352.41 X server crashes on resume from suspend with Dell P2715Q in power-save mode

I have an ASUS Strix Geforce GTX 960 GPU connected to a Dell P2715Q
UHD monitor. Everything works great except when resuming from
suspend. For resume to succeed, the monitor must be fully powered on.
If the monitor is in power-save mode or is entirely powered off,
resume fails.

The failure mode is that the X server crashes with the error you’ll
see in the log. The monitor does power up properly at a Linux console,
and the server can be restarted using startx.

This bug has been reported against Debian Linux at
and by another user on Ubuntu at

I believe I’ve confirmed through many days worth of experiments that
the sole predictor of the failure is the state of the monitor on
startup. If the monitor is fully powered on, resume always succeeds.
If it is not, resume always fails. Factors that do not predict
success or failure include the number of suspends since boot or the
time elapsed since the last suspend. (Please note however that a very
rapid resume is likely to succeed because the monitor takes time to
transition to power-save mode.)

Also note that on a failed resume it is only the X server that
crashes. Nothing goes wrong with the rest of the system. Also, even
on a failed resume, the GPU successfully instructs the monitor to
power on—just not in time to keep X from crashing.

About the monitor: the Ubuntu bug report mentions the Dell P2415Q.
These two monitors are siblings; they are both 16:9 UHD monitors
differing only in the size of the display panel. One assumes they have
all the same electronics inside.

I have tried to attach the relevant logs and also the nvidia-bug-report.log.gz
that I have generated. This information is also duplicated at

Xorg.0.log-before-suspend.txt (20 KB)
Xorg.0.log-after-failed-suspend.txt (20.9 KB)
nvidia-bug-report.log.gz (83.1 KB)

The log files before and after suspend, as well as the
nvidia-bug-report.log.gz, are at http://www.cs.tufts.edu/~nr/nvidia-report-2015-11-20/

Do you have any workaround? The problem persists for me in 367.44-3