Mistral-Small-3.1-24B-Instruct on NVIDIA RTX 4090 (24GB VRAM)
Full FP16 (51.35 GB) exceeds 24 GB VRAM, but 4-bit GGUF Q4_K_M / AWQ requires 16.9 GB (14.99 GB weights + 1.25 GB KV cache at 8K context), delivering ~43 tokens/sec.
🎛️ Interactive VRAM & KV-Cache Tuner
📊 Real-Time Memory Footprint
16.9 GB / 24 GB🚀 Ready-to-Run Launch Command
ollama run mistral-small:24b # Or with llama.cpp server (8K context, Flash Attention): llama-server -hf mistralai/Mistral-Small-3.1-24B-Instruct-2503-GGUF -c 8192 -ngl 99 -fa
Need 16.9 GB+ VRAM to run Mistral-Small-3.1-24B-Instruct?
Recommended cloud setup for 4-bit: RTX 4090 (24GB) (or 51.35 GB for full FP16). Spin up a pre-configured vLLM / Ollama instance in 60 seconds:
| Provider | Best For | RTX 4090 (24GB) | A100 (80GB) | H100 (80GB) | Bonus / Trial | Action |
|---|---|---|---|---|---|---|
| RunPod | 1-Click vLLM / Ollama Pods | $0.44 / hr | $1.39 / hr | $2.39 / hr | Instant PyTorch / vLLM Pods + Serverless GPU | Deploy on RunPod → |
| Vast.ai | Cheapest Spot GPU Market | $0.29 / hr | $0.95 / hr | $1.79 / hr | Save 50–70% on RTX 4090 / 5090 / A100 rentals | Check Spot GPUs → |
| Vultr Cloud GPU | Enterprise SLA & Bare Metal | — | $1.79 / hr | $2.69 / hr | $300 Free Credit for New Users | Claim $300 Credit → |
| DigitalOcean GPU Droplets | Managed H100 Droplets | — | — | $2.99 / hr | $200 Free Trial (60 Days) | Get $200 Free → |
All Quantization Formats: Mistral-Small-3.1-24B-Instruct on NVIDIA RTX 4090
| Quantization Format | Bits/Weight | Weights VRAM | KV Cache (8K) | Total VRAM | Fits on 1x NVIDIA RTX 4090 (24GB)? | Est. Speed | GPUs Needed |
|---|---|---|---|---|---|---|---|
| FP16 / BF16 (Unquantized) | 16 bpw | 49.44 GB | 1.25 GB | 51.35 GB | No (Over by 27.4GB) | ~1.7 tok/s | 3x NVIDIA RTX 4090 |
| FP8 (W8A8 / E4M3) | 8.5 bpw | 26.27 GB | 1.25 GB | 28.17 GB | No (Over by 4.2GB) | ~3.2 tok/s | 2x NVIDIA RTX 4090 |
| GGUF Q8_0 (8-bit) | 8.5 bpw | 26.27 GB | 1.25 GB | 28.17 GB | No (Over by 4.2GB) | ~3.2 tok/s | 2x NVIDIA RTX 4090 |
| GGUF Q6_K (6-bit) | 6.56 bpw | 20.27 GB | 1.25 GB | 22.18 GB | Tight (92%) | ~31.8 tok/s | 1x NVIDIA RTX 4090 |
| GGUF Q5_K_M (5-bit) | 5.69 bpw | 17.58 GB | 1.25 GB | 19.49 GB | Yes (81%) | ~36.6 tok/s | 1x NVIDIA RTX 4090 |
| GGUF Q4_K_M / AWQ (4-bit) | 4.85 bpw | 14.99 GB | 1.25 GB | 16.9 GB | Yes (70%) | ~43 tok/s | 1x NVIDIA RTX 4090 |
| GGUF IQ4_XS (4.25 bpw) | 4.25 bpw | 13.13 GB | 1.25 GB | 15.04 GB | Yes (63%) | ~49 tok/s | 1x NVIDIA RTX 4090 |
| GGUF Q3_K_M (3-bit) | 3.91 bpw | 12.08 GB | 1.25 GB | 13.99 GB | Yes (58%) | ~53.3 tok/s | 1x NVIDIA RTX 4090 |
Frequently Asked Questions (Mistral-Small-3.1-24B-Instruct + NVIDIA RTX 4090)
Can NVIDIA RTX 4090 (24GB) run Mistral-Small-3.1-24B-Instruct locally?
Yes — NVIDIA RTX 4090 (24GB) can run Mistral-Small-3.1-24B-Instruct using GGUF Q6_K (6-bit). Full FP16 (51.35 GB) exceeds 24 GB VRAM, but 4-bit GGUF Q4_K_M / AWQ requires 16.9 GB (14.99 GB weights + 1.25 GB KV cache at 8K context), delivering ~43 tokens/sec.
How is KV Cache calculated for Mistral-Small-3.1-24B-Instruct?
Mistral-Small-3.1-24B-Instruct uses Dense-GQA with 40 transformer layers, 32 attention heads, and 8 KV heads (head dimension 128). At 8,192 tokens in FP16, KV cache consumes 1.25 GB. Enabling 8-bit KV cache (-ctk q8_0 -ctv q8_0 in llama.cpp or --kv-cache-dtype fp8 in vLLM) cuts this in half to 0.63 GB.