkin_: oh
14:51 imirkin_: no, that won't work
14:52 imirkin_: i was complaining about that
14:52 imirkin_: there's no way to select which card you want
14:52 imirkin_: i think i lobbied for an env var to specify the rendernode device
14:52 imirkin_: i guess you really do need the dri2 offloading
14:52 karolherbst: what?
14:52 sarnex: ok is there a way to disable my bad radeon card then
14:52 karolherbst: you can't choose which card?
14:52 karolherbst: sarnex: remove the driver :D
14:53 sarnex: my main card is radeon
14:53 imirkin_: in which case you must not have an xorg.conf device at all :(
14:53 karolherbst: sarnex: how mayn gpus do you have?
14:53 sarnex: 3
14:53 sarnex: 2 radeon 1 nouveau
14:53 karolherbst: ohhh
14:53 karolherbst: I see
14:53 karolherbst: you could mhhh
14:53 karolherbst: well
14:53 karolherbst: maybe you can remove the card from the pcie bus through the kernel
14:54 imirkin_: } else if (!strcmp(default_device_id_path_tag, prime))
14:54 sarnex: is that a patch i can apply somewhere
14:54 imirkin_: looks like you can use the 'id_path_tag', e.g. pci-0000_02_00_0
14:54 karolherbst: sarnex: inside /sys/bus/pci/devices
14:54 imirkin_: adjust to taste
14:54 imirkin_: sarnex: listen to me, not karolherbst :p
14:54 sarnex: imirkin_: is that an envvar also?
14:54 karolherbst: :p
14:54 karolherbst: :D
14:54 imirkin_: yeah, DRI_PRIME=pci-0000_02_00_0
14:55 sarnex: oh ok neat
14:55 karolherbst: ohh
14:55 imirkin_: in theory
14:55 karolherbst: that's actually pretty smart
14:55 imirkin_: if i read the right loader
14:55 imirkin_: yep, works for me
14:55 sarnex: imirkin_: yep libGL: OpenDriver: trying /usr/lib64/dri/nouveau_dri.so
14:55 sarnex: good call
14:56 karolherbst: imirkin_: but it doesn't fail if you write garbage in it
14:56 karolherbst: :D
14:56 sarnex: it says using DRI3 but i didnt set that for nouveau, is that seperate
14:56 imirkin_: karolherbst: no, it just picks the default device in that case, which is your main card
14:56 sarnex: eg is the nouveau ddx really using dri2
14:56 karolherbst: imirkin_: I know
14:56 imirkin_: sarnex: you're nto using the nouveau ddx
14:56 karolherbst: sarnex: you don't have a ddx with DRI3
14:56 sarnex: oh
14:56 sarnex: cool
14:56 karolherbst: it's just some kernel magic as far as I know :D
14:57 imirkin_: dri2 requires an X-server side ddx integratino. dri3 doesn't
14:58 sarnex: imirkin_: how can i tell if im reclocking
14:59 imirkin_: sarnex: cat /sys/class/drm/card*/device/pstate
14:59 karlmag: hmm... I guess the blob doesn't give out a vbios...
14:59 imirkin_: sarnex: last line should tell you your current state
14:59 imirkin_: the lines before it, the available states
15:00 imirkin_: echo the level id into the file to switch states
15:01 sarnex: unless im an idiot there is no pstate in card{0,1,2}/device
15:02 imirkin_: boot with nouveau.pstate=1
15:03 sarnex: ok ill add that
15:03 imirkin_: (have i mentioned that reclocking is experimental?)
15:04 sarnex: dont worry i wont sue you when my dorm burns down
15:04 sarnex: brb
15:06 sarnex: ok i see pstate now
15:06 vita_cell: nouveau.debug=volt=debug can work for reclock core?
15:07 sarnex: does it clock down when not in use?
15:07 imirkin_: vita_cell: that just prints more debug messages
15:07 imirkin_: sarnex: there's no automatic reclocking
15:07 imirkin_: sarnex: the reclocking isn't reliable enough to be automatic :)
15:09 sarnex: ok cool
15:09 vita_cell: what is the difference: GRUB_CMDLINE_LINUX_DEFAULT vs GRUB_CMDLINE_LINUX
15:12 sarnex: imirkin_: what was the game that had the virtual memory overflow
15:12 imirkin_: metro: ll
15:13 gryffus: and GW2 https://github.com/iXit/Mesa-3D/issues/153
15:13 sarnex: well i have to leave GW2 to install overnightn or something
15:13 sarnex: because the installer is so awful
15:14 gryffus: sarnex: yeah it is
15:14 gryffus: btw you have my apitrace
15:15 gryffus: does it help to reproduce the bug?
