NVIDIA Collective Communication Library (NCCL) Documentation

Contents:

• Overview of NCCL
• Setup
• Using NCCL
  • Creating a Communicator
    • Creating a communicator with options
    • Creating a communicator using multiple ncclUniqueIds
    • Creating more communicators
    • Using multiple NCCL communicators concurrently
    • Finalizing a communicator
    • Destroying a communicator
  • Error handling and communicator abort
    • Asynchronous errors and error handling
  • Fault Tolerance
  • Collective Operations
    • AllReduce
    • Broadcast
    • Reduce
    • AllGather
    • ReduceScatter
  • Data Pointers
  • CUDA Stream Semantics
    • Mixing Multiple Streams within the same ncclGroupStart/End() group
  • Group Calls
    • Management Of Multiple GPUs From One Thread
    • Aggregated Operations (2.2 and later)
    • Nonblocking Group Operation
  • Point-to-point communication
    • Sendrecv
    • One-to-all (scatter)
    • All-to-one (gather)
    • All-to-all
    • Neighbor exchange
  • Thread Safety
  • In-place Operations
  • Using NCCL with CUDA Graphs
  • User Buffer Registration
    • NVLink Sharp Buffer Registration
    • IB Sharp Buffer Registration
    • General Buffer Registration
    • Memory Allocator
• NCCL API
  • Communicator Creation and Management Functions
    • ncclGetLastError
    • ncclGetErrorString
    • ncclGetVersion
    • ncclGetUniqueId
    • ncclCommInitRank
    • ncclCommInitAll
    • ncclCommInitRankConfig
    • ncclCommInitRankScalable
    • ncclCommSplit
    • ncclCommFinalize
    • ncclCommDestroy
    • ncclCommAbort
    • ncclCommGetAsyncError
    • ncclCommCount
    • ncclCommCuDevice
    • ncclCommUserRank
    • ncclCommRegister
    • ncclCommDeregister
    • ncclMemAlloc
    • ncclMemFree
  • Collective Communication Functions
    • ncclAllReduce
    • ncclBroadcast
    • ncclReduce
    • ncclAllGather
    • ncclReduceScatter
  • Group Calls
    • ncclGroupStart
    • ncclGroupEnd
    • ncclGroupSimulateEnd
  • Point To Point Communication Functions
    • ncclSend
    • ncclRecv
  • Types
    • ncclComm_t
    • ncclResult_t
    • ncclDataType_t
    • ncclRedOp_t
    • ncclScalarResidence_t
    • ncclConfig_t
    • ncclSimInfo_t
  • User Defined Reduction Operators
    • ncclRedOpCreatePreMulSum
    • ncclRedOpDestroy
• Migrating from NCCL 1 to NCCL 2
  • Initialization
  • Communication
  • Counts
  • In-place usage for AllGather and ReduceScatter
  • AllGather arguments order
  • Datatypes
  • Error codes
• Examples
  • Communicator Creation and Destruction Examples
    • Example 1: Single Process, Single Thread, Multiple Devices
    • Example 2: One Device per Process or Thread
    • Example 3: Multiple Devices per Thread
    • Example 4: Multiple communicators per device
  • Communication Examples
    • Example 1: One Device per Process or Thread
    • Example 2: Multiple Devices per Thread
• NCCL and MPI
  • API
    • Using multiple devices per process
    • ReduceScatter operation
    • Send and Receive counts
    • Other collectives and point-to-point operations
    • In-place operations
  • Using NCCL within an MPI Program
    • MPI Progress
    • Inter-GPU Communication with CUDA-aware MPI
• Environment Variables
  • System configuration
    • NCCL_SOCKET_IFNAME
      • Values accepted
    • NCCL_SOCKET_FAMILY
      • Values accepted
    • NCCL_SOCKET_NTHREADS
      • Values accepted
    • NCCL_NSOCKS_PERTHREAD
      • Values accepted
    • NCCL_CROSS_NIC
      • Values accepted
    • NCCL_IB_HCA
      • Values accepted
    • NCCL_IB_TIMEOUT
      • Values accepted
    • NCCL_IB_RETRY_CNT
      • Values accepted
    • NCCL_IB_GID_INDEX
      • Values accepted
    • NCCL_IB_ADDR_FAMILY
      • Values accepted
    • NCCL_IB_ADDR_RANGE
      • Values accepted
    • NCCL_IB_ROCE_VERSION_NUM
      • Values accepted
    • NCCL_IB_SL
      • Values accepted
    • NCCL_IB_TC
