Skip to content

feat: Build NCCL from source for more up-to-date versions #35

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

Draft
Eta0 wants to merge 2 commits into
base: master
Choose a base branch
from
Draft

feat: Build NCCL from source for more up-to-date versions #35

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
123e40c
feat: Build NCCL from source
Eta0 Apr 8, 2024
b5faf2f
feat: Build NCCL 2.21.5-1 for all targets
Eta0 Apr 8, 2024
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
8 changes: 4 additions & 4 deletions .github/workflows/ubuntu-20.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ jobs:
base-tag: 11.8.0-cudnn8-devel-ubuntu20.04
cuda-version-minor: "11.8.0"
cuda-version-major: "11.8"
nccl-version: 2.16.5-1
nccl-version: 2.21.5-1
cuda-samples-version: "11.6"
hpcx-distribution: "hpcx-v2.14-gcc-MLNX_OFED_LINUX-5-ubuntu20.04-cuda11-gdrcopy2-nccl2.16-x86_64"

Expand All @@ -29,7 +29,7 @@ jobs:
base-tag: 12.0.1-cudnn8-devel-ubuntu20.04
cuda-version-minor: "12.0.1"
cuda-version-major: "12.0"
nccl-version: 2.19.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.0"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu20.04-cuda12-x86_64"

Expand All @@ -42,7 +42,7 @@ jobs:
base-tag: 12.1.1-cudnn8-devel-ubuntu20.04
cuda-version-minor: "12.1.1"
cuda-version-major: "12.1"
nccl-version: 2.18.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.1"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu20.04-cuda12-x86_64"

Expand All @@ -68,7 +68,7 @@ jobs:
base-tag: 12.3.2-cudnn9-devel-ubuntu20.04
cuda-version-minor: "12.3.2"
cuda-version-major: "12.3"
nccl-version: 2.20.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.3"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu20.04-cuda12-x86_64"

Expand Down
8 changes: 4 additions & 4 deletions .github/workflows/ubuntu-22.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ jobs:
base-tag: 12.0.1-cudnn8-devel-ubuntu22.04
cuda-version-minor: "12.0.1"
cuda-version-major: "12.0"
nccl-version: 2.18.5-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.0"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu22.04-cuda12-x86_64"

Expand All @@ -29,7 +29,7 @@ jobs:
base-tag: 12.1.1-cudnn8-devel-ubuntu22.04
cuda-version-minor: "12.1.1"
cuda-version-major: "12.1"
nccl-version: 2.18.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.1"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu22.04-cuda12-x86_64"

Expand All @@ -42,7 +42,7 @@ jobs:
base-tag: 12.2.2-cudnn8-devel-ubuntu22.04
cuda-version-minor: "12.2.2"
cuda-version-major: "12.2"
nccl-version: 2.19.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.2"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu22.04-cuda12-x86_64"

Expand All @@ -55,7 +55,7 @@ jobs:
base-tag: 12.3.2-cudnn9-devel-ubuntu22.04
cuda-version-minor: "12.3.2"
cuda-version-major: "12.3"
nccl-version: 2.20.3-1
nccl-version: 2.21.5-1
cuda-samples-version: "12.3"
hpcx-distribution: "hpcx-v2.18-gcc-mlnx_ofed-ubuntu22.04-cuda12-x86_64"

Expand Down
18 changes: 16 additions & 2 deletions Dockerfile.ubuntu20
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,14 +12,13 @@ RUN apt-get -qq update && \
--no-install-recommends \
--allow-downgrades \
build-essential libtool autoconf automake autotools-dev unzip \
devscripts debhelper fakeroot \
ca-certificates \
wget curl openssh-server vim environment-modules \
iputils-ping net-tools \
libnuma1 libsubunit0 libpci-dev \
libpmix-dev \
datacenter-gpu-manager \
libnccl2=$TARGET_NCCL_VERSION+cuda${CUDA_VERSION_MAJOR} \
libnccl-dev=${TARGET_NCCL_VERSION}+cuda${CUDA_VERSION_MAJOR} \
git

# Mellanox OFED (latest)
Expand All @@ -32,6 +31,21 @@ RUN apt-get -qq update \
&& rm -rf /var/lib/apt/lists/*
# mlnx-ofed-hpc-user-only

ARG NCCL_ARCH_LIST='70 75 80 86 89 90'
RUN mkdir /tmp/build && \
cd /tmp/build && \
NCCL_TAG="$(echo $TARGET_NCCL_VERSION | tr -dc '[:digit:].-')" && \
wget -qO - "https://github.com/NVIDIA/nccl/archive/refs/tags/v${NCCL_TAG}.tar.gz" \
| tar --strip-components=1 -xzf - && \
REAL_ARCH() { printf -- '-gencode=arch=compute_%s,code=sm_%s ' "$1" "$1"; } && \
VIRTUAL_ARCH() { printf -- '-gencode=arch=compute_%s,code=compute_%s' "$1" "$1"; } && \
LAST_ARCH="${NCCL_ARCH_LIST##* }" && \
make -j20 pkg.debian.build \
"NVCC_GENCODE=$(for I in $NCCL_ARCH_LIST; do REAL_ARCH "$I"; done && VIRTUAL_ARCH "$LAST_ARCH")" && \
dpkg -i ./build/pkg/deb/libnccl*.deb && \
cd /tmp && \
rm -r /tmp/build

# IB perftest with GDR
ENV PERFTEST_VERSION_HASH=5b47ede

Expand Down
18 changes: 16 additions & 2 deletions Dockerfile.ubuntu22
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,14 +12,13 @@ RUN apt-get -qq update && \
--no-install-recommends \
--allow-downgrades \
build-essential libtool autoconf automake autotools-dev unzip \
devscripts debhelper fakeroot \
ca-certificates \
wget curl openssh-server vim environment-modules \
iputils-ping net-tools \
libnuma1 libsubunit0 libpci-dev \
libpmix-dev \
datacenter-gpu-manager \
libnccl2=$TARGET_NCCL_VERSION+cuda${CUDA_VERSION_MAJOR} \
libnccl-dev=${TARGET_NCCL_VERSION}+cuda${CUDA_VERSION_MAJOR} \
git && \
apt-get clean && \
rm -rf /var/lib/apt/lists/*
Expand All @@ -34,6 +33,21 @@ RUN apt-get -qq update \
&& rm -rf /var/lib/apt/lists/*
# mlnx-ofed-hpc-user-only

ARG NCCL_ARCH_LIST='70 75 80 86 89 90'
RUN mkdir /tmp/build && \
cd /tmp/build && \
NCCL_TAG="$(echo $TARGET_NCCL_VERSION | tr -dc '[:digit:].-')" && \
wget -qO - "https://github.com/NVIDIA/nccl/archive/refs/tags/v${NCCL_TAG}.tar.gz" \
| tar --strip-components=1 -xzf - && \
REAL_ARCH() { printf -- '-gencode=arch=compute_%s,code=sm_%s ' "$1" "$1"; } && \
VIRTUAL_ARCH() { printf -- '-gencode=arch=compute_%s,code=compute_%s' "$1" "$1"; } && \
LAST_ARCH="${NCCL_ARCH_LIST##* }" && \
make -j20 pkg.debian.build \
"NVCC_GENCODE=$(for I in $NCCL_ARCH_LIST; do REAL_ARCH "$I"; done && VIRTUAL_ARCH "$LAST_ARCH")" && \
dpkg -i ./build/pkg/deb/libnccl*.deb && \
cd /tmp && \
rm -r /tmp/build

# IB perftest with GDR
ENV PERFTEST_VERSION_HASH=5b47ede

Expand Down