Search results

[Question]: cudaMemcpyBatchAsync with cudaMemcpyFlagPreferOverlapWithCompute: D2D (local copy) bandwidth is unexpectedly low compared to SM-based copy

question

Status: Open.

#2177 In NVIDIA/nccl;

· pkuleo opened on May 14, 2026

[Question]: Understanding when GIN symmetric kernels improve performance

question

Status: Open.

#2176 In NVIDIA/nccl;

· WWeiOne opened on May 14, 2026

NCCL within Cloud Hypervisor VM is 25% of baremetal speed.

question

Status: Open.

#2171 In NVIDIA/nccl;

· tchaton opened on May 12, 2026

[Issue]: vLLM serve with tensor-parallel-size=8 on Kubernetes + vGPU fails: NCCL TCPStore broken pipe, EngineCore initialization failed #42389

Status: Open.

#2170 In NVIDIA/nccl;

· sumufuyun opened on May 12, 2026

[Question]: alltoall throughput drops sharply above 512 MB message size

question

Status: Open.

#2169 In NVIDIA/nccl;

· erfiboy opened on May 11, 2026

[BUG] NCCL 2.30.4 NVLS code path hangs MoE expert weight loading on 8× DGX Spark (sm_121, no NVLink); 2.30.3 works

Status: Open.

#2167 In NVIDIA/nccl;

· idonati opened on May 10, 2026

[Issue]: GIN may choose wrong NIC when net plugin is used

Status: Open.

#2165 In NVIDIA/nccl;

· fishautumn opened on May 8, 2026

[RFE]: Replace assert() with proper error handling

enhancement

Status: Open.

#2163 In NVIDIA/nccl;

· LeviMee opened on May 8, 2026

[Question]: Do we have any plan to integrate the nccl ep to megatron?

question

Status: Open.

#2162 In NVIDIA/nccl;

· gangxie112 opened on May 8, 2026

NCCL_TOPO_XML_MAX_NODES=256 limit hit during intra-node XML fusion on 32-NIC hosts (AWS p5.48xlarge)

Status: Open.

#2160 In NVIDIA/nccl;

· dmvevents opened on May 7, 2026

[Question]: NCCL 2.28.9 compatibility with V100 (sm_70) — segfault in ncclCommInitRank

question

Status: Open.

#2159 In NVIDIA/nccl;

· Tchoupi2001 opened on May 6, 2026

[Issue]: NIC recovery failure

Status: Open.

#2157 In NVIDIA/nccl;

· Mitchelljdaw opened on May 4, 2026