If you're unsure what process to kill (process doesn't show up with nvidia-smi):

1. Use nvtop to find PIDs of the dead processes still hogging VRAM and which device index (use nvtop because nvidia-smi may have filtered it)

2. Check fuser -v /dev/nvidia<device index> to find user (change device index to relevant integer)

3. Use htop -u <user> and kill processes that seem to have hanged

This is a complement to [Kenan's answer](https://stackoverflow.com/a/46597252/15399131).