System:
Truenas Scale 25.04.2.4
RX 7900 XTX
Core Ultra 9 285k
W880 MB
Issue
Ollama experiences very slow model loading time even when running on a fast nvme. E.g. for a 32B model it takes over 20s to load. Just towards the end of loading the model I get the printout attached in the truenas console, suggesting there is something wrong with the way the models are loaded.
Tried
- Installed fresh ubuntu 24.04 on the same hardware (no vm, no truenas, just ubuntu on nvme), and ollama runs like a charm.
- Some online search suggests updating the bios, which didn’t help
- Deploying docker container in a portainer truenas app instead, same issue
Not tried
- messing with the driver on the host system, don’t want to break the install and not really sure what I am doing.