15:15 sarnex: gryffus: can you remind me of the one that should crash
15:17 gryffus: sarnex: apitrace/nine_gw2_output/native_directx/apitrace-nine/
15:17 sarnex: thanks
15:17 gryffus: sarnex: no problem
15:17 vita_cell: nouveau.debug=volt=debug does nothing
15:23 karlmag: I read DumpingVideoBios on the wiki. It gives several ways of dumping it. I tried both the "using /sys" and through debugfs... It produced different sized files on the same machine. Is this to be expected?
15:28 sarnex: strange i cant get it to replay under nouveau, NINE_DEBUG=all says its doing stuff but the screen is just gray
15:31 sarnex: oh i just had to wait like 45 seconds
15:31 imirkin_: err... why?
15:31 sarnex: gryffus: it indeed crashes for me
15:32 imirkin_: oh, coz you were waitnig for it to exhaust your memory space :)
15:32 sarnex: no idea i dont have that issue on radeon it just sits at a gray screen for a while
15:32 sarnex: well it started playing after that, like i saw the game
15:32 imirkin_: a glimpse :)
15:32 imirkin_: anwyays... any issues you identify i'm happy to look into
15:32 sarnex: how can i tell if my address space is being eaten
15:33 imirkin_: look at /proc/<pid>/maps?
15:33 imirkin_: btw if you guys put together some way for me to run traces, and find traces that repro issues on nouveau, feel free to file bugs on bugs.freedesktop.org
15:34 sarnex: well you have to install wine patched with nine
15:34 imirkin_: that stinks
15:34 sarnex: yeah
15:36 sarnex: does nouveau not implement VRAM-usage?
15:37 sarnex: gallium_hud: unknown driver query 'VRAM-usage'
15:37 imirkin_: dunno... there are a ton of various things you look at in the hud
15:37 imirkin_: i forget what they are
15:38 sarnex: ah i see
15:38 sarnex: apparently not
15:39 imirkin_: send patches ;)
15:39 sarnex: only work i can do is trivial
15:41 imirkin_: too bad
15:41 sarnex: there are 680 lines in maps when it dies, so i guess it isnt eating address space for me
15:41 gryffus: sarnex: glad it does not crash only for me
15:41 imirkin_: a single line can account for any amount of memory
15:42 sarnex: oh
15:42 sarnex: is one of the volumns the size or do i have to subtract the values
15:42 sarnex: columns
15:46 imirkin_: i forget
15:47 sarnex: gryffus: i guess ill let it install overnight and try myself
15:49 sarnex: imirkin_: i just tried WOW, i get the same UI corruption as https://github.com/iXit/Mesa-3D/issues/149
15:50 sarnex: is there a way i can help with that
15:50 imirkin_: sarnex: sure, figure out what's causing it, and fix it
15:50 sarnex: i mean in reference to your comment on the bug, something about finding the bad format?
15:50 sarnex: if its up to me to fix it there is no hope
15:51 gryffus: sarnex: ok
15:51 imirkin_: sarnex: look at what formats are being used, and start removing support for non-standard ones :)
15:51 sarnex: and these are the mesa formats, not the wine formats right
15:51 imirkin_: oh also... i dunno if st/nine respects the state tracker's format list. if not, that's a huge problem :)
15:52 gryffus: sarnex: do you also get wow crash if you run it with -opengl switch with nouveau?
15:52 imirkin_: well, with st/nine no mesa formats. just pipe formats.
