Okay, I take that back: It also happens in Chromium Version 63.0.3239.108 but less often. I can get it to happen in Netflix after a few seconds (whole tab flickers) with this output in the console:

[18730:18730:0203/133945.997882:ERROR:gles2_cmd_decoder.cc(16234)] Offscreen context lost via ARB/EXT_robustness. Reset status = GL_GUILTY_CONTEXT_RESET_KHR
[18730:18730:0203/133945.997932:ERROR:gles2_cmd_decoder.cc(4394)]   GLES2DecoderImpl: Context reset detected after MakeCurrent.
[18730:18730:0203/133945.997962:ERROR:gpu_channel_manager.cc(204)] Exiting GPU process because some drivers cannot recover from problems.
[18730:18730:0203/133945.998009:ERROR:gpu_channel_manager.cc(204)] Exiting GPU process because some drivers cannot recover from problems.

I tried unstable Version 65.0.3325.31 (Offizieller Build) dev (64-Bit); same results.

--disable-gpu on the commandline fixes the problems. The attached file is the page chrome://gpu/

Deleted: gpu-chrome-64.html 123 KB

gpu-chrome-64.html 123 KB View Download

I checked my systemd journal and found these:

kernel: NVRM: Xid (PCI:0000:01:00): 31, Ch 00000001, engmask 00000101, intr 10000000

right when the NVIDIA driver is loaded and then many of these in roughly 10 second intervals:

kernel: NVRM: Xid (PCI:0000:01:00): 8, Channel 00000008

https://docs.nvidia.com/deploy/xid-errors/ says:

Error 31: GPU memory page fault
Error 8: GPU stopped processing

Error 31 could indicate a hardware failure, so I tried a few GL benchmarks. They looked good. Weird. I'll do a reboot with power off, now.

Seems to be a one-time glitch. Reboot fixed it.

My system is now stable again. I think it's was a one time glitch or maybe aging hardware.

Suggested fix: When the GPU process fails several times, how about deactivating it automatically plus giving the user a notification about the problem. Then the browser would feel more stable and I could continue to use it to google for a solution :-)

Thanks for filing the issue!

As per comments#5 and #7 by reporter it's clearly understood that the issue has been resolved. Requesting some one from Blink>WebGL team to have a look into this and respond for comment#7 regarding the notifications. 	

We disable the GPU after 3 crashes already. I don't think there's anything further we can/should do.

As I said, I had a ton of errors, ugly UI flashes and the browser was hanging for a moment every now and then. So something in the "disable after 3 crashes" isn't working in this case.

Can you send us the complete contents of chrome://gpu after you've seen this issue? If you can repro easily, it would be good to copy it after you've seen it crash more than 3 times in one session.

In particular, there's a "GPU process crash count" field and I'm wondering if it's really going above 3, or if we're failing to increment it in this case.

See the file attached in comment #3. The crash count is 0. For some reason, "Exiting GPU process because some drivers cannot recover from problems." doesn't count as a crash. It's 6 times in the log followed by "The GPU process exited normally. Everything is okay."

I think you should count unrecoverable errors as "crashes", too.

Ah, sorry, I tried to find that file but I missed it.

From a security standpoint, I'm not it's necessary to treat GL_GUILTY_CONTEXT_RESET_KHR as a crash. But just to improve user experience, we probably should.

I'm not sure* it's necessary

Since this doesn't seem to be affecting many users, this is P3 for now.