MODEL ARCHITECTURE SPEC • MISTRAL AI24B Dense
Mistral Small 3 24B Instruct VRAM & inference speed
24B Total ParamsDense Architecture40 Layers • 8 KV Heads32K Default Ctx (Max 128K)Ollama: mistral-small:24b
Accurate memory requirements, KV cache footprint, and tokens/sec generation benchmarks for running Mistral Small 3 24B Instruct across local GPUs, Apple Silicon Macs, and cloud instances.
• TIGHT FIT18.33 GB (92%)~31 t/s
Mistral Small 3 24B Instruct on AMD Radeon RX 7900 XT (20GB GDDR6)
• Tight Fit — Runs 100% on GPU (Low Headroom)
• TIGHT FIT
Required VRAM
18.33 GB
92% of hardware capacity (20 GB)Est. Decode Speed
~31 tok/s
Dense model @ 800 GB/sAvailable Headroom
+1.67 GB
Tight margin — close background appsMemory Allocation & Headroom
18.33 GB allocated of 20 GB capacity
92% utilized
Weights: 14.4 GB
KV: (32K): 2.68 GB
Overhead: 1.25 GB
Free Headroom: +1.67 GB
Architecture & Throughput Transparency: Single-user decode speed (~31 tok/s) is memory-bandwidth bound at batch size = 1 on AMD Radeon RX 7900 XT (20GB GDDR6) (800 GB/s). Prompt prefill / TTFT (~450 tok/s) is compute-bound and utilizes GPU Tensor cores. Actual latency varies with FlashAttention-2 and runtime engine (Ollama/vLLM/llama.cpp).
Cloud Instance Options for Mistral Small 3 24B Instruct
On-DemandDeploy this model on an on-demand cloud GPU without local hardware constraints or power usage:
RunPod Cloud GPUsBest 1-Click Ollama & vLLM
$0.38/hr (RTX 4090 Spot) – $2.39/hr (H100)Vast.ai GPU MarketplaceCheapest Dual RTX 3090 / 4090
$0.18/hr (RTX 3060) – $0.78/hr (2x RTX 3090)Lambda GPU CloudEnterprise 8x H100 Clusters
$2.49/hr (H100) – $19.92/hr (8x H100)Quantization Precision Matrix
32K ctx • KV: Q8_0| Quantization | Weights | KV Cache | Total VRAM | Est. Speed | Fit Status |
|---|---|---|---|---|---|
| FP1616 bpw | 48 GB | 2.68 GB | 53.11 GB (266%) | 0 t/s | OOM |
| Q8_08.5 bpw | 25.44 GB | 2.68 GB | 29.76 GB (149%) | 0 t/s | OOM |
| Q6_K6.56 bpw | 19.68 GB | 2.68 GB | 23.8 GB (119%) | ~5 t/s | PARTIAL |
| Q5_K_M5.68 bpw | 17.04 GB | 2.68 GB | 21.07 GB (105%) | ~6 t/s | PARTIAL |
| Q4_K_M4.8 bpw | 14.4 GB | 2.68 GB | 18.33 GB (92%) | ~31 t/s | FITS |
| FP44.25 bpw | 12.72 GB | 2.68 GB | 16.6 GB (83%) | ~35 t/s | FITS |
| IQ3_XS3.3 bpw | 9.84 GB | 2.68 GB | 13.61 GB (68%) | ~44 t/s | FITS |
Hardware Compatibility Matrix for Mistral Small 3 24B Instruct
Standard Q4_K_M precision with 32K context window and Q8_0 KV cache
| Hardware System | VRAM Envelope | Required VRAM | Headroom | Est. Decode Speed | Compatibility | Test Fit |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 (32GB GDDR7) Consumer GPU • 1792 GB/s | 32 GB | 18.33 GB | +13.67 GB | ~81 tok/s | ✓ FITS | Test Fit → |
| NVIDIA GeForce RTX 4090 (24GB GDDR6X) Consumer GPU • 1008 GB/s | 24 GB | 18.33 GB | +5.67 GB | ~45 tok/s | ✓ FITS | Test Fit → |
| 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM) Multi-GPU Rig • 1648 GB/s | 48 GB | 19.03 GB | +28.97 GB | ~68 tok/s | ✓ FITS | Test Fit → |