15:52 sarnex: gryffus: yes but i got this issue on radeon also
15:52 gryffus: for me it crashes with "state_tracker/st_atom_framebuffer.c:60:update_framebuffer_size: Assertion `surface' failed." message
15:53 gryffus: sarnex: oh... ok
15:53 imirkin_: if you guys can come up with a trace, that'd be useful to have a bug reported
15:55 sarnex: gryffus: if you have time :P
15:55 gryffus: imirkin_: i have tried only traces with nine/directx, if you point me to some guide how to do it for opengl, i can du it
15:55 sarnex: imirkin_: stupid question how can i find out what pipe format it's using
15:55 imirkin_: sarnex: mmmm... you can troll GALLIUM_TRACE=foo.xml output
15:55 sarnex: ok
15:57 sarnex: 5000 matches of PIPE_FORMAT
15:57 sarnex: worry
15:57 imirkin_: there's a tools/something/format_dump.py i think
15:57 imirkin_: or... similar name
16:04 sarnex: im getting alot of formats
16:06 sarnex: ill just start disabling stuff i guess
16:07 imirkin_: make a list
16:07 imirkin_: i'll point the weird ones
16:08 sarnex: the output is 12000 lines grepping for PIPE_FORMAT im not sure of a good way to extract just PIPE_FORMAT_* and remove duplicates
16:08 imirkin_: sed + sort
16:15 sarnex: oh neat
16:15 imirkin_: sort -u of course
16:15 sarnex: imirkin_: https://bpaste.net/show/e13eba49a5aa
16:16 imirkin_: PIPE_FORMAT_L4A4_UNORM -- kill that one
16:17 gryffus: i cannot find 32bit apitrace for openSUSE so that WoW trace could be a problem :(
16:17 sarnex: will do
16:18 imirkin_: i should have killed off all the 4 formats, i guess i missed those
16:19 imirkin_: sarnex: http://hastebin.com/dumususaze.cs
16:19 sarnex: ok ill test now
16:26 aaaa: imirkin_: multiple monitors still crashes X(but not my whole system), i booted with nouveau.noaccel=1 and nouveau.nofbaccel=1 and i am not getting some errors that were in dmesg. dmesg: https://bpaste.net/show/ef7715d466fd Xorg log: https://bpaste.net/show/228ae6e97463
16:27 sarnex: imirkin_: same issue
16:27 sarnex: ill generate a new list of formats
16:28 imirkin_: aaaa: that xorg log is not from that boot
16:32 aaaa: imirkin_: how do you know? i tried using 2 monitors, it crahsed, i coped /var/log/Xorg.0.log and thats it before i started x again
16:32 aaaa: this is Xorg.log.old https://bpaste.net/show/5fdaeb343143
16:32 imirkin_: aaaa: coz it doesn't have the kernel params in question
16:32 imirkin_: aaaa: it logs the kernel cmdline on start
16:33 aaaa: BOOT_IMAGE=/vmlinuz-4.2.0-1-amd64 root=/dev/mapper/root-root ro quiet nouveau.noaccel=1 nouveau.nofbaccel=1
16:33 aaaa: thats /proc/cmdline
16:33 imirkin_: aaaa: in your dmesg, yes.
16:33 imirkin_: aaaa: now look in the xorg log you pasted
16:33 imirkin_: note hwo it's not there
16:33 aaaa: hmmmmm, i see
16:34 sarnex: imirkin_: i grepped out the is_format_supported calls this time https://bpaste.net/show/d383b22cd00b
16:35 imirkin_: sorry dunno
16:35 aaaa: i will try multiple monitors again, should i then upload /var/log/Xorg.0.log?
16:35 imirkin_: those all look fine
16:35 imirkin_: you need to identify which draw is messing up
16:35 imirkin_: and what all it's drawing
16:36 sarnex: ill create an apitrace i guess
16:36 sarnex: and try to see what its doing
16:38 aaaa: nevermind, found the right log
16:39 imirkin_: sarnex: you could try nuking PIPE_FORMAT_I8_UNORM
16:40 sarnex: will do
16:40 imirkin_: sarnex: and maybe the L* variants too. but i've never had issues with them.
16:40 aaaa: nevermind, still does not have the right arguments...
16:40 aaaa: it should be /var/log/Xorg.0.log right?
16:40 sarnex: imirkin_: F3B,I3B and C4B?
16:41 imirkin_: sarnex: well, the formats defined are more important
16:41 imirkin_: those F3B things just mean like... format with 3 components, type b (which iirc means no vertex)
16:41 sarnex: ok ill just nuke all L*
16:41 aaaa: the timestamp on /var/log/Xorg.0.log is from yesterday...
