Thanks for the update. I haven't seen any other bug report against the nvme-pci driver which show similar symptoms. If default_ps_max_latency_us workaround yields positive results we could try to get a quirk added to the kernel. Though I don't know if this possible though without starring at the code first :)