Symmetric memory API and symmetric kernels
* Redesign from the ground up, enabling major latency and bandwidth
improvements.
* Add new API calls to register user-allocated memory among communicator
ranks into a NCCL window: ncclCommWindowRegister() and
ncclCommWindowDeregister(). The calls currently support symmetric
registration for P2P and NVLS, and require VMM memory buffers (i.e.,
CUMEM must be operational).
* Implement specialized kernels taking advantage of symmetrically
registered memory, with performance gains expected particularly for
small to medium message sizes.
* The kernels support 32 bit floating point types and smaller, and sum as
the reduction operator, with no more than one collective operation per
group.
* Floating point summation is always done in fp32 accumulators (with the
exception of fp8 on NVLS, where it uses fp16 inside the switch). Thus,
the accuracy with fp8 and fp16 data types should be much improved.
* This initial implementation supports non-network communicators only (P2P
and NVLS transports).
* To explore this functionality users need to use the new memory
registration API calls with the NCCL_WIN_COLL_SYMMETRIC flag and all
ranks of a communicator must pass buffers at the same offset in the same
registration when invoking a collective NCCL operation.
Add support for DGX Spark.
Add support for DirectNIC (CX8) to the internal IB plugin.
Add a new ncclCommShrink() API call
* It is a non-collective call similar to ncclCommSplit(), which makes it
possible to exclude some (possibly unresponsive) ranks from the parent
communicator.
Add support for loading multiple network plugins
* This enables the creation of generic containers that can work across a
range of providers.
* Allow NCCL_NET_PLUGIN to accept a comma-separated list of plugins to
load.
NVLink SHARP (NVLS) improvements
* Implement NVLS+IB SHARP support for AllGather and ReduceScatter with
user buffer registration. This improves performance and reduces the
number of CTAs needed to achieve peak bandwidth.
* Gracefully fall back by default to other transports if NVLS
initialization fails (the old behavior of returning an error code from a
NCCL call can be preserved by setting NCCL_NVLS_ENABLE=1).
* Decrease the NVLS channel count to 24 on Blackwell systems with multiple
NVLink domains per communicator.
* Enable fine-tuning of NCCL behavior per communicator using new
"ncclConfig_t" members "collnetEnable", "CTAPolicy", and "nvlsCTAs".
Profiler improvements
* Extend the init function by adding communicator name, comm id (hash),
rank, number of ranks, number of nodes, and the NCCL log function to the
argument list. This makes the name and the comm id available to all
events in the communicator without explicitly passing them to each
individual event. Add the communicator id and rank to the profiler trace
filename. Now, the communicator name can be set via a new "ncclConfig_t"
member "commName".
* Improve the accuracy of the GPU kernel events by providing GPU-generated
timestamps for the start and stop of every NCCL operation.
* Harmonize proxy events, removing overlaps between ProxyOp and ProxyStep
states.
* Add support for network-defined event updates (through
"recordEventState").
* Report the correct number of channels used by every collective/p2p
operation (used to be set to nMaxChannels for collectives and absent for
p2ps).
* Fix the logic on proxyCtrl Idle/Active events (Issue #1162).
* Fix an issue where the network proxy profiler could lose track of an
event identifier (Issue #1682).
* Improve the backward compatibility with plugins older than v4.
* Ensure that the work counters are 0-initialized.
* Fix a potential race condition in the network profiler that could result
in an event being linked to a wrong parent.
MNNVL improvements
* Increase to 16 the number of NICs used to communicate between MNNVL
domains on GB200 systems, to optimize the performance of collective
operations.
* Add support for more complex MNNVL topologies with up to 32 NICs per
node.
* If the MNNVL fabric initialization was unsuccessful, NCCL will now fail
by default, so as to avoid inadvertently falling back to a potentially
much slower network transport. Such failures are typically due to a
misconfigured IMEX support on the system. To continue without MNNVL,
restart the job with NCCL_MNNVL_ENABLE=0.
* Fix a potential hang in alltoall-like communication patterns at a scale
of over 80 ranks.
* Make NCCL_P2P_DISABLE=1 imply NCCL_MNNVL_ENABLE=0 (so the latter no
longer needs to be specified on MNNVL systems).
* Fix an initialization failure when NCCL_TOPO_FILE is used on MNNVL
systems.
* Fix the graph search to exclude non-local NICs.
* Fix the SHM transport to use fabric handles on MNNVL systems.
NIC Fusion improvements
* Disable the creation of fused NICs for physical devices that haven't
been merged.
* Flatten multiple ports to a single PCI device within the internal IB
plugin and reparent dual-port NICs under the first PCI parent. If the
parent is not a PCI switch, PCI devices for fused NICs won't be
duplicated.
* Route traffic on GB200-CX8 systems through DirectNIC, not the host
interface.
Improve support for platforms with C2C connectivity (e.g., GB200)
* Enable GPUDirect RDMA for the NICs by default.
