Skip to content

Commit 7d1f6a2

Browse files
matthewdouglasmatthewdouglas
authored
docs: update AMD section (#2015)
1 parent 6721333 commit 7d1f6a2

1 file changed

Lines changed: 105 additions & 105 deletions

File tree

‎docs/source/installation.mdx‎

Lines changed: 105 additions & 105 deletions
Original file line numberDiff line numberDiff line change
@@ -10,16 +10,16 @@ We provide official support for NVIDIA GPUs, CPUs, Intel XPUs, and Intel Gaudi.
1010
- [NVIDIA CUDA](#cuda)
1111
- [Installation via PyPI](#cuda-pip)
1212
- [Compile from Source](#cuda-compile)
13+
- [AMD ROCm](#rocm)
14+
- [Installation via PyPI](#rocm-pip)
15+
- [Compile from Source](#rocm-compile)
1316
- [Intel XPU](#xpu)
1417
- [Installation via PyPI](#xpu-pip)
1518
- [Intel Gaudi](#gaudi)
1619
- [Installation via PyPI](#gaudi-pip)
1720
- [CPU](#cpu)
1821
- [Installation via PyPI](#cpu-pip)
1922
- [Compile from Source](#cpu-compile)
20-
- [AMD ROCm (Preview)](#rocm)
21-
- [Installation via PyPI](#rocm-pip)
22-
- [Compile from Source](#rocm-compile)
2323
- [Preview Wheels](#preview-wheels)
2424

2525
## System Requirements[[requirements]]
@@ -126,107 +126,8 @@ Big thanks to [wkpark](https://github.com/wkpark), [Jamezo97](https://github.com
126126
</hfoption>
127127
</hfoptions>
128128

129-
## Intel XPU[[xpu]]
130-
131-
* A compatible PyTorch version with Intel XPU support is required. The current minimum is **PyTorch 2.6.0**. It is recommended to use the latest stable release. See [Getting Started on Intel GPU](https://docs.pytorch.org/docs/stable/notes/get_start_xpu.html) for guidance.
132-
133-
### Installation via PyPI[[xpu-pip]]
134-
135-
This is the most straightforward and recommended installation option.
136-
137-
The currently distributed `bitsandbytes` packages are built with the following configurations:
138-
139-
| **OS** | **oneAPI Toolkit** | **Kernel Implementation** |
140-
|--------------------|------------------|----------------------|
141-
| **Linux x86-64** | 2025.1.3 | SYCL + Triton |
142-
| **Windows x86-64** | 2025.1.3 | SYCL + Triton |
143-
144-
The Linux build has a minimum glibc version of 2.34.
145-
146-
Use `pip` or `uv` to install the latest release:
147-
148-
```bash
149-
pip install bitsandbytes
150-
```
151-
152-
## Intel Gaudi[[gaudi]]
153-
154-
* A compatible PyTorch version with Intel Gaudi support is required. The current minimum is **Gaudi v1.21** with **PyTorch 2.6.0**. It is recommended to use the latest stable release. See the Gaudi software [installation guide](https://docs.habana.ai/en/latest/Installation_Guide/index.html) for guidance.
155-
156-
157-
### Installation from PyPI[[gaudi-pip]]
158-
159-
Use `pip` or `uv` to install the latest release:
160-
161-
```bash
162-
pip install bitsandbytes
163-
```
164-
165-
## CPU[[cpu]]
166-
167-
### Installation from PyPI[[cpu-pip]]
168-
169-
This is the most straightforward and recommended installation option.
170-
171-
The currently distributed `bitsandbytes` packages are built with the following configurations:
172-
173-
| **OS** | **Host Compiler** | Hardware Minimum
174-
|--------------------|----------------------|----------------------|
175-
| **Linux x86-64** | GCC 11.4 | AVX2 |
176-
| **Linux aarch64** | GCC 11.4 | |
177-
| **Windows x86-64** | MSVC 19.51+ (VS2026) | AVX2 |
178-
| **Windows arm64** | MSVC 19.43+ (VS2022) | ARM NEON |
179-
| **macOS arm64** | Apple Clang 17 | |
180-
181-
The Linux build has a minimum glibc version of 2.24.
182-
183-
Use `pip` or `uv` to install the latest release:
184-
185-
```bash
186-
pip install bitsandbytes
187-
```
188-
189-
### Compile from Source[[cpu-compile]]
190-
191-
To compile from source, simply install the package from source using `pip`. The package will be built for CPU only at this time.
192-
193-
<hfoptions id="cpu-source">
194-
<hfoption id="Linux / macOS">
195-
196-
```bash
197-
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
198-
pip install -e .
199-
```
200-
201-
</hfoption>
202-
<hfoption id="Windows (x86-64)">
203-
204-
Requires Visual Studio 2022 or 2026.
205-
206-
```bash
207-
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
208-
pip install -e .
209-
```
210-
211-
</hfoption>
212-
<hfoption id="Windows (ARM64)">
213-
214-
Requires Visual Studio 2022 or 2026 with the **ARM64 C++ build tools** component, Python >= **3.12**, and PyTorch >= **2.12**.
215-
216-
```bash
217-
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
218-
pip install -e .
219-
```
220-
221-
> [!NOTE]
222-
> The build system will detect the ARM64 architecture automatically via CMake. Only the CPU backend is supported on Windows ARM64 at this time (no CUDA).
223-
224-
</hfoption>
225-
</hfoptions>
129+
## AMD ROCm [[rocm]]
226130

