Yes — with 31.1 GB to spare
Devstral Small 2 24B at Q4_K_M fits your L40S entirely on the GPU at 8K context, at an estimated 39 tokens per second. Past 206K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully on GPU
8K context
Q4_K_M · 13.5 GB
Apache 2.0
Released Dec 2025
Vision
Not in the Ollama library
Built for software-engineering agents (OpenHands, Cline). Dense 24B, 384K window. Not in the Ollama library.
The VRAM budget
weights 13.5 GB
Weights 13.5 GB
KV cache @ 8K 1.25 GB
Runtime overhead 0.6 GB
Free 31.1 GB of 46.4 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 23.7 GB | 25.6 GB | 141K | 22 | −0.1% ppl | Long context |
| Q6_K | 18.3 GB | 20.2 GB | 175K | 29 | −0.4% ppl | Long context |
| Q5_K_M | 15.8 GB | 17.7 GB | 191K | 33 | −0.8% ppl | Long context |
| Q4_K_M | 13.5 GB | 15.3 GB | 206K | 39 | −1.9% ppl | Recommended |
| Q3_K_M | 10.9 GB | 12.8 GB | 223K | 48 | −5.4% ppl | Long context |
| Q2_K | 9.4 GB | 11.2 GB | 233K | 56 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ llama-server \
-hf mistralai/Devstral-Small-2-24B-Instruct-2512:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 13.5 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03There is room to go to 206K context on this card.