What could change the answer
- Long context and KV cache
- Batch size or concurrent users
- Runtime overhead and driver support
- CPU/RAM offload
- Image encoders and auxiliary models
MODEL × HARDWARECan RTX 5080 16GB
Can RTX 5080 16GB
run Mistral Small 24B?
Only with optimization or offload.
Q4_K_M MEMORY VERDICTTIGHT
18.3 GB estimated requirement vs 16.0 GB usable-memory assumption.
Model weights14.6 GB
Runtime reserveIncluded
Hardware stackCUDA
Speed claimNot estimated