So I got this "NVMe" VPS from Hetzner for my inference side project right, and im running llama.cpp on it. Thing is HORRIBLY slow for the model size, like 2 tok/sec when I expected 15+. Check df -T and it says ext4 on /dev/sda1. Not nvme0n1. Not even virtio. Just straight sda.
Screenshot: https://imgur.com/fake-link-123
I opened a ticket and they said "the disk is NVMe-backed" but like... that could mean anything. Anyone know how to actually verify what im getting? Ngl this feels scammy.
CUDA cores are my love language