16:42 imirkin_: sarnex: the issue could be something else entirely... that screenshot just reminded me of the other bug :)
16:42 sarnex: yeah
16:47 imirkin_: skeggsb: wasn't sure what your sor comment was asking me to do...
16:49 imirkin_: skeggsb: did you mean creating a gf119_sor_get_some_info() type of call?
16:50 imirkin_: skeggsb: and/or looking at 616798 + hoff?
16:50 imirkin_: [for gf119]
16:53 gryffus: imirkin_: i was able to get some apitrace of that crashing wow... where do you want me to upload it? It has only 10,5MB
16:54 imirkin_: xz -9 it, file a bug at bugs.freedesktop.org against mesa core
16:54 aaaa: imirkin_: it does not look like /var/log/Xorg*log is being created.
16:54 imirkin_: aaaa: ok
16:54 imirkin_: aaaa: well you need to convey the information one way or another
16:54 aaaa: imirkin_: any idea why?
16:56 imirkin_: aaaa: dunno, maybe you're using some weirdo system
16:56 imirkin_: aaaa: i don't do distro support... if X is crashing, happy to look at the crash log. you're on your own for finding it.
16:57 aaaa: imirkin_: they were created yesterday...i have not done anything on this system since than other than change the boot paramaters....
16:57 imirkin_: aaaa: i believe you. that doesn't change the fact that i don't do distro support.
16:58 imirkin_: aaaa: based on the dmesg log you uploaded, everything's working swimmingly at the kernel level
16:58 imirkin_: i'm a bit unhappy about those DP condition messages, but they're *probably* ok... a few people have reported them with no ill effect
16:58 sarnex: imirkin_: yeah still dies with those formats, maybe another issue
16:59 imirkin_: sarnex: oh, one last one to nuke -- A8_UNORM
16:59 sarnex: ok
17:00 sarnex: B8G8R8A8_UNORM etc dosent count right
17:00 imirkin_: sarnex: yeah, just the A8_UNORM one
17:00 sarnex: ok cool
17:03 aaaa: imirkin_: found it in .local/share/xorg/ https://dpaste.de/xKKz
17:04 imirkin_: aaaa: looks all happy. remind me what the issue is?
17:04 aaaa: X crashes
17:04 aaaa: when try to use 2 monitors
17:04 imirkin_: i don't see a crash in that log
17:05 aaaa: nor do i but its on the tty, let me type it out manualy
17:05 Niautanor: imirkin_, the patches you linked me worked perfectly by the way. Thanks again
17:05 imirkin_: Niautanor: what frequency did you end up using?
17:06 Niautanor: 225Mhz
17:06 imirkin_: Niautanor: k cool
17:07 aaaa: /usr/lib/Xorg: symbol lookup error: /usr/lib/xorg/modules/drivers/nouveau_drv.so undefined symbol: exaGetPixmapDriver
17:07 imirkin_: hmmmmmmmm
17:08 aaaa: Privatexinit: connection to x server lost
17:08 imirkin_: aaaa: here's a quick fix -- uninstall xf86-video-nouveau
17:08 imirkin_: might be called xorg-server-driver-nouveau or something for you, dunno what distros like to rename it to
17:08 imirkin_: aaaa: the modesetting ddx should be able to do everything you need just fine
17:09 sarnex: imirkin_: yeah still dead rip
17:09 sarnex: thanks for trying
17:09 imirkin_: well, good to know it's not a format thing
17:09 imirkin_: ;)
17:10 aaaa: uninstalling it now, it was under the name xserver-xorg-video-nouveau
17:12 aaaa: it works!, thank you for your help imirkin_
17:12 imirkin_: aaaa: you're welcome :)
17:12 aaaa: do you think i still need to disable accel?
17:13 imirkin_: aaaa: yes. the issue you were hitting was that you're using a kernel which doesn't init your card's accel units properly
17:13 imirkin_: aaaa: the secondary issue is that apparently the nouveau DDX doesn't handle the case where accel has been disabled too well in certain cases
17:14 aaaa: imirkin_: so if im not using nouveau what am i using?
17:14 imirkin_: skeggsb: does drm_detect_hdmi_monitor detect that a monitor has been plugged into HDMI? it seemed like it just poked at some bits in the EDID...
