Recently some of our workstations at the office have been experiencing odd memory related crashes. Some of the resulting events explicitly call out a lack of video RAM. But when I try to profile RAM usage by the app it doesn’t use that much VRAM at all … at least according to nvidia-smi. But I’m seeing error messages in the crash log about ‘memory page eviction’ which, again, points to a lack of VRAM.
I think the missing link here is knowing how much system memory the nvidia WDDM driver is trying to use. Does it crash when trying to allocate any memory from system RAM, or is it exceeding the maximum allocated (dedicated VRAM + available system VRAM?)
Any suggestions on how to troubleshoot or monitor this?