Skip to content

add README for AMD ROCm #32

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Apophis3158 wants to merge 1 commit into
base: master
Choose a base branch
from
+12 −3
Open

add README for AMD ROCm #32

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 12 additions & 3 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,18 @@ This project is a pytorch VRAM allocator that implements on-demand offloading of

## Support:

* **Nvidia GPUs only**
* **Nvidia GPUs**
* **CUDA 12.8+**
* **AMD GPUs (Experimental)**
* **Linux: ROCm 7.2.2+** <!-- NOTE: VMM fix: https://github.com/ROCm/rocm-systems/pull/4363 -->
* <details><summary><strong>Windows: TheRock 7.13+</strong></summary>

`ROCm 7.2.1` used in the [official installation guide](https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html#install-pytorch-via-pip) has been proven to have defects with VMM. And in ROCm SDK `hip_runtime_api.h`, many memory related APIs are noted: **implemented on Linux** and **under development on Microsoft Windows**. <!-- TODO: update after next ROCm release -->

Therefore, using [TheRock](https://github.com/ROCm/TheRock) releases is strongly recommended ([TheRock installation guide](https://github.com/ROCm/TheRock/blob/main/RELEASES.md#installing-per-family-releases-using-pip)). <!-- TODO: switch to https://github.com/ROCm/TheRock/blob/main/RELEASES.md#installing-multi-arch-pytorch-python-packages after multi-arch torch wheels pass staging tests -->

</details>
* **Pytorch 2.8+**
* **Cuda 12.8+**
* **Windows 11+** / **Linux** as per python ManyLinux support

---
Expand Down Expand Up @@ -46,7 +55,7 @@ see examples/example.py

## Backend:

* VBAR allocation is done with `cuMemAddressReserve()`, faulting with `cuMemCreate()` and `cuMemMap()` and all frees done with appropriate converse APIs.
* VBAR allocation is done with `cuMemAddressReserve()` (`hipMemAddressReserve()` for ROCm), faulting with `cuMemCreate()` and `cuMemMap()` and all frees done with appropriate converse APIs.
* For consistency with VBAR memory management, main pytorch allocator plugin is also implemented with `cuMemAddressReserve` -> `cuMemCreate` -> `cuMemMap`. This also behaves a lot better on Windows systems with System Memory fallback.

## Caveats:
Expand Down
Loading