17:14 imirkin_: aaaa: you're using the 'modesetting' ddx, which is a hardware-agnostic ddx
17:15 aaaa: imirkin_: is that just framebuffer?
17:15 airlied: imirkin_: it's just for detecting if it's a hdmi monitor plugged in
17:15 airlied: not what it's plugged into
17:16 imirkin_: airlied: right, so it could just as well be plugged into DVI or DP right?
17:16 imirkin_: [not in those cases since we check for TMDS first, but in general]
17:16 skeggsb: imirkin_: well, the edid is associated with a particular connector..
17:16 imirkin_: skeggsb: this is what you want, right? if (pclk >= 165000 && (ctrl & 0xf) == 5)
17:16 airlied: yes a DVI or DP->BLAH connector I assume
17:16 skeggsb: imirkin_: nope, drop the pclk bit completely from those places
17:17 imirkin_: k
17:17 imirkin_: if ((ctrl & 0xf) == 5)
17:17 imirkin_: yay?
17:17 skeggsb: yep
17:17 aaaa: imirkin_: and is this specific to the package that my distro provides? should i just compile them myself?
17:18 skeggsb: er, well, (ctrl & 0xf00) == 0x500, but, yeah :P
17:18 imirkin_: aaaa: if you compile a drm-next kernel, you should be good.
17:18 imirkin_: aaaa: http://cgit.freedesktop.org/~airlied/linux/log/?h=drm-next
17:18 imirkin_: skeggsb: errrr.... i see. so should i just do *conf == 5?
17:19 skeggsb: either way, it makes no difference, whatever you think looks prettiest :P
17:19 imirkin_: *conf == 5 has fewer parentheses
17:19 aaaa: well thankyou for your help
17:19 imirkin_: i think i'll go with that
17:20 imirkin_: skeggsb: ok, so should i use detect_hdmi_monitor() in nv50_sor_mode_set?
17:20 skeggsb: yeah, you should be able to use it on nv_connector->edid and get the right result
17:21 imirkin_: so i want to set it to 1 if hdmi monitor || rate < 165
17:21 skeggsb: or wherever the edid lives from nv_connector.. it's there somewhere ;)
17:21 imirkin_: hehehe
17:21 sarnex: imirkin_: any idea whats going on here https://i.imgur.com/LWSB3DY.jpg
17:22 skeggsb: you basically want: if (!hdmi && rate > tmds_limit) proto = DUAL_TMDS
17:22 imirkin_: oh this is so going to backfire
17:22 imirkin_: if (nv_connector->edid &&
17:22 imirkin_: !drm_detect_hdmi_monitor(nv_connector->edid) &&
17:22 imirkin_: but... what if there's no edid? no dual-link?
17:22 imirkin_: i guess i can live with that
17:22 imirkin_: thoughts?
17:23 skeggsb: there's *supposed* to always be one for dvi/hdmi, but, i guess the sane fallback is to assume !hdmi if !edid
17:23 imirkin_: sarnex: wrong stride set somewhere...
17:24 sarnex: first let me remove the format patch and see if i still get it
17:24 imirkin_: skeggsb: ok
17:30 sarnex: imirkin_: yeah still happens, is the problem in mesa? if not nvm
17:30 imirkin_: sarnex: ?
17:30 sarnex: the picture i posted still happens if i use mesa git vs mesa -weird formats
17:31 skeggsb: imirkin_: did you figure out anything better for the glamor serialize patch?
17:32 imirkin_: skeggsb: didn't try yet
17:32 imirkin_: skeggsb: too many broken things, glamor is the least of my worries :p
17:33 imirkin_: sarnex: ok... i have no idea what the context is.
17:33 sarnex: ok nvm
17:34 skeggsb: imirkin_: well, it's not strictly a glamor bug :P just one thing we know hits it hehe
17:34 skeggsb: but, ack :)
17:34 imirkin_: skeggsb: maybe. i *think* what glamor is doing is fine, but i can't think of why serialize would help at all there
17:34 imirkin_: hmmmmmmmmm
17:34 imirkin_: i wonder if this happens when it rotates buffers....
17:35 imirkin_: skeggsb: if (nouveau_duallink && nv_encoder->dcb->duallink_possible)
17:35 imirkin_: max_clock *= 2;
17:35 imirkin_: probably not great to have there :)
17:39 imirkin_: skeggsb: so your claim is that if you plug a monitor with DVI and HDMI outputs over DVI, it will return false for drm_detect_hdmi_monitor ?
