Can I Run Devstral Small 2 on a NVIDIA RTX 3090?
Runs at full precision (fp16). Zero quality loss.
13 quantizations fit your 24.0GB
| Quant | Min VRAM | Recommended | File size | Headroom |
|---|---|---|---|---|
| fp16BEST | 15.0 GB | 16.5 GB | 0.9 GB | +9.0 GB |
| Q8_0 | 8.4 GB | 9.9 GB | 25.1 GB | +15.6 GB |
| Q6_K | 6.8 GB | 8.3 GB | 19.4 GB | +17.2 GB |
| Q5_K_M | 6.0 GB | 7.5 GB | 16.8 GB | +18.0 GB |
| Q5_K_S | 5.8 GB | 7.3 GB | 16.3 GB | +18.2 GB |
| Q4_1 | 5.4 GB | 6.9 GB | 14.9 GB | +18.6 GB |
| Q4_K_M | 5.2 GB | 6.7 GB | 14.3 GB | +18.8 GB |
| Q4_K_S | 5.0 GB | 6.5 GB | 13.6 GB | +19.0 GB |
| Q4_0 | 4.9 GB | 6.4 GB | 13.5 GB | +19.1 GB |
| Q3_K_L | 4.1 GB | 5.6 GB | 12.4 GB | +19.9 GB |
| Q3_K_M | 3.9 GB | 5.4 GB | 11.5 GB | +20.1 GB |
| Q3_K_S | 3.7 GB | 5.2 GB | 10.4 GB | +20.3 GB |
| Q2_K | 3.3 GB | 4.8 GB | 8.9 GB | +20.7 GB |
Try it in the cloud first
Don't want to download Devstral Small 2 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 24.0GB.
FAQ
Can the NVIDIA RTX 3090 run Devstral Small 2?
Yes. The NVIDIA RTX 3090's 24.0GB of VRAM is enough to run Devstral Small 2 at fp16 quantization (15.0GB required).
What's the best quantization to use?
fp16 is the highest-precision quantization that fits in your 24.0GB. It uses about 15.0GB of memory and 16.5GB 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.