microsoft/mscclpp

Issues

• [Bug] Hang when using ProxyChan and one GPU is sending zeros bytes.

  #432 opened 3 days ago by FC-Li
  1

• [Bug] C++ exception with description "ibv_modify_qp failed (errno 19) (Ib failure: No such device)" thrown in the test body.

  #431 opened 2 days ago by FC-Li
  9

• [Bug] Proxy channel over CudaIPC on AMD GPUs

  #418 opened 15 days ago by liangyuRain
  1

• [Feature] Using the AMD Infinity Fabric in SMChannels

  #408 opened 16 days ago by ThomasNing
  4

• [Feature] Cross-node communication without InfiniBand

  #398 opened 18 days ago by chenhongyu2048
  4

• [Bug] There is not any InfiniBand or NVLink in my 4-GPU machine, how can I use mscclpp to communicate?

  #397 opened a month ago by Maphsge4
  1

• Question about DeviceSyncer and Ensuring Synchronization

  #395 opened a month ago by hidva
  1

• [Bug] Proxy chan hang at cudaMemcpyAsync

  #394 opened a month ago by FC-Li
  1

• [Bug] Hang when one GPU get a new BasePtr while others not.

  #388 opened a month ago by FC-Li
  1

• [Bug] run nccl_api_test failed

  #387 opened a month ago by yizhang2077
  1

• [Bug] flush() hang bug.

  #377 opened 2 months ago by TonyWu199
  0

• [Perf] Failed to reproduce the performance result for Single-node AllReduce mentioned in README.md

  #362 opened 2 months ago by FC-Li
  5

• [Feature] Immediate data upon signal & wait

  #368 opened 2 months ago by chhwang
  0

• [Feature] Can NVIDIA and AMD communicate?

  #361 opened 3 months ago by liuyang6055
  1

• [Bug] Can't launch allreduce test

  #359 opened 3 months ago by chenhongyu2048
  1

• [Bug] libmscclpp_nccl fails linking using ROCm 6.0

  #349 opened 3 months ago by corey-derochie-amd
  2

• Is there exist some documentation to explain the difference between allreduce algorithm in mscclpp?

  #350 opened 4 months ago by MARD1NO
  4

• [Feature] How to run on a machine having multi-gpus with one infiniband?

  #254 opened a year ago by jhlee508
  9

• [Bug] `mscclpp/concurrency_device.hpp: No such file or directory`

  #335 opened 5 months ago by TZHelloWorld
  2

• [Bug] (Ib failure: Cannot allocate memory) reported when run mscclpp-test/allreduce_test_perf with MPI on 2 nodes

  #323 opened 5 months ago by dong-liuliu
  4

• [Bug]mscclpp::DeviceSyncer::sync Assertion failed

  #320 opened 5 months ago by linstreamer
  1

• [Feature] NPKit support

  #206 opened 6 months ago by chhwang
  0

• [Bug] Bugs in mp_unit_test.

  #315 opened 6 months ago by TonyWu199
  3

• How to use mscclpp as a backend in pytorch

  #311 opened 6 months ago by wangfakang
  4

• [Bug] Program hangs at proxy channel `wait()`

  #285 opened 6 months ago by liangyuRain
  1

• [Bug]mscclpp-tests dont exit after test.

  #282 opened 6 months ago by TonyWu199
  3

• [Bug] __assert_fail declaration in mscclpp breaks "assert()" usage in host functions.

  #302 opened 7 months ago by Alkaid-Benetnash
  1

• MSCCL++ Low-priority Work Items

  #199 opened a year ago by chhwang
  0

• MSCCL++ v0.5.0 Release Plan

  #281 opened 8 months ago by chhwang
  1

• [Feature] `deviceHandle()` interface is counter-intuitive

  #289 opened 9 months ago by chhwang
  0

• Using MSCCL++ smchannel/proxy_channel within Triton Kernels

  #288 opened 9 months ago by rajagond
  1

• [Feature] Usage as backend in Pytorch

  #287 opened 9 months ago by azharlightelligence
  1

• [Bug] Is there a known bug with `Driver Version: 535.129.03` which cases `MscclppAllReduce3` to hang?

  #260 opened a year ago by saeedmaleki
  5

• [Bug]Meet IB problem in single node experiment?

  #274 opened 10 months ago by TonyWu199
  1

• [Doc] Inquiry on MSCCL++ Algorithms

  #269 opened 10 months ago by jhlee508
  1

• [Bug] getting error for `allreduce_bench.py`

  #266 opened 10 months ago by saeedmaleki
  1

• [Feature] `CommGroup` method names are confusing

  #265 opened a year ago by chhwang
  0

• [Bug] python benchmark hang for small message size on AMD GPU

  #263 opened a year ago by Binyang2014
  0

• [Doc] What are the differences between this repo mscclpp and msccl?

  #253 opened a year ago by YJHMITWEB
  4

• [Feature] gets get rid of make pylib-copy

  #216 opened a year ago by saeedmaleki
  0

• [Bug] Error when creating many proxy channels

  #242 opened a year ago by liangyuRain
  7

• [Perf] Relaxed atomic for FIFO push

  #226 opened a year ago by chhwang
  0

• [Doc] In the quickstart.md file, the argument for the mpirun should be '--bind-to numa' instead of '--bind-to-numa'.

  #228 opened a year ago by sphish
  1

• [Perf] Have you tuned the performance of single-node all-to-all communication？

  #227 opened a year ago by sphish
  2

• [Bug] Get error when using IB to transfer large size of message

  #224 opened a year ago by Binyang2014
  1

• [Feature] Improving code from mscclpp-test and examples

  #207 opened a year ago by chhwang
  2

• [Perf] Two-node allreduce perf improvement

  #210 opened a year ago by Binyang2014
  1

• [Feature] Enhance Python benchmark

  #215 opened a year ago by Binyang2014
  0

• [Bug] When addMemory for more than 2^8 registered memories to a single proxy service, the system should throw an exception.

  #212 opened a year ago by saeedmaleki
  0

• [Feature] Topology detection from the topo XML file

  #205 opened a year ago by chhwang
  0