17:43 airlied: it better
17:44 imirkin_: ok cool. i thought 1 monitor = 1 edid
17:44 imirkin_: i'm sure that's so 1990 :)
17:45 imirkin_: airlied: what about connecting HDMI monitor input to DVI gpu output?
17:45 airlied: you think monitors with VGA and DVI returned a single EDID
17:45 imirkin_: i think DVI didn't exist in 1990
17:46 airlied: DVI link speed limitations are mostly in the cable I think
17:46 airlied: maybe the connectors
17:46 imirkin_: i get that
17:46 imirkin_: i'm talking about which edid we'd get
17:46 imirkin_: by hooking a DVI gpu output into an HDMI monitor input
17:46 airlied: you get the EDID from the port
17:47 airlied: so that would look like a HDMI monitor
17:47 imirkin_: so we'd think it was an hdmi monitor
17:47 imirkin_: ok
17:47 imirkin_: skeggsb: anything to worry about?
17:48 skeggsb: no, generally even the DVI-shaped ports are HDMI-capable
17:48 imirkin_: yeah, but not the higher speeds right?
17:48 skeggsb: if you want to be strictly correct, there is a bit in the DCB that says whether that's true or not
17:48 imirkin_: due to connector/cable issues
17:48 skeggsb: i suspect it'd be fine
17:48 imirkin_: we respect that dcb bit
17:48 imirkin_: that's the dcb->can_dual_link right?
17:48 skeggsb: nope, there's a "hdmi supported" bit too
17:48 skeggsb: which we don't use
17:49 imirkin_: yeah, but that says nothing about the cabling...
17:49 imirkin_: presumably a regular DVI cable + connectors can't do the higher speeds
17:49 imirkin_: even if you stick a little hdmi thingie at the end of it
17:49 skeggsb: not much we can do about that really, i'm sure the connectors will be fine if DCB allows it
17:49 skeggsb: and, it's the user's own fault if they use a cable that can't
17:49 imirkin_: ;)
17:50 imirkin_: yes, i'm sure users are inteimately familiar with the various bw-related issues that result in them not getting a terminal
17:50 imirkin_: since we pick the "best" mode
17:50 skeggsb: istr that the earlier hdmi-capable nvidia gpus were via an adaptor
17:50 airlied: you have to go out of your way to make that happen
17:50 airlied: like get a DVI->HDMI adapter on the monitor end
17:50 imirkin_: yeah
17:51 imirkin_: when there's a will, there's a way ;)
17:51 skeggsb: which brings us back to "their own fault" :)
17:51 airlied: yes but that's just a better idiot detector
17:54 imirkin_: skeggsb: v2 sent... probably could stand to use some testing
17:54 imirkin_: i have none of the hw in question
17:56 imirkin_: oh fun. derm_detect_hdmi_monitor can take a null
17:56 skeggsb: imirkin_: you can drop the !nv_connector->edid part it seems, the drm does that
17:56 skeggsb: hah, good timing :P
18:00 imirkin_: all better?
18:00 skeggsb: yeah, i'll merge those
18:01 imirkin_: there appears to have been an influx of people coming in here with that problem
18:01 imirkin_: no idea why all of a sudden
18:05 skeggsb: guessing the higher resolution monitors are becoming a lot more accessible
18:05 imirkin_: probably
20:14 sarnex: is there a way to get the current temp?
20:14 imirkin_: run 'sensors'
20:14 imirkin_: look for nouveau-pci-NNNN
20:14 sarnex: oh thanks
20:14 sarnex: 51C reclocked not bad
20:15 imirkin_: oh so reclocking worked for you? neat
20:15 imirkin_: did the core clock up and everything?
20:15 sarnex: imirkin_: well how can i tell, i see pstate
20:15 imirkin_: cat the pstate file, compare the AC: line with the level you wanted to be at
20:16 imirkin_: ignore the * thing -- that's just there to mislead you
20:17 sarnex: hmm says core clock 405mhz website says it should be 905
20:17 sarnex: so i guess it doesnt work
20:17 imirkin_: pastebin the pstate file
20:17 sarnex: https://bpaste.net/show/afce16143daf
20:17 imirkin_: you're on the low pstate
20:18 sarnex: is there a way to change it or reclocking doesnt work for this card
20:18 imirkin_: you need to 'echo 0f > path-to-pstate' if you want to switch
20:18 sarnex: oh
20:18 imirkin_: like i said, it's not automatic
20:18 sarnex: oh wow it says 902 now
20:18 sarnex: so it works
20:18 imirkin_: cool
20:18 imirkin_: does memory go up too?
