Can the GeForce RTX 5070 run Mistral Small 3.2 24B Instruct (2506)?
Only with CPU offload at Q4_K_M.
Mistral Small 3.2 24B Instruct (2506) is 24B parameters. At Q4_K_M the weights are about 13.4 GiB; with the 1.5 GiB runtime reserve that is 14.9 GiB against the card's 12GB, about 2.9 GiB more than the card has.
All three quants
12GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 13.4 GiB | 14.9 GiB | Only with CPU offload | short 2.9 GiB |
| Q8_0near-lossless | 23.7 GiB | 25.2 GiB | Does not fit | short 13.2 GiB |
| FP16full weights | 44.7 GiB | 46.2 GiB | Does not fit | short 34.2 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
Not this card. These fit it comfortably at Q4.
- GeForce RTX 5090 32GB · headroom 17.1 GiB
- GeForce RTX 3090 (used) 24GB · headroom 9.1 GiB
- GeForce RTX 4090 (used) 24GB · headroom 9.1 GiB
- Radeon RX 7900 XTX 24GB · headroom 9.1 GiB
What the GeForce RTX 5070 does run well → · Mistral Small 3.2 24B Instruct (2506) on every card →
About the model
Mistral Small 3.2 24B Instruct (2506) by Mistral AI: 131,072-token native context, text + vision, released 2025 under the Apache-2.0. Model card → · Hosted pricing on PicksByModel →