Ollama library
Devstral Small 2 24B
Ollama library·24B·Transformer·384K ctx·2026-08
Model size→Hardware requirement→Actual fit
Devstral Small 2 24B needs ~16.2–29.8 GB of VRAM depending on quantization. The bar is what it needs; the marker is the card's VRAM. Speed figures below are planning estimates.
8 GB
RTX 4060 · 3060 Ti
16.28
Q4_K_M · Est. ~3 tok/sSlow · CPU offload
12 GB
RTX 4070 · 3060 12G
16.212
Q4_K_M · Est. ~5 tok/sSlow · CPU offload
16 GB
RTX 4080 · M-series 16G
16.216
Q4_K_M · Est. ~8 tok/sSlow · CPU offload
24 GB
RTX 4090 · 3090
18.924
Q5_K_M · Est. ~66 tok/sComfortable
48 GB
A6000 · 2×24G
29.848
Q8_0 · Est. ~37 tok/sComfortable
02 — Quantizations
QuantFile sizeEst. VRAM
Q4_K_M15 GB16.2 GB
Q5_K_M17.7 GB18.9 GB
Q8_028.6 GB29.8 GB
03 — Runtimes & use
llama.cpp · Ollama · LM Studio
General
Memory and speed figures are planning estimates, not measured benchmarks. Memory includes a fixed allowance for KV cache and runtime overhead. Long-context prompts push VRAM use above the numbers shown here.