20:18 sarnex: yes doubles
20:19 imirkin_: so now it matches the 0f line right?
20:19 imirkin_: (pretty much)
20:19 sarnex: https://bpaste.net/show/b527da4195a8
20:19 imirkin_: excellent.
20:19 imirkin_: everything works as intended
20:19 imirkin_: should give you a nice perf boost
20:19 imirkin_: experience suggests you'll get 60-80% of blob speed, depending on the sepcific application
20:21 sarnex: 53c relocked too so temps are good
20:21 sarnex: great work everyone
20:21 imirkin_: well, start running glxspheres and you'll see temp go up :)
20:22 imirkin_: keep in mind that your thing is fanless, and we don't automatically clock down or anything
20:22 imirkin_: so... be careful ;)
20:24 sarnex: 58c doing 354.459649 frames/sec - 395.576969 Mpixels/sec
20:25 sarnex: with vblank_mode=0
20:25 sarnex: pretty good assuming the temps are anything like modern cards that can do 80c stable
22:46 pq: imirkin, yeah, unhandled x86 instruction in mmiotrace is a good guess, but IIRC it might be also if the same instruction accesses more than one ioremapped address
22:46 imirkin: ah yeah
22:46 imirkin: like a rep movsb?
22:46 pq: imirkin, I've forgot almost everything of mmiotrace, but I have a feeling you might get in trouble if there's e.g. a mov from one ioremapped addr to another
22:47 pq: imirkin, hmm, rep shouldn't be fatal problem - either the cpu single-steps each rep iteration or it doesn't (you lose data), but I don't recall it could cause a fatal issue
22:48 imirkin: right, the rep wasn't the main bit of it, more the movsb :)
22:48 pq: oh, ok, I don't really know any asm :-p
23:02 imirkin: skeggsb: what do you think about doing the c800 workaround on all mobile chips?
23:02 imirkin: skeggsb: iirc there's a "mobile" flag in the vbios
23:04 skeggsb: imirkin: i'm probably going to just bite the bullet and flip it on by default in 4.4, see what happens during the -rc
23:05 imirkin: skeggsb: probably only for mobile...
23:05 skeggsb: well, i'm pretty sure every trace i've looked at, nvidia do it
23:05 skeggsb: it's not like they're different chips...
23:05 imirkin: i thought you said you found boards where it broke things
23:06 skeggsb: no, i said screwing it up broke things, and i wasn't confident enough that it was the same everywhere, i'm pretty sure what's implemented is ok.. maybe.. possibly
23:06 skeggsb: i've been trying to get that confirmed... buuuut yeah
23:09 skeggsb: imirkin: the earlier code only did the gk104 version of the writes, i've looked at a heap of traces now and only see the three variants we do
23:09 imirkin: ah ok
23:09 imirkin: GK208 doesn't need anything?
23:10 skeggsb: nope, i'm guessing that got fixed
23:10 skeggsb: i did check. well, on gk110, but, close enough
23:12 imirkin: haha
23:12 imirkin: oh, did you see someone having trouble with a GK110?
23:12 skeggsb: nope? i have two of them that worked last i checked
23:13 imirkin: skeggsb: https://bugs.freedesktop.org/show_bug.cgi?id=92761
23:14 skeggsb: oh, i did see that one go by
23:14 skeggsb: i'm somewhat suspicious of "GPC4"
23:15 imirkin: yeah
23:15 imirkin: only high GPC numbers
23:15 skeggsb: my bet is that we're reading the GPC count from the wrong place, and it doesn't exist
23:18 imirkin: sounds reasonable
23:18 imirkin: i bet that guy would nvapeek if you provided the relevant commands
23:20 imirkin: oh, and unrelatedly, someone confirms that the hdmi patches i sent worked -- nice to know that i didn't mess it up :)
23:22 skeggsb: i noticed that :)