I have several machines running Ubuntu 16.04, with latest nvidia drivers 375.51, with dual 1080 Ti, or dual 970 GTX. I use GPUs for computations, and two monitors are attached to one of them. When both monitors are 4K, after the screen turns off due to inactivity, it doesn’t wake up. After this freezup, restarting lightdm may help, or may not (screens remain blank) Also, sometimes nvidia-smi stops working after the freezup.
The problem happens on every desktop I’ve ever had with dual 4K setup. It happened on earlier driver versions. It does not happen with two HD monitors, or with one HD and one 4K monitor.
Attached is a nvidia debug log during the frozen state.
The problem is accompanied by the following record in kern.log.
May 25 12:37:56 wiz kernel: [1038599.126144] snd_hda_codec_hdmi hdaudioC3D0: HDMI: invalid ELD data byte 45
May 25 12:37:57 wiz kernel: [1038599.618183] snd_hda_codec_hdmi hdaudioC3D0: HDMI: invalid ELD data byte 16
May 25 12:37:57 wiz kernel: [1038599.702177] snd_hda_codec_hdmi hdaudioC3D0: HDMI: invalid ELD data byte 0
May 25 12:37:58 wiz kernel: [1038600.856403] NVRM: GPU at PCI:0000:81:00: GPU-8d74d21a-eeae-dba2-6016-6e1b6d8202bf
May 25 12:37:58 wiz kernel: [1038600.856407] NVRM: GPU Board Serial Number: 0321117032013
May 25 12:37:58 wiz kernel: [1038600.856409] NVRM: Xid (PCI:0000:81:00): 31, Ch 00000010, engmask 00000101, intr 10000000
< The last line then repeats ~80 times>
nvidia-bug-report.log.gz (371 KB)