Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 21, 2025

• Switched to NVTX3

  jirikraus

  committed
  2dc777a

  Copy full SHA for 2dc777a

Commits on Oct 30, 2024

• Added User Buffer Registration option to NCCL variants.

  jirikraus

  committed
  2aaa5ce

  Copy full SHA for 2aaa5ce

Commits on Aug 14, 2024

• Update README.md

  Show description for e21d9c2
  jirikraus

  authored
  e21d9c2

  Copy full SHA for e21d9c2

Commits on Jul 22, 2024

• Added multi_node_p2p variant.

  jirikraus

  committed
  584eb2b

  Copy full SHA for 584eb2b

Commits on Mar 14, 2024

• Added MPI_Barrier after MPI_Bcast to make NCCL initi more robust.

  jirikraus

  committed
  83a5970

  Copy full SHA for 83a5970

• GTC 2024 update.

  jirikraus

  committed
  c23b12e

  Copy full SHA for c23b12e

Commits on Oct 17, 2022

• Improved device affinity handling

  Show description for f90a770
  jirikraus

  committed
  f90a770

  Copy full SHA for f90a770

• Added NCCL version combined with CUDA Graphs.

  jirikraus

  committed
  88d4e61

  Copy full SHA for 88d4e61

Commits on May 2, 2022

• Added documentation for use_hp_streams and warning for nccheck overlap interference effects in MPI overlap variant.

  jirikraus

  committed
  9740968

  Copy full SHA for 9740968

Commits on Apr 29, 2022

• Fixed typo in device affinity handling comment.

  

  mhrywniak

  authored and

  jirikraus

  committed
  5dd3f4d

  Copy full SHA for 5dd3f4d

• Added support to handle GPU affinity with CUDA_VISIBLE_DEVICES. Requires NVSHMEM 2.5 or later.

  jirikraus

  committed
  d02153d

  Copy full SHA for d02153d

Commits on Mar 9, 2022

• Linked NVSHMEM version to nvidia-ml which is equired starting with NVSHMEM 2.5.

  jirikraus

  committed
  46df3e9

  Copy full SHA for 46df3e9

Commits on Feb 23, 2022

• Moved MPIX_Query_cuda_support to after MPI_Init required by some CUDA-aware MPI implementations.

  jirikraus

  committed
  2573846

  Copy full SHA for 2573846

• Improved device affinity handling to allow using CUDA_VISIBLE_DEVICES to handle device affinity.

  jirikraus

  committed
  1636262

  Copy full SHA for 1636262

Commits on Jan 24, 2022

• Added exit on CUDA RT, MPI or NCCL error (fixes Issue #14).

  jirikraus

  committed
  4c9806c

  Copy full SHA for 4c9806c

Commits on Nov 29, 2021

• README.md: fixed typo and improvements.

  

  mhrywniak

  authored and

  jirikraus

  committed
  d18ee2d

  Copy full SHA for d18ee2d

Commits on Nov 26, 2021

• Fixing issue #12: Use of uninitialized variables?

  jirikraus

  committed
  3a7090a

  Copy full SHA for 3a7090a

• Added check for CUDA-aware MPI. To avoid issues like #13. Thanks @eyalroz for the suggestion.

  jirikraus

  committed
  1aed265

  Copy full SHA for 1aed265

Commits on Nov 11, 2021

• Merge pull request #11 from NVIDIA/jirikraus-tutorial-multi-gpu

  Show description for bc1d24f
  jirikraus

  authored
  bc1d24f

  Copy full SHA for bc1d24f

• Added reference to tutorial-multi-gpu repo

  jirikraus

  authored
  81dc7b8

  Copy full SHA for 81dc7b8

Commits on Oct 5, 2021

• Replaced nvshmemx_long_signal with nvshmemx_signal_op due to deprecation of the latter.

  jirikraus

  committed
  0c65eee

  Copy full SHA for 0c65eee

• Fixes issue #10: cannot compile multi_threaded_copy{,_overlap} because of argument mismatch to std::swap

  jirikraus

  committed
  79a3614

  Copy full SHA for 79a3614

Commits on Oct 1, 2021

• Fixed communication warm up loop.

  jirikraus

  committed
  98bbe89

  Copy full SHA for 98bbe89

Commits on Sep 29, 2021

• Fixed typo for NCCL overlap variant

  jirikraus

  committed
  ae4d83a

  Copy full SHA for ae4d83a

• Fixed typo for multi threaded copy overlap variant

  jirikraus

  committed
  d386ed4

  Copy full SHA for d386ed4

• Fixed typo for MPI overlap variant.

  jirikraus

  committed
  3dd5d9a

  Copy full SHA for 3dd5d9a

• Added high priority stream option

  jirikraus

  committed
  0bed88a

  Copy full SHA for 0bed88a

Commits on Apr 13, 2021

• Added more links to GTC Talk recordings.

  jirikraus

  authored
  f5a4de1

  Copy full SHA for f5a4de1

Commits on Mar 24, 2021

• Fixed NVSHMEM_SYMMETRIC_SIZE

  jirikraus

  committed
  d6ee330

  Copy full SHA for d6ee330

Commits on Mar 23, 2021

• DGX A100 updates for bench.sh

  jirikraus

  committed
  c85a514

  Copy full SHA for c85a514

Commits on Mar 12, 2021

• Made NCCL 2.8+ a requriment

  jirikraus

  committed
  144408f

  Copy full SHA for 144408f

Commits on Mar 10, 2021

• Added NCCL overlapp variant to bensh script.

  jirikraus

  committed
  776b234

  Copy full SHA for 776b234

• Updated sanitize and profile make targets for compute-sanitizer and NSight Systems

  jirikraus

  committed
  3dbb24a

  Copy full SHA for 3dbb24a

• Added NCCL overlapp variant.

  jirikraus

  committed
  16743b2

  Copy full SHA for 16743b2

• Added NCCL version check

  jirikraus

  committed
  ad8a885

  Copy full SHA for ad8a885

Pagination

Previous

Next