File tree Expand file tree Collapse file tree
Expand file tree Collapse file tree Original file line number Diff line number Diff line change @@ -16,15 +16,11 @@ RUN cuda_version=$(echo ${CUDA_VERSION} | awk -F . '{ print $1"-"$2 }') \
1616 && apt-get install -y --no-install-recommends \
1717 cuda-libraries-dev-${cuda_version} \
1818 cuda-nvcc-${cuda_version} \
19- libhwloc-dev \
20- autoconf \
21- automake \
22- libtool \
2319 && rm -rf /var/lib/apt/lists/*
2420
2521# EFA
2622
27- ARG EFA_VERSION=1.38.1
23+ ARG EFA_VERSION=1.48.0
2824
2925RUN cd /tmp \
3026 && apt-get update \
@@ -36,34 +32,14 @@ RUN cd /tmp \
3632
3733# NCCL
3834
39- ARG NCCL_VERSION=2.26.2 -1
35+ ARG NCCL_VERSION=2.27.7 -1
4036
4137RUN cd /tmp \
4238 && git clone https://github.com/NVIDIA/nccl.git -b v${NCCL_VERSION} \
4339 && cd nccl \
4440 && make -j$(nproc) src.build BUILDDIR=${NCCL_HOME} \
4541 && rm -rf /tmp/nccl
4642
47- # AWS OFI NCCL
48-
49- ARG OFI_VERSION=1.14.0
50-
51- RUN cd /tmp \
52- && git clone https://github.com/aws/aws-ofi-nccl.git -b v${OFI_VERSION} \
53- && cd aws-ofi-nccl \
54- && ./autogen.sh \
55- && ./configure \
56- --with-cuda=${CUDA_HOME} \
57- --with-libfabric=${LIBFABRIC_PATH} \
58- --with-mpi=${OPEN_MPI_PATH} \
59- --with-cuda=${CUDA_HOME} \
60- --with-nccl=${NCCL_HOME} \
61- --disable-tests \
62- --prefix=${NCCL_HOME} \
63- && make -j$(nproc) \
64- && make install \
65- && rm -rf /tmp/aws-ofi-nccl /var/lib/apt/lists/*
66-
6743# NCCL Tests
6844
6945RUN cd /opt \
You can’t perform that action at this time.
0 commit comments