Mistral AIMixtralMixtralForCausalLM

Mixtral 8x22B v0.1 — Hardware Requirements & GPU Compatibility

Chat

Mixtral-8x22B-v0.1 is Mistral AI's large sparse mixture-of-experts base model, a pretrained checkpoint with no instruction tuning and no built-in moderation, intended as the foundation for fine-tuned or instruct derivatives rather than direct chat use. It routes each token through 2 of 8 experts, giving roughly 39 billion active parameters out of about 141 billion total, so its per-token compute is far lighter than its total size at the cost of having to hold every expert in memory. Its pretraining data covers English, French, German, Spanish, and Italian. Because all experts must be resident even though only a fraction activate per token, it still needs a multi-GPU workstation even when quantized. Context length is 65,536 tokens. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in April 2024, ahead of an instruction-tuned Mixtral-8x22B-Instruct release.

140.6B params · MoE39B active66K contextApache 2.0Apr 202412.0K downloads250 likes401 quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Mixtral 8x22B v0.1 runs on a 96 GB+ Mac or a workstation GPU

Best local pick → Mac Studio (M3 Ultra, 96GB) · Q4_K_M · 88.4 GB · ~19 tok/s

Macs from 48 GB (IQ2_XXS · ~27 tok/s)

Apache 2.0

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Mixtral 8x22B v0.1

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

Ollama

one command · needs Ollama installed

$ollama run hf.co/mradermacher/Mixtral-8x22B-v0.1-i1-GGUF

Q4_K_M · needs ~88.4 GB · 16K context

Download: Mixtral-8x22B-v0.1-i1-Q4_K_M.gguf84.37 GBmradermacher

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on H200 (141 GB) (52 GB headroom) · Q4_K_M

Generation speed: ~94 tok/s; generation speed
Cost per 1M output tokens: $10.64; per 1M output tokens

Deployon RunPodBest valuefrom $3.59/hr

$5+ bonus for new users

Compare GPUs →

or

How Much VRAM Does Mixtral 8x22B v0.1 Need?

Select a quantization level; the listed hardware adheres to it.

Quantization	VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
	39.4 GB	46.5 GB	—	37.9 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Does not fit: needs 42.7 GB
	43.0 GB	50 GB	—	42 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Does not fit: needs 46.3 GB
	44.7 GB	51.8 GB	—	42.6 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Does not fit: needs 48 GB
	48.2 GB	55.3 GB	—	46.7 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Does not fit: needs 51.5 GB
	60.5 GB	67.6 GB	—	~59.8 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Does not fit: needs 63.8 GB
	69.3 GB	76.4 GB	—	~68.6 GB	3-bit medium — 3-bit medium quantization	Does not fit: needs 72.6 GB
	85.1 GB	92.2 GB	—	~84.4 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Does not fit: needs 88.4 GB
	101.0 GB	108 GB	—	~100.2 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Does not fit: needs 104.3 GB
	116.8 GB	123.8 GB	—	~116 GB	6-bit, very good — 6-bit quantization, very good quality	Does not fit: needs 120.1 GB
	141.4 GB	148.4 GB	—	~140.6 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Does not fit: needs 144.7 GB
	282.0 GB	289.1 GB	—	~281.2 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 285.3 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

88.4GBat 16K context

Weights: 84.4 GB
KV cache: 3.8 GB
Overhead: 0.3 GB

Doesn't fit RTX 4090 (reference) · needs 64.4 GB more

≥10 tok/s needs ≥374 GB/s · M3 Ultra ~19, M2 Ultra ~19

Or .

Which Hardware Can Run Mixtral 8x22B v0.1?

Q4_K_M · 85.1 GB

Mixtral 8x22B v0.1 (Q4_K_M) requires 85.1 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 111+ GB is recommended. Using the full 66K context window can add up to 14.6 GB, bringing total usage to 99.7 GB. No consumer GPU has enough memory. 30 devices with unified memory can run Mixtral 8x22B v0.1, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 42.7 GB8 GB tier →12 GB tier →16 GB tier →

24 GB · RTX 4090referenceDoesn't fit — smallest quant needs 42.7 GB24 GB tier →

Mac 32 GB · Mac Mini M4Doesn't fit — smallest quant needs 42.7 GBMacs →

Mac 64 GB · Mac Studio M4 MaxIQ2_M · 51.5 GB · ~22 tok/s · 32K51.5~2232KMacs →

How good is Mixtral 8x22B v0.1?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

MMLU
Epoch AI · Apr 2024Lab-reported
77.8%
#17/85 open
Lab-reported

Mixtral 8x22B v0.1 benchmark results

Alternatives to Mixtral 8x22B v0.1

Mixtral family · 7 models →

Similar models

1tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
WizardLM 2 8x22Balpindale	—	88.4 GB	—needs > 24 GB	same size

Fine-tunes & variants(1)community builds of Mixtral 8x22B v0.1

Mixtral 8x22B Instruct v0.1

Specifications

Publisher: Mistral AI
Family: Mixtral
Parameters: 140.6B
Active parameters: 39.2B per token
Architecture: MixtralForCausalLM
Context Length: 65,536 tokens
Vocabulary Size: 32,000
Release Date: 2024-04-16
License: Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Mixtral 8x22B v0.1 need?

