Mixtral-8x22B-Instruct (MoE) on NVIDIA H200 141GB SXM (141GB VRAM)
Full FP16 (292.87 GB) exceeds 141 GB VRAM, but 4-bit GGUF Q4_K_M / AWQ requires 90.46 GB (88.05 GB weights + 1.75 GB KV cache at 8K context), delivering ~125.9 tokens/sec.
🎛️ Interactive VRAM & KV-Cache Tuner
📊 Real-Time Memory Footprint
90.46 GB / 141 GB🚀 Ready-to-Run Launch Command
ollama run mixtral:8x22b # Or with llama.cpp server (8K context, Flash Attention): llama-server -hf mistralai/Mixtral-8x22B-Instruct-v0.1-GGUF -c 8192 -ngl 99 -fa
Need 90.46 GB+ VRAM to run Mixtral-8x22B-Instruct (MoE)?
Recommended cloud setup for 4-bit: Multi-GPU H100 / H200 Node (or 292.87 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: Mixtral-8x22B-Instruct (MoE) on NVIDIA H200 141GB SXM
| Quantization Format | Bits/Weight | Weights VRAM | KV Cache (8K) | Total VRAM | Fits on 1x NVIDIA H200 141GB SXM (141GB)? | Est. Speed | GPUs Needed |
|---|---|---|---|---|---|---|---|
| FP16 / BF16 (Unquantized) | 16 bpw | 290.46 GB | 1.75 GB | 292.87 GB | No (Over by 151.9GB) | ~4.9 tok/s | 3x NVIDIA H200 141GB SXM |
| FP8 (W8A8 / E4M3) | 8.5 bpw | 154.31 GB | 1.75 GB | 156.72 GB | No (Over by 15.7GB) | ~9.3 tok/s | 2x NVIDIA H200 141GB SXM |
| GGUF Q8_0 (8-bit) | 8.5 bpw | 154.31 GB | 1.75 GB | 156.72 GB | No (Over by 15.7GB) | ~9.3 tok/s | 2x NVIDIA H200 141GB SXM |
| GGUF Q6_K (6-bit) | 6.56 bpw | 119.09 GB | 1.75 GB | 121.5 GB | Yes (86%) | ~93.1 tok/s | 1x NVIDIA H200 141GB SXM |
| GGUF Q5_K_M (5-bit) | 5.69 bpw | 103.29 GB | 1.75 GB | 105.71 GB | Yes (75%) | ~107.3 tok/s | 1x NVIDIA H200 141GB SXM |
| GGUF Q4_K_M / AWQ (4-bit) | 4.85 bpw | 88.05 GB | 1.75 GB | 90.46 GB | Yes (64%) | ~125.9 tok/s | 1x NVIDIA H200 141GB SXM |
| GGUF IQ4_XS (4.25 bpw) | 4.25 bpw | 77.15 GB | 1.75 GB | 79.56 GB | Yes (56%) | ~143.6 tok/s | 1x NVIDIA H200 141GB SXM |
| GGUF Q3_K_M (3-bit) | 3.91 bpw | 70.98 GB | 1.75 GB | 73.39 GB | Yes (52%) | ~156.1 tok/s | 1x NVIDIA H200 141GB SXM |
Frequently Asked Questions (Mixtral-8x22B-Instruct (MoE) + NVIDIA H200 141GB SXM)
Can NVIDIA H200 141GB SXM (141GB) run Mixtral-8x22B-Instruct (MoE) locally?
Yes — NVIDIA H200 141GB SXM (141GB) can run Mixtral-8x22B-Instruct (MoE) using GGUF Q6_K (6-bit). Full FP16 (292.87 GB) exceeds 141 GB VRAM, but 4-bit GGUF Q4_K_M / AWQ requires 90.46 GB (88.05 GB weights + 1.75 GB KV cache at 8K context), delivering ~125.9 tokens/sec.
How is KV Cache calculated for Mixtral-8x22B-Instruct (MoE)?
Mixtral-8x22B-Instruct (MoE) uses MoE-GQA with 56 transformer layers, 48 attention heads, and 8 KV heads (head dimension 128). At 8,192 tokens in FP16, KV cache consumes 1.75 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 0.88 GB.