microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-05-11 17:00:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

prerequisites update

Browse Source
This commit is contained in:
Changho Hwang
2026-04-01 21:18:47 +00:00
parent ff4d825652
commit 94d0508ec2
1 changed files with 3 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
docs/quickstart.md
View File

@@ -31,6 +31,9 @@
```
If you don't want to build Python module, you need to set `-DMSCCLPP_BUILD_PYTHON_BINDINGS=OFF` in your `cmake` command (see details in [Install from Source](#install-from-source)).
* (Optional, for benchmarks) MPI
* (Optional, for NVIDIA platforms) [GDRCopy](https://github.com/NVIDIA/gdrcopy) >= 2.5.0
* GDRCopy is required for IB `HostNoAtomic` mode, which uses CPU-side signal forwarding to GPU memory via BAR1 mappings. This mode is used on platforms where RDMA atomics are not available (e.g., when using Data Direct Virtual Functions).
* Install GDRCopy from source or via packages. See the [GDRCopy installation guide](https://github.com/NVIDIA/gdrcopy#installation).
* Others
* For RDMA (InfiniBand or RoCE) support on NVIDIA platforms, [GPUDirect RDMA](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-operator-rdma.html#gpudirect-rdma-and-gpudirect-storage) should be supported by the system. See the detailed prerequisites from [this NVIDIA documentation](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-operator-rdma.html#common-prerequisites).
* For NVLink SHARP (NVLS) support on NVIDIA platforms, the Linux kernel version should be 5.6 or above.