| NVIDIA GeForce RTX 5080 (16GB GDDR7) Consumer GPU • 960 GB/s | 16 GB | 18.33 GB | -2.33 GB | ~10 tok/s | ! PARTIAL | Test Fit → |
| NVIDIA RTX 4070 Ti Super (16GB) Consumer GPU • 672 GB/s | 16 GB | 18.33 GB | -2.33 GB | ~7 tok/s | ! PARTIAL | Test Fit → |
| NVIDIA GeForce RTX 3060 (12GB) Consumer GPU • 360 GB/s | 12 GB | 18.33 GB | -6.33 GB | 0 (OOM) | ✕ OOM | Test Fit → |
| Apple MacBook Pro/Air M1 (8GB Unified Memory) Apple Silicon • 68 GB/s | 7 GB | 18.33 GB | -11.33 GB | 0 (OOM) | ✕ OOM | Test Fit → |
| Apple MacBook Pro/Air (16GB Unified Memory) Apple Silicon • 100 GB/s | 14 GB | 18.33 GB | -4.33 GB | ~1 tok/s | ! PARTIAL | Test Fit → |
| Apple MacBook Pro M4 Pro (24GB Unified Memory) Apple Silicon • 273 GB/s | 20 GB | 18.33 GB | +1.67 GB | ~12 tok/s | • TIGHT FIT | Test Fit → |
| Apple Mac Mini M4 Pro (64GB Unified Memory) Apple Silicon • 273 GB/s | 54 GB | 18.33 GB | +35.67 GB | ~11 tok/s | ✓ FITS | Test Fit → |
| Apple Mac Studio M4 Max (128GB Unified Memory) Apple Silicon • 546 GB/s | 112 GB | 18.33 GB | +93.67 GB | ~23 tok/s | ✓ FITS | Test Fit → |
| Apple Mac Studio M3 Ultra (256GB Unified Memory) Apple Silicon • 819 GB/s | 236 GB | 18.33 GB | +217.67 GB | ~35 tok/s | ✓ FITS | Test Fit → |
| AMD Strix Halo Mini-PC (Ryzen AI Max+ 395 128GB) Unified APU / Mini-PC • 256 GB/s | 110 GB | 18.33 GB | +91.67 GB | ~10 tok/s | ✓ FITS | Test Fit → |
| NVIDIA DGX Spark (128GB Grace-Blackwell Unified) AI Appliance • 273 GB/s | 118 GB | 18.33 GB | +99.67 GB | ~12 tok/s | ✓ FITS | Test Fit → |
| NVIDIA H100 SXM (80GB HBM3 Cloud GPU) Datacenter Cloud GPU • 3350 GB/s | 80 GB | 18.33 GB | +61.67 GB | ~160 tok/s | ✓ FITS | Test Fit → |
| NVIDIA GeForce RTX 5070 Ti (16GB GDDR7) Consumer GPU • 896 GB/s | 16 GB | 18.33 GB | -2.33 GB | ~9 tok/s | ! PARTIAL | Test Fit → |
| NVIDIA GeForce RTX 5070 (12GB GDDR7) Consumer GPU • 672 GB/s | 12 GB | 18.33 GB | -6.33 GB | 0 (OOM) | ✕ OOM | Test Fit → |
| NVIDIA GeForce RTX 4080 Super (16GB GDDR6X) Consumer GPU • 736 GB/s | 16 GB | 18.33 GB | -2.33 GB | ~7 tok/s | ! PARTIAL | Test Fit → |
| AMD Radeon RX 7900 XTX (24GB GDDR6) Consumer GPU • 960 GB/s | 24 GB | 18.33 GB | +5.67 GB | ~37 tok/s | ✓ FITS | Test Fit → |
| AMD Radeon RX 7900 XT (20GB GDDR6) Consumer GPU • 800 GB/s | 20 GB | 18.33 GB | +1.67 GB | ~31 tok/s | • TIGHT FIT | Test Fit → |
| Apple Mac Studio M2 Ultra (192GB Unified Memory) Apple Silicon Workstation • 800 GB/s | 144 GB | 18.33 GB | +125.67 GB | ~35 tok/s | ✓ FITS | Test Fit → |
| NVIDIA GeForce RTX 3090 (24GB GDDR6X) Consumer GPU • 936 GB/s | 24 GB | 18.33 GB | +5.67 GB | ~40 tok/s | ✓ FITS | Test Fit → |
| NVIDIA GeForce RTX 4060 Ti (16GB GDDR6) Consumer GPU • 288 GB/s | 16 GB | 18.33 GB | -2.33 GB | ~3 tok/s | ! PARTIAL | Test Fit → |
| Apple Mac Mini M4 (16GB Unified Memory) Apple Silicon • 120 GB/s | 13 GB | 18.33 GB | -5.33 GB | 0 (OOM) | ✕ OOM | Test Fit → |
| Apple Mac Mini M4 (24GB Unified Memory) Apple Silicon • 120 GB/s | 20 GB | 18.33 GB | +1.67 GB | ~5 tok/s | • TIGHT FIT | Test Fit → |