27 lines
1.3 KiB
Plaintext
27 lines
1.3 KiB
Plaintext
Source: nccl
|
|
Section: universe/science
|
|
Maintainer: Boris Fomitchev <bfomitchev@nvidia.com>
|
|
Priority: optional
|
|
Build-depends: debhelper(>=9), nvidia-352-dev, cuda-toolkit-7-0,
|
|
cuda-curand-dev-7-0, cuda-cublas-dev-7-0, cuda-cudart-dev-7-0, cuda-ld-conf-7-0
|
|
Standards-Version: 3.9.5
|
|
|
|
Package: nccl
|
|
Section: universe/tools
|
|
Architecture: amd64
|
|
Depends: ${misc:Depends}, ${shlibs:Depends}, nvidia-352,
|
|
cuda-curand-7-0, cuda-cublas-7-0, cuda-cudart-7-0, cuda-ld-conf-7-0
|
|
Description: Optimized primitives for collective multi-GPU communication
|
|
NCCL (pronounced "Nickel") is a stand-alone library of standard collective communication routines,
|
|
such as all-gather, reduce, broadcast, etc., that have been optimized to achieve high bandwidth over PCIe.
|
|
NCCL supports up to eight GPUs and can be used in either single- or multi-process (e.g., MPI) applications.
|
|
|
|
Package: nccl-dev
|
|
Section: universe/tools
|
|
Architecture: amd64
|
|
Depends: ${misc:Depends}, ${shlibs:Depends}, nccl (= ${binary:Version})
|
|
Description: Dev package for NCCL library
|
|
NCCL (pronounced "Nickel") is a stand-alone library of standard collective communication routines,
|
|
such as all-gather, reduce, broadcast, etc., that have been optimized to achieve high bandwidth over PCIe.
|
|
NCCL supports up to eight GPUs and can be used in either single- or multi-process (e.g., MPI) applications.
|