* Add support for P2C (PXN over C2C) and the LL128 protocol.
Extend NCCL fault tolerance in multithreaded scenarios
* Support the creation of multiple nonblocking communicators within a
single group and polling in parallel for the completion using multiple
threads (one per communicator).
Enable ncclImplicitOrderLaunch for CUDA 12.9+
* This can potentially speed up NCCL_IMPLICIT_LAUNCH_ORDER.
Improve the netSocket transport latency and control
* Provide finer control over the size of the socket send/receive buffers,
the task size, and the number of sockets that a single peer can open.
* Add support for the inlining of small messages behind the header when
using multiple sockets per connection.
Improve the readability of the CPU affinity in the debug output
* Print it as a range string rather than a bitmask.
Fix a potential race condition in graph execution
* A contention could arise when mixing graph and non-graph execution.
Improve PXN connection code
* Avoid duplicate and unused connections.
RAS fixes
* Fix a memory corruption at job termination time in case of a previously
failed initialization of a RAS socket connection.
* Fix a race condition leading to a crash when generating a RAS report
during communicator initialization (Issues #1669, #1718).
* Fix a potential race condition when gathering data for a RAS status
report.
Fix a potential memory corruption in ncclCommSplit()
* Memory could get corrupted when resource sharing was in use and the size
of the NVLink domain in the new communicator was smaller than in the old
one.
Fix asynchronous graph upload
* Fix a small memory leak.
* Fix oversychronization.
Add a check for out-of-memory conditions in ncclMemAlloc()
Clean up the NCCL socket code
* accept() will retry also if just reading the magic failed (Issue #1613).
* connect() will retry also if poll() did not return a POLLOUT event
(Issue #1618).
* Add error checking in a few instances (Issue #1539).
* Fix the loop condition in ncclFindInterfaceMatchSubnet() (Issue #1574).
* Clean up the debug output, downgrading WARN messages to INFO in
non-critical cases, and printing the peer's address where relevant.
Switch NCCL_DEBUG_FILE to line buffering
* This should help avoid mixed-up partial output lines in multithreaded
cases.
Other minor fixes
* Improve the checks for buffer overflows in the graph code (Issue #1585).
* Extend logging and state clearing to all four events in the internal IB
plugin (Issue #1650).
* Fix the error path in case IB communication is not ready (Issue #1489).
* Add ECE logging for IB fabric.
* Fix various minor issues in the graph module (Issue #1635).
* Clean up the debug output in the graph code, downgrading WARN messages
to INFO in non-critical cases.
* Add a missing argument to a directSend() call (Issue #1628).
* Remove duplicate code in sendProxySetup() (Issue #1420).
* Fix the order of arguments of cudaDeviceCanAccessPeer() (Issue #1507).
* Fix compiler warnings with GCC 14.
* Fix a typo in a comment (Issue #1236).
Profiler improvements
* Add events for CUDA kernel start and end.
* Allow network plugins to generate profiling events
* Enable profiling on a per-operation basis, rather than per-communicator.
* Add support for graph capturing.
Add implicit launch order
* Allow to prevent deadlocks when using multiple NCCL communicators per
device by implicitly ordering NCCL operations using the host program
order. Disabled by default, set NCCL_LAUNCH_ORDER_IMPLICIT=1 to enable.
* Add a complementary mechanism to detect host threads racing to launch
to the same device. Enabled by default, set NCCL_LAUNCH_RACE_FATAL=0 to
disable.
Optimize the PAT algorithm
* Separate the computation and execution of PAT steps on different warps,
allowing to run up to 16 PAT steps in parallel to significantly
accelerate PAT and reduce its linear part.
Add support for setting QoS per communicator
* Add a new trafficClass field to the communicator configuration, to
allow the application to select a particular traffic class for a
given communicator. The meaning of the traffic class is
network-specific and should be set in accordance with the network
configuration.
* For the IB/RoCE plugin, existing config variables such as NCCL_IB_SL
and NCCL_IB_TC take precedence.
Allow to enable GPU Direct RDMA specifically on C2C platforms
* Disabled by default, set NCCL_NET_GDR_C2C=1 to enable.
Do not disable user buffer registration unless PXN is really used
* Only disable UB when a communicator has more than one rank per
node on any node.
RAS subsystem improvements
* Report operation counts separately for each collective operation type.
* Provide details about missing communicator ranks and reliably
distinguish ranks that are no longer a given communicator's members
(now reported as NOCOMM) from those that failed to respond.
Add support for timestamps to NCCL diagnostic messages
* On by default for WARN messages; NCCL_DEBUG_TIMESTAMP_LEVELS can be
used to enable them for other debug levels as well.
* The format can be changed using the NCCL_DEBUG_TIMESTAMP_FORMAT config
variable.
Reduce the memory usage with NVLink SHARP (NVLS)
* Potentially save hundreds of MBs of device memory, considering the
multicast buffer size granularity separately from the address alignment.