227-
## AMD ROCm (Preview)[[rocm]]
228-
229-
* Support for AMD GPUs is currently in a preview state.
230131
* All features are supported for both consumer RDNA devices and Data Center CDNA products.
231132
* A compatible PyTorch version with AMD ROCm support is required. It is recommended to use the latest stable release. On Linux, see [PyTorch on ROCm](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/3rd-party/pytorch-install.html) for guidance. On Windows, ROCm-enabled PyTorch wheels are available from:
232133
- [repo.radeon.com/rocm/windows/](https://repo.radeon.com/rocm/windows/) — official AMD releases
@@ -257,14 +158,14 @@ pip install bitsandbytes
257158

258159
### Compile from Source[[rocm-compile]]
259160

260-
bitsandbytes can be compiled from ROCm 6.4 - ROCm 7.14.0. See the `CMakeLists.txt` for additional options.
161+
bitsandbytes can be compiled from ROCm 6.3 - ROCm 7.14.0. See the `CMakeLists.txt` for additional options.
261162

262163
<hfoptions id="rocm-source">
263164
<hfoption id="Linux">
264165

265166
To compile from source, you need CMake >= **3.31.6** and Python >= **3.10** installed. Make sure you have a compiler installed to compile C++ (`gcc`, `make`, headers, etc.).
266167

267-
You should also have a ROCm installation (system-wide or via Docker). The current minimum supported version is **6.4**.
168+
You should also have a ROCm installation (system-wide or via Docker). The current minimum supported version is **6.3**.
268169

269170
```bash
270171
# Install bitsandbytes from source
@@ -332,6 +233,105 @@ pip install .
332233
</hfoption>
333234
</hfoptions>
334235

236+
237+
## Intel XPU[[xpu]]
238+
239+
* A compatible PyTorch version with Intel XPU support is required. The current minimum is **PyTorch 2.6.0**. It is recommended to use the latest stable release. See [Getting Started on Intel GPU](https://docs.pytorch.org/docs/stable/notes/get_start_xpu.html) for guidance.
240+
241+
### Installation via PyPI[[xpu-pip]]
242+
243+
This is the most straightforward and recommended installation option.
244+
245+
The currently distributed `bitsandbytes` packages are built with the following configurations:
246+
247+
| **OS** | **oneAPI Toolkit** | **Kernel Implementation** |
248+
|--------------------|------------------|----------------------|
249+
| **Linux x86-64** | 2025.1.3 | SYCL + Triton |
250+
| **Windows x86-64** | 2025.1.3 | SYCL + Triton |
251+
252+
The Linux build has a minimum glibc version of 2.34.
253+
254+
Use `pip` or `uv` to install the latest release:
255+
256+
```bash
257+
pip install bitsandbytes
258+
```
259+
260+
## Intel Gaudi[[gaudi]]
261+
262+
* A compatible PyTorch version with Intel Gaudi support is required. The current minimum is **Gaudi v1.21** with **PyTorch 2.6.0**. It is recommended to use the latest stable release. See the Gaudi software [installation guide](https://docs.habana.ai/en/latest/Installation_Guide/index.html) for guidance.
263+
264+
265+
### Installation from PyPI[[gaudi-pip]]
266+
267+
Use `pip` or `uv` to install the latest release:
268+
269+
```bash
270+
pip install bitsandbytes
271+
```
272+
273+
## CPU[[cpu]]
274+
275+
### Installation from PyPI[[cpu-pip]]
276+
277+
This is the most straightforward and recommended installation option.
278+
279+
The currently distributed `bitsandbytes` packages are built with the following configurations:
280+
281+
| **OS** | **Host Compiler** | Hardware Minimum
282+
|--------------------|----------------------|----------------------|
283+
| **Linux x86-64** | GCC 11.4 | AVX2 |
284+
| **Linux aarch64** | GCC 11.4 | |
285+
| **Windows x86-64** | MSVC 19.51+ (VS2026) | AVX2 |
286+
| **Windows arm64** | MSVC 19.43+ (VS2022) | ARM NEON |
287+
| **macOS arm64** | Apple Clang 17 | |
288+
289+
The Linux build has a minimum glibc version of 2.24.
290+
291+
Use `pip` or `uv` to install the latest release:
292+
293+
```bash
294+
pip install bitsandbytes
295+
```
296+
297+
### Compile from Source[[cpu-compile]]
298+
299+
To compile from source, simply install the package from source using `pip`. The package will be built for CPU only at this time.
300+
301+
<hfoptions id="cpu-source">
302+
<hfoption id="Linux / macOS">
303+
304+
```bash
305+
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
306+
pip install -e .
307+
```
308+
309+
</hfoption>
310+
<hfoption id="Windows (x86-64)">
311+
312+
Requires Visual Studio 2022 or 2026.
313+
314+
```bash
315+
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
316+
pip install -e .
317+
```
318+
319+
</hfoption>
320+
<hfoption id="Windows (ARM64)">
321+
322+
Requires Visual Studio 2022 or 2026 with the **ARM64 C++ build tools** component, Python >= **3.12**, and PyTorch >= **2.12**.
323+
324+
```bash
325+
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
326+
pip install -e .
327+
```
328+
329+
> [!NOTE]
330+
> The build system will detect the ARM64 architecture automatically via CMake. Only the CPU backend is supported on Windows ARM64 at this time (no CUDA).
331+
332+
</hfoption>
333+
</hfoptions>
334+
335335
## Preview Wheels[[preview-wheels]]
336336

337337
If you would like to use new features even before they are officially released and help us test them, feel free to install the wheel directly from our CI (*the wheel links will remain stable!*):

0 commit comments

Comments
 (0)