# Unsloth Requirements
Unsloth can be used in two ways: through [Unsloth Studio](https://unsloth.ai/docs/new/studio/install), the web UI, or through [Unsloth Core](#unsloth-core-requirements), the original code-based version. Each has different requirements.
## **Unsloth Studio Requirements**
* **Mac:** Like CPU - [Chat](https://unsloth.ai/docs/new/studio/chat#using-unsloth-studio-chat) + [Data Recipes](https://unsloth.ai/docs/new/studio/data-recipe) works for now. **MLX** training coming very soon.
* **CPU: Unsloth still works without a GPU**, for Chat + Data Recipes.
* **Training:** Works on **NVIDIA**: RTX 30, 40, 50, Blackwell, DGX Spark/Station etc. + **Intel** GPUs
* **Coming soon:** Support for **Apple MLX** and **AMD**.
### :windows: Window**s**
Unsloth Studio works directly on Windows without WSL. To train models, make sure your system satisfies these requirements:
**Requirements**
* Windows 10 or Windows 11 (64-bit)
* NVIDIA GPU with drivers installed
* **App Installer** (includes `winget`): [here](https://learn.microsoft.com/en-us/windows/msix/app-installer/install-update-app-installer)
* **Git**: `winget install --id Git.Git -e --source winget`
* **Python**: version 3.11 up to, but not including, 3.14
* Work inside a Python environment such as **uv**, **venv**, or **conda/mamba**
### :apple: MacOS
Unsloth Studio works on Mac devices for [Chat](#run-models-locally) for GGUF models and [Data Recipes](https://unsloth.ai/docs/new/studio/data-recipe) ([Export](https://unsloth.ai/docs/new/studio/export) coming very soon). **MLX training coming soon!**
* macOS 12 Monterey or newer (Intel or Apple Silicon)
* Install Homebrew: [here](https://brew.sh/)
* Git: `brew install git`
* cmake: `brew install cmake`
* openssl: `brew install openssl`
* Python: version 3.11 up to, but not including, 3.14
* Work inside a Python environment such as **uv**, **venv**, or **conda/mamba**
### :linux: Linux & WSL
* Ubuntu 20.04+ or similar distro (64-bit)
* NVIDIA GPU with drivers installed
* CUDA toolkit (12.4+ recommended, 12.8+ for blackwell)
* Git: `sudo apt install git`
* Python: version 3.11 up to, but not including, 3.14
* Work inside a Python environment such as **uv**, **venv**, or **conda/mamba**
### :microchip: CPU only
Unsloth Studio supports CPU devices for [Chat](#run-models-locally) for GGUF models and [Data Recipes](https://unsloth.ai/docs/new/studio/data-recipe) ([Export](https://unsloth.ai/docs/new/studio/export) coming very soon)
* Same as the ones mentioned above for Linux (except for NVIDIA GPU drivers) and MacOS.
### **Training**
Unsloth Studio Training currently works on NVIDIA GPUs, with AMD, MLX, Intel support coming very soon. You can still use the [original Unsloth Core](#unsloth-requirements) to train on AMD and Intel devices. **Python 3.11–3.13** is required.
| Requirement | Linux / WSL | Windows |
| ---------------- | ---------------------------------------- | --------------------------------------------- |
| **Git** | Usually preinstalled | Installed by setup script (`winget`) |
| **CMake** | Preinstalled or `sudo apt install cmake` | Installed by setup script (`winget`) |
| **C++ compiler** | `build-essential` | Visual Studio Build Tools 2022 |
| **CUDA Toolkit** | Optional; `nvcc` auto-detected | Installed by setup script (matched to driver) |
## Unsloth Core Requirements
* **Operating System**: Works on Linux and [Windows](https://docs.unsloth.ai/get-started/install-and-update/windows-installation)
* Supports NVIDIA GPUs since 2018+ including [Blackwell RTX 50](https://unsloth.ai/docs/blog/fine-tuning-llms-with-blackwell-rtx-50-series-and-unsloth) and [DGX Spark](https://unsloth.ai/docs/blog/fine-tuning-llms-with-nvidia-dgx-spark-and-unsloth)
* Minimum CUDA Capability 7.0 (V100, T4, Titan V, RTX 20 & 50, A100, H100, L40 etc) [Check your GPU!](https://developer.nvidia.com/cuda-gpus) GTX 1070, 1080 works, but is slow.
* The official [Unsloth Docker image](https://hub.docker.com/r/unsloth/unsloth) `unsloth/unsloth` is available on Docker Hub
* [docker](https://unsloth.ai/docs/get-started/install/docker "mention")
* Unsloth works on [AMD](https://unsloth.ai/docs/get-started/install/amd) and [Intel](https://unsloth.ai/docs/get-started/install/intel) GPUs (follow our [specific guides](https://unsloth.ai/docs/get-started/install)). Apple/Silicon/MLX is in the works
* Your device should have `xformers`, `torch`, `BitsandBytes` and `triton` support.
* If you have different versions of torch, transformers etc., `pip install unsloth` will automatically install all the latest versions of those libraries so you don't need to worry about version compatibility.
{% hint style="info" %}
Python 3.13 is supported!
{% endhint %}
### Fine-tuning VRAM requirements:
How much GPU memory do I need for LLM fine-tuning using Unsloth?
{% hint style="info" %}
A common issue when you OOM or run out of memory is because you set your batch size too high. Set it to 1, 2, or 3 to use less VRAM.
**For context length benchmarks, see** [**here**](https://unsloth.ai/docs/basics/unsloth-benchmarks#context-length-benchmarks)**.**
{% endhint %}
Check this table for VRAM requirements sorted by model parameters and fine-tuning method. QLoRA uses 4-bit, LoRA uses 16-bit. Keep in mind that sometimes more VRAM is required depending on the model so these numbers are the absolute minimum:
| Model parameters | QLoRA (4-bit) VRAM | LoRA (16-bit) VRAM |
| ---------------- | ------------------ | ------------------ |
| 3B | 3.5 GB | 8 GB |
| 7B | 5 GB | 19 GB |
| 8B | 6 GB | 22 GB |
| 9B | 6.5 GB | 24 GB |
| 11B | 7.5 GB | 29 GB |
| 14B | 8.5 GB | 33 GB |
| 27B | 22GB | 64GB |
| 32B | 26 GB | 76 GB |
| 40B | 30GB | 96GB |
| 70B | 41 GB | 164 GB |
| 81B | 48GB | 192GB |
| 90B | 53GB | 212GB |
| 405B | 237 GB | 950 GB |