      • Values accepted
    • NCCL_IB_RETURN_ASYNC_EVENTS
      • Values accepted
    • NCCL_OOB_NET_ENABLE
      • Values accepted
    • NCCL_OOB_NET_IFNAME
      • Values accepted
    • NCCL_UID_STAGGER_THRESHOLD
      • Values accepted
    • NCCL_UID_STAGGER_RATE
      • Values accepted
    • NCCL_NET
      • Values accepted
    • NCCL_NET_PLUGIN
      • Values accepted
    • NCCL_TUNER_PLUGIN
      • Values accepted
    • NCCL_PROFILER_PLUGIN
      • Values accepted
    • NCCL_IGNORE_CPU_AFFINITY
      • Values accepted
    • NCCL_CONF_FILE
      • Values accepted
    • NCCL_DEBUG
      • Values accepted
    • NCCL_DEBUG_FILE
      • Values accepted
    • NCCL_DEBUG_SUBSYS
      • Values accepted
    • NCCL_COLLNET_ENABLE
      • Value accepted
    • NCCL_COLLNET_NODE_THRESHOLD
      • Value accepted
    • NCCL_TOPO_FILE
      • Value accepted
    • NCCL_TOPO_DUMP_FILE
      • Value accepted
    • NCCL_SET_THREAD_NAME
      • Value accepted
  • Debugging
    • NCCL_P2P_DISABLE
      • Values accepted
    • NCCL_P2P_LEVEL
      • Values accepted
      • Integer Values (Legacy)
    • NCCL_P2P_DIRECT_DISABLE
      • Values accepted
    • NCCL_SHM_DISABLE
      • Values accepted
    • NCCL_BUFFSIZE
      • Values accepted
    • NCCL_NTHREADS
      • Values accepted
    • NCCL_MAX_NCHANNELS
      • Values accepted
    • NCCL_MIN_NCHANNELS
      • Values accepted
    • NCCL_CHECKS_DISABLE
      • Values accepted
    • NCCL_CHECK_POINTERS
      • Values accepted
    • NCCL_LAUNCH_MODE
      • Values accepted
    • NCCL_IB_DISABLE
      • Values accepted
    • NCCL_IB_AR_THRESHOLD
      • Values accepted
    • NCCL_IB_QPS_PER_CONNECTION
      • Values accepted
    • NCCL_IB_SPLIT_DATA_ON_QPS
      • Values accepted
    • NCCL_IB_CUDA_SUPPORT
      • Values accepted
    • NCCL_IB_PCI_RELAXED_ORDERING
      • Values accepted
    • NCCL_IB_ADAPTIVE_ROUTING
      • Values accepted
    • NCCL_MEM_SYNC_DOMAIN
      • Values accepted
    • NCCL_CUMEM_ENABLE
      • Values accepted
    • NCCL_CUMEM_HOST_ENABLE
      • Values accepted
    • NCCL_NET_GDR_LEVEL (formerly NCCL_IB_GDR_LEVEL)
      • Values accepted
      • Integer Values (Legacy)
    • NCCL_NET_GDR_READ
      • Values accepted
    • NCCL_NET_SHARED_BUFFERS
      • Value accepted
    • NCCL_NET_SHARED_COMMS
      • Value accepted
    • NCCL_SINGLE_RING_THRESHOLD
      • Values accepted
    • NCCL_LL_THRESHOLD
      • Values accepted
    • NCCL_TREE_THRESHOLD
      • Values accepted
    • NCCL_ALGO
      • Values accepted
    • NCCL_PROTO
      • Values accepted
    • NCCL_NVB_DISABLE
      • Value accepted
    • NCCL_PXN_DISABLE
      • Value accepted
    • NCCL_P2P_PXN_LEVEL
      • Value accepted
    • NCCL_RUNTIME_CONNECT
      • Value accepted
    • NCCL_GRAPH_REGISTER
      • Value accepted
    • NCCL_LOCAL_REGISTER
      • Value accepted
    • NCCL_SET_STACK_SIZE
      • Value accepted
    • NCCL_GRAPH_MIXING_SUPPORT
      • Value accepted
    • NCCL_DMABUF_ENABLE
      • Value accepted
    • NCCL_P2P_NET_CHUNKSIZE
      • Values accepted
    • NCCL_P2P_LL_THRESHOLD
      • Values accepted
    • NCCL_ALLOC_P2P_NET_LL_BUFFERS
      • Values accepted
    • NCCL_COMM_BLOCKING
      • Values accepted
    • NCCL_CGA_CLUSTER_SIZE
      • Values accepted
    • NCCL_MAX_CTAS
      • Values accepted
    • NCCL_MIN_CTAS
      • Values accepted
    • NCCL_NVLS_ENABLE
      • Values accepted
    • NCCL_IB_MERGE_NICS
      • Values accepted
    • NCCL_MNNVL_ENABLE
      • Values accepted
• Troubleshooting
  • Errors
  • GPU Direct
    • GPU-to-GPU communication
    • GPU-to-NIC communication
    • PCI Access Control Services (ACS)
  • Topology detection
  • Shared memory
    • Docker
    • Systemd
  • Networking issues
    • IP Network Interfaces
    • IP Ports
    • InfiniBand
    • RDMA over Converged Ethernet (RoCE)

Indices and tables

• Index
• Module Index
• Search Page