NVIDIA nccl issues - Githubissues

NVIDIA / nccl

Optimized primitives for collective multi-GPU communication

Other

3.27k stars 826 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

local access violation work queue error when upgrade to v2.20.3-1

#1524 gangxie112 opened 13 hours ago
7
Questions about the FIFO of simple protocol

#1523 JK-Jiagn opened 1 day ago
0
Any possibility/plan to support fused kernels?

#1522 dearsxx0918 opened 1 day ago
0
Why group calls (`ncclGroupStart()` and `ncclGroupEnd()`) are invoked in `ncclSend()` and `ncclRecv()`

#1521 ZhiyiHu1999 opened 4 days ago
0
Is it safe or recommended to use multiple communicators for real distributed training

#1520 ZhiyiHu1999 opened 1 week ago
0
Unable to use multiple NICs

#1519 thecodingwizard opened 1 week ago
4
Problem with SHMEM creation on startup.

#1518 davies-w closed 1 week ago
4
torch.distributed.DistBackendError: NCCL error

#1517 Chevolier opened 1 week ago
1
Could not enable P2P between devices

#1516 ZhiyiHu1999 opened 1 week ago
1
Nccl socketStartConnect: Connect to x.x.x.x<xxxx> failed : Software caused connection abort

#1515 913871734 opened 1 week ago
4
torch.distributed.DistBackendError: NCCL error in ProcessGroupNCCL.cpp:1275

#1514 shenshaowei opened 1 week ago
1
nccl capture error

#1513 freshduer opened 1 week ago
0
How does nccl regtister the memory region by ibv_reg_xxxx ?

#1512 jinhao2 closed 1 week ago
1
nccl-test Indicates a performance problem

#1511 yalbaba opened 2 weeks ago
13
GPU Direct RDMA Disabled for HCA

#1510 hiennguyennq opened 2 weeks ago
3
NCCL hung witih NCCL_P2P_USE_CUDA_MEMCPY=1 by pytorch

#1509 adofirst2018 opened 2 weeks ago
5
what is the count value in allreduce function?

#1508 BSkim26 opened 2 weeks ago
0
align the peer mem access check with API document

#1507 changchengx opened 2 weeks ago
3
allgather performance using NVLS is poor

#1506 telala opened 2 weeks ago
13
Question on DGX A100 GPU topologies

#1505 YJHMITWEB closed 2 weeks ago
2
Encounter NCCL error when runing Pytorch example code

#1504 Noblezhong opened 3 weeks ago
5
Difference between readLL() and readLLFinish() in prims_ll.h

#1503 ZhiyiHu1999 opened 3 weeks ago
0
nccl topo about PHB and NODE

#1502 jianzi123 opened 3 weeks ago
1
Makefile: Add nccl_common.h and nccl_tuner.h to INCEXPORTS

#1501 martin-belanger closed 3 weeks ago
4
Can NCCL and model forward propagation (CUDA matrix operations) be executed simultaneously, and if so, how can this be achieved?

#1500 liweiqing1997 opened 3 weeks ago
0
ncclInternalError: Internal check failed

#1499 whiteyn opened 3 weeks ago
3
Question about ncclTopoConnectNodes in topo.cc

#1498 JK-Jiagn closed 3 weeks ago
2
same data all reduce on H20, but results are different

#1497 Rainlin007 opened 4 weeks ago
10
result of sendrecv_perf is wrong

#1496 yanminjia opened 4 weeks ago
1
P2P in non-blocking mode

#1495 kwen2501 opened 1 month ago
3
Alternating rings cause bad performance (NIC sending PFC) in a cluster with mixed crossNic=0/1 nodes

#1494 huzhiwen93 opened 1 month ago
0
A question about sequences of functions called in nccl/src/transport /net.cc

#1493 ZhiyiHu1999 opened 1 month ago
0
NCCL Logs Questions

#1492 gjit-juniper opened 1 month ago
0
Question about tree channel

#1491 networkResearcher opened 1 month ago
0
Question about the topology of double binary tree

#1490 networkResearcher opened 1 month ago
0
Fix comm ready check in ncclIbIsend and ncclIbIrecv

#1489 WWeiOne opened 1 month ago
1
nvlsAllocateMem report error

#1488 shanleo2024 closed 4 weeks ago
7
1000x latency with `ncclSend` and `ncclRecv`

#1487 goelayu closed 1 month ago
10
NCCL WARN socketProgress: Connection closed by remote peer

#1486 ganyu1992 opened 1 month ago
2
The internal errors caused by host-side multi-threads

#1485 themoonstone opened 1 month ago
0
How to use the profiler plugin?

#1484 jxh314 opened 1 month ago
1
what is the nvidia IMEX channels in Multi-Node NVLink (MNNVL)

#1483 vvmex opened 1 month ago
1
The reason for breakdown "NCCL WARN comm 0x55ec10690820 has already been destroyed"

#1482 chenhongyu2048 closed 1 month ago
2
what will happen if i call ncclAbort after ncclSend

#1481 freshduer opened 1 month ago
0
The nsys profile will hang when NCCL_P2P_USE_CUDA_MEMCPY is enabled

#1480 PhdShi opened 1 month ago
5
question about PXN

#1479 anchenchai opened 1 month ago
0
Some question about NCCL_IB_SCA

#1478 shanleo2024 closed 1 month ago
2
How to interpret NCCL's NVTX data

#1477 martin-belanger closed 1 month ago
1
Question about hostStream, deviceStream and userStream

#1476 MC952-arch opened 1 month ago
0
Does FabricManager support isolation at GPU card granularity in MNNVL envrionmant?

#1475 hailiyidishui opened 1 month ago
1