Yes — with 80.7 GB to spare
Devstral Small 2 24B at Q4_K_M fits your Ryzen AI Max+ 395 · 128 GB entirely in unified memory at 8K context, at an estimated 11 tokens per second. There is room for its full 384K window.
Fully in unified memory
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 80.7 GB of 96.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 23.7 GB | 25.6 GB | 384K | 6.0 | −0.1% ppl | Long context |
| Q6_K | 18.3 GB | 20.2 GB | 384K | 7.8 | −0.4% ppl | Long context |
| Q5_K_M | 15.8 GB | 17.7 GB | 384K | 9.0 | −0.8% ppl | Long context |
| Q4_K_M | 13.5 GB | 15.3 GB | 384K | 11 | −1.9% ppl | Recommended |
| Q3_K_M | 10.9 GB | 12.8 GB | 384K | 13 | −5.4% ppl | Long context |
| Q2_K | 9.4 GB | 11.2 GB | 384K | 15 | −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.
02macOS caps what the GPU may wire down at about 96.0 GB of 128 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to the model's full 384K context on this card.