From 351b95b926ed4e29b5e0e15b7d183661c8d7d03b Mon Sep 17 00:00:00 2001 From: Changho Hwang Date: Fri, 24 Nov 2023 17:00:18 +0800 Subject: [PATCH] Update documents (#225) Adding AMD supports on the docs --- README.md | 4 +++- docs/quickstart.md | 3 ++- 2 files changed, 5 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index ff97ad887..9796179d3 100644 --- a/README.md +++ b/README.md @@ -9,6 +9,8 @@ | Unit Tests (CUDA) | [![Build Status](https://dev.azure.com/binyli/HPC/_apis/build/status%2Fmscclpp-ut?branchName=main)](https://dev.azure.com/binyli/HPC/_build/latest?definitionId=4&branchName=main) | | Integration Tests (CUDA) | [![Build Status](https://dev.azure.com/binyli/HPC/_apis/build/status%2Fmscclpp-test?branchName=main)](https://dev.azure.com/binyli/HPC/_build/latest?definitionId=3&branchName=main) | +*NOTE (Nov 2023): Azure pipelines for ROCm will be added soon.* + A GPU-driven communication stack for scalable AI applications. See [Quick Start](docs/quickstart.md) to quickly get started. @@ -75,7 +77,7 @@ __device__ void barrier() { } ``` -MSCCL++ provides consistent interfaces, i.e., the above interfaces are used regardless of the location of the remote GPU (either on the local node or on a remote node) or the underlying link (either NVLink or InfiniBand). +MSCCL++ provides consistent interfaces, i.e., the above interfaces are used regardless of the location of the remote GPU (either on the local node or on a remote node) or the underlying link (either NVLink/xGMI or InfiniBand). ### ProxyChannel and SmChannel diff --git a/docs/quickstart.md b/docs/quickstart.md index 19b32d6ec..0a3f0625f 100644 --- a/docs/quickstart.md +++ b/docs/quickstart.md @@ -10,7 +10,8 @@ * Non-Azure Systems * NVIDIA A100 GPUs + CUDA >= 11.8 * NVIDIA H100 GPUs + CUDA >= 12.0 - * AMD support is underway. + * AMD MI250X GPUs + ROCm >= 5.7 + * AMD MI300X GPUs + ROCm >= 5.7 * OS: tested over Ubuntu 18.04 and 20.04 * Libraries: [libnuma](https://github.com/numactl/numactl), MPI (optional) * Others