Mixtral 8x22B v0.1 requires 85.1 GB of VRAM at Q4_K_M, or 282.0 GB at BF16. Full 66K context adds up to 14.6 GB (99.7 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 140.6B × 4.8 bits ÷ 8 = 84.4 GB

KV Cache + Overhead ≈ 0.7 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 15.3 GB (at full 66K context)

VRAM usage by quantization

Q4_K_M

85.1 GB

Q4_K_M + full context

99.7 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Mixtral 8x22B v0.1?

No — Mixtral 8x22B v0.1 requires at least 39.4 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Mixtral 8x22B v0.1?

For Mixtral 8x22B v0.1, Q4_K_M (85.1 GB) offers the best balance of quality and VRAM usage. Q5_K_M (101.0 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 39.4 GB.

VRAM requirement by quantization

IQ2_XXS

39.4 GB~53%

IQ2_M

48.2 GB~62%

Q3_K_M

69.3 GB~83%

Q4_K_M ★

85.1 GB~89%

Q6_K

116.8 GB~95%

BF16

282.0 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Mixtral 8x22B v0.1 on a Mac?

Yes — Mac Studio M5 Max (48 GB) and 55 other Macs can run Mixtral 8x22B v0.1, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 85.1 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio (M3 Ultra, 96GB).

Can I run Mixtral 8x22B v0.1 locally?

Yes — Mixtral 8x22B v0.1 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Mixtral 8x22B v0.1?

At Q4_K_M, Mixtral 8x22B v0.1 can reach ~62 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 23.7 GB active ÷ (8000 × 0.65) = 4.56 ms, plus 56 layers × 0.055 ms = 3.08 ms, so 1000 ÷ 7.64 ms = ~131 tok/s

Estimated speed at Q4_K_M (85.1 GB)

NVIDIA B200

~131 tok/s

NVIDIA B300

~131 tok/s

NVIDIA GH200 Grace Hopper Superchip

~95 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Mixtral 8x22B v0.1?

At Q4_K_M, the download is about 84.37 GB. The full-precision BF16 version is 281.24 GB. The smallest option (IQ2_XXS) is 38.67 GB.

Which GPUs can run Mixtral 8x22B v0.1?

No single consumer GPU has enough VRAM to run Mixtral 8x22B v0.1 at Q4_K_M (85.1 GB). Multi-GPU or professional hardware is required.

Which devices can run Mixtral 8x22B v0.1?

30 devices with unified memory can run Mixtral 8x22B v0.1 at Q4_K_M (85.1 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

Run Mixtral 8x22B v0.1

88.4 GB VRAM at 16K context

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/mradermacher/Mixtral-8x22B-v0.1-i1-GGUF
mradermacher/Mixtral-8x22B-v0.1-i1-GGUF does not list Q4_K_M; Ollama picks a quant itself (Q4_K_M if present). Check the repo's Files tab.
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/mradermacher/Mixtral-8x22B-v0.1-i1-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/mradermacher/Mixtral-8x22B-v0.1-i1-GGUF
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf mradermacher/Mixtral-8x22B-v0.1-i1-GGUF -c 16384 -ngl 99 --port 8080
mradermacher/Mixtral-8x22B-v0.1-i1-GGUF does not list Q4_K_M; llama.cpp falls back to Q4_K_M or the first file.

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/mradermacher/Mixtral-8x22B-v0.1-i1-GGUF
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf mradermacher/Mixtral-8x22B-v0.1-i1-GGUF -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve mistralai/Mixtral-8x22B-v0.1 --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~309 GB of GPU memory.

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$mradermacher/Mixtral-8x22B-v0.1-i1-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
H200 SXM (141 GB)CheapestCheapest52 GB free	~94 tok/s	$10.64	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 52 GB free · ~94 tok/s · $10.64 / 1M tokens
H200 NVL (141 GB)52 GB free	~94 tok/s	$11.18	Deploy on Vast.aifrom $3.77/hr (opens in new tab)	141 GB · 52 GB free · ~94 tok/s · $11.18 / 1M tokens
H200 SXM (141 GB)52 GB free	~94 tok/s	$14.85	Deploy on Vast.aifrom $5.01/hr (opens in new tab)	141 GB · 52 GB free · ~94 tok/s · $14.85 / 1M tokens
B200 (192 GB)103 GB free	~131 tok/s	$12.69	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 103 GB free · ~131 tok/s · $12.69 / 1M tokens
B300 (288 GB)199 GB free	~131 tok/s	$14.73	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 199 GB free · ~131 tok/s · $14.73 / 1M tokens
B200 (192 GB)103 GB free	~131 tok/s	$15.92	Deploy on Vast.aifrom $7.50/hr (opens in new tab)	192 GB · 103 GB free · ~131 tok/s · $15.92 / 1M tokens

Compare GPUs →