Can I Run Devstral Small 2 24B Instruct 2512 on a NVIDIA DGX Spark?
Runs at full precision (f32). Zero quality loss.
14 quantizations fit your 128GB
| Quant | Min VRAM | Recommended | File size | Headroom |
|---|---|---|---|---|
| f32BEST | 97.0 GB | 98.5 GB | 1.8 GB | +31.0 GB |
| fp16 | 49.0 GB | 50.5 GB | 0.9 GB | +79.0 GB |
| Q8_0 | 26.5 GB | 28.0 GB | 25.1 GB | +101.5 GB |
| Q6_K | 20.8 GB | 22.3 GB | 19.4 GB | +107.2 GB |
| Q5_K_M | 18.0 GB | 19.5 GB | 16.8 GB | +110.0 GB |
| Q5_K_S | 17.6 GB | 19.1 GB | 16.3 GB | +110.4 GB |
| Q4_1 | 16.0 GB | 17.5 GB | 14.9 GB | +112.0 GB |
| Q4_K_M | 15.6 GB | 17.1 GB | 14.3 GB | +112.5 GB |
| Q4_K_S | 14.7 GB | 16.2 GB | 13.6 GB | +113.3 GB |
| Q4_0 | 14.5 GB | 16.0 GB | 13.5 GB | +113.5 GB |
| Q3_K_L | 11.7 GB | 13.2 GB | 12.4 GB | +116.3 GB |
| Q3_K_M | 11.1 GB | 12.6 GB | 11.5 GB | +117.0 GB |
| Q3_K_S | 10.2 GB | 11.7 GB | 10.4 GB | +117.8 GB |
| Q2_K | 8.9 GB | 10.4 GB | 8.9 GB | +119.1 GB |
Try it in the cloud first
Don't want to download Devstral Small 2 24B Instruct 2512 just to try it? Use a hosted API or rent a GPU by the second.
Affiliate links — we earn a commission at no cost to you.
All quant variants, benchmark scores, and use-case tags.
Top-ranked open-source models that fit in 128GB.
FAQ
Can the NVIDIA DGX Spark run Devstral Small 2 24B Instruct 2512?
Yes. The NVIDIA DGX Spark's 128GB of unified memory is enough to run Devstral Small 2 24B Instruct 2512 at f32 quantization (97.0GB required).
What's the best quantization to use?
f32 is the highest-precision quantization that fits in your 128GB. It uses about 97.0GB of memory and 98.5GB recommended for comfortable inference.
What if I need more headroom for context length?
KV cache memory grows with context length. The numbers above assume a baseline 2K-4K context. For long-context use (32K+), add another 2-6GB depending on the model architecture.