can it run
Yes, but close to the edge. 24GB against a 19GB working set: it runs, with little room for long context windows or other processes.
weights 14.6GB + KV cache 0.62GB at 4k context vs ~24GB usable → full. 15.2 GB needed of 24 GB usable — headroom for context
computed roofline: 1008 GB/s × 0.46–0.76 efficiency window / 14.6GB (Q4_K_M) — bands, never points.
| model file | 14.6GB (GGUF Q4_K_M) |
| minimum memory | 19GB working set |
| this machine | 24GB GDDR6X |
| params | 24B |
| license | apache-2.0 |
| price | $2000 |
easy run: ollama pull devstral:24b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ Tesla V100 32GB (used) — cheapest machine that runs it ($700)
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine
→ rtx-5090 — the next step up from this machine's GPU
eBay (used) ↗ · Amazon (new) ↗