Some basic information, like kernel version used would be nice. It works fine for me with 4.3.3.
I'm using kernel 4.1.15, xorg 1.17.4 and Chromium 48.0.2564.82 (64-bit).
Hi petterg and Intel forum followers,
Do anyone already work with X.org 1.17.4 Server OS on NUCs? Intel has not tested yet this operating system on NUCs.
I'm having this exact issue and it's driving me mad. This is persistent on kernel 4.2, 4.3, 4.4 and 4.5-rc1 with (as far as I know) all drivers installed.
Strange, I don't have any such problem with 4.3.3. Do you have the GUC and DMC firmware installed (under /lib/firmware/i915 on my system)?
I'll check when I get back home, but I did install the skl_guc_ver43.bin into that folder. Didn't change anything though.
I will try to post a more specific breakdown of what I have installed and not. It would be great if this just came down to some missing drivers and/or files. Otherwise I might be affected by the Corsair Value ram thing that has bugged a lot of users. This seems oddly specific to GPU acceleration though...
Do you remember what you installed on top of vanilla Ubuntu 15.10?
I use Arch Linux, not Ubuntu. Did you install the firmware using the script? The script also sets some symlinks, which I guess are needed for the FW to be loaded.
Anyway, I ran into a different problem with SKL graphics: Video playback with mpv using vaapi hw decoding (tried vaapi and vaapi-copy) and vaapi or opengl output all lead to hangs, massive frame drops and even complete hard lockups of the whole system. I was able to work around these issues by setting i915.enable_rc6=0 on the kernel command line. I tried some less extreme measures (like disable_power_well=0) but no other fixed the issue for me. Seems the Linux drivers have indeed quite some issues, but maybe disabling rc6 (low power gfx c-state) helps in your case, too.
This will cause the system to use more power in idle, but by my measurements using turbostat it's not that much more.
[36945.207535] [drm] stuck on render ring
[36945.207943] [drm] GPU HANG: ecode 9:0:0x85dfbfff, in chrome , reason: Ring hung, action: reset
[36945.210087] drm/i915: Resetting chip after gpu hang
[36947.208026] [drm] RC6 on
Is there a work-around this as it makes it impossible to use Chrome since it will stop for several seconds every few seconds?
There is an advice from ArchLinux wikia:
Note: The latest generation of integrated GPUs (Skylake/HD 530 for instance) may require the i915.preliminary_hw_support=1 kernel parameter to boot properly. As of kernel 4.3.x, this should no longer be necessary.
Maybe this will help
I've been suffering from this exact problem as well on the NUC6i5SYK + Arch Linux. Multiple GPU hangs that can even result in complete system freeze. For me it reproduces quickly when any browser (Firefox, Chromium) tries to do WebGL intensive work.
Right now the only workaround seems to be to disable hardware rendering, either on Xorg or direclty in the relevant application.
It seems several bugs are already open on similar issues: Bug List
I'm also trying to keep the Arch Wiki updated on these issues so please do share any further knowledge on this issue.
Can you try the work-around which works for me (at least for video playback)? i915.enable_rc6=0 on the kernel command line. I didn't have problem with Chrome or Firefox, but I also don't use WebGL, AFAIK.
Definitely, I'll try this later today and report back.
So that kernel options seems to do *something* since now Chromium can run WebGL stuff without crashes!
Firefox however has no WebGL support at all with i915.enable_rc6=0 and doesn't have hardware acceleration either. So, partial success, and it's good to have another workaround, but a double edged sword.
i915.enable_rc6=0 cannot possibly disable Firefox WebGL support. All it does is disable support for the low power RC6 state in the kernel's DRM driver. It does absolutely nothing with regard to GL support or any other feature support. So either this did not work before or Firefox support is flakey.
RC6 support was disabled by default for a long time on (then new) Ivy Bridge graphics, because it caused issues (hangs etc.) and all that disabling did was increase idle power consumption.
You're right, this was just Firefox activating a crash guard. Apparently i915.enable_rc6=0 is a good workaround for this specific problem!