Update performance tuning for recent Intel CPUs
* Improve algorithm/protocol selection on recent CPUs such as Emerald
Rapids and Sapphire Rapids.
Improve channel scheduling when mixing LL and Simple operations.
* Make LL operations account for 4x more traffic to ensure LL and simple
operations complete at the same time.
Refactor the plugin code
* Clean up and harmonize the support code across the network, tuner,
and profiler plugins.
Add support for comment lines (starting with #) in the nccl.conf file
* Issue #1540.
Make user buffer registration problems print an INFO instead of a WARN.
Drop support for network plugin interface version 5.
Fix a race condition with split-shared communicators
* NCCL could hang during connection setup if multiple communicators
were grouped together that share resources.
Fix a performance regression when using NCCL_CROSS_NIC=1
* NCCL would unnecessarily alternate rings, breaking the GPU-NIC
associations.
Make GID index detection code more resilient
* Dynamic GID detection code was giving up too soon if the
detected index was not available (e.g., wasn't mapped to the
container's sysfs).
* Issues #1538, #1573.
Fix a race condition with non-blocking operation
* Fix issue when creating a non-blocking communicator after a non-
blocking collective operation on another communicator.
Fix shared memory usage on recent Blackwell GPUs.
* Issues NVIDIA/nccl-tests#287, NVIDIA/nccl-tests#291, #1637.
Fix an error with NIC fusion and IB SHARP when recreating communicators
* Disable the unloading of network plugins
Make the auto-merge failures in the NIC fusion non-fatal
* This could happen when trying to merge IB and RoCE devices.
Fixes to ncclCommAbort
* Fix hangs due to the progress thread spinning indefinitely on the
network progress.
* Reduce the abort time by up to two orders of magnitude.
Fix a crash when libnccl.so was dynamically unloaded
* The RAS subsystem was missing a clean-up handler.
Fix a hang if the network plugin's test() call returns an error.
Fix a hang on heterogeneous architectures
* Ensure we harmonize the tuning to avoid different tuning choices,
causing a hang.
Fix double-free on failed ncclCommInitRank and ncclCommFinalize.
Fix a potential list traversal bug during a group launch of multiple
communicators
* Issue #1599.
Unify the handling of NCCL configuration variables
* Under rare circumstances, some variables specified in the config file
could be ignored.
Rework core for NVIDIA Trusted Computing
* Compress work structs so that they are shared between channels
* Utilize the full amount of kernel argument space permitted (4k)
before resorting to work fifo.
* Rework the task preprocessing phase.
* Use a separate abortDevFlag which is kept in sync with abortFlag
using cudaMemcpy operations.
* Rename src/include/align.h to src/include/bitops.h
Add lazy connection establishment for collective operations
* Move buffer allocation and connection establishment to the first
collective operation using that algorithm.
* Accelerate init time and reduce memory usage.
* Avoid allocating NVLS buffers if all calls are registered.
* Compute algo/proto in ncclLaunchCollTasksInfo early on.
* Connect peers in ncclCollPreconnectFunc if not connected already.
* Also move shared buffer creation to the first send/recv call.
Accelerate intra-node NVLink detection
* Make each rank only detect NVLinks attached to its GPU.
* Fuse XMLs to reconstruct the full NVLink topology
Add init profiling to report time spend in different init phases.
* Report timings of bootstrap, allgather, search, connect, etc.
* Add new "PROFILE" category for NCCL_DEBUG_SUBSYS.
Add support for PCI p2p on split PCI switches
* Detect split PCI switches through a kernel module exposing
switch information.
* Update the topology XML and graph to add those inter-switch
connections.
Add cost estimation API
* Add a new ncclGroupEndSimulate primitive to return the estimated
time a group would take.
Net/IB: Add separate traffic class for fifo messages
* Add NCCL_IB_FIFO_TC to control the traffic class of fifo messages
independently from NCCL_IB_TC.
Merges PR #1194
Net/IB: Add support for IB router
* Use flid instead of lid if subnets do not match
* Warn if flid is 0
Optimizations and fixes for device network offload (unpack)
* Double the default number of channels
* Cache netDeviceType
* Fix save/increment head logic to enable Tree support.
Support ncclGroupStart/End for ncclCommAbort/Destroy
* Allow Abort/Destroy to be called within a group when managing
multiple GPUs with a single process.
Improve Tuner API
* Provide to the plugin the original cost table so that the plugin
can leave unknown or disabled algo/proto combinations untouched.
* Remove nvlsSupport and collnetSupport.
Do not print version to stdout when using a debug file
* Also print version from all processes with INFO debug level.
Fixes issue #1271
Fix clang warnings in NVTX headers
* Update NVTX headers to the latest version
Fixes issue #1270
Disable port fusion in heterogeneous systems
* Do not fuse ports if a mix of multi-port and single port are detected.
Fix NVLS graphs search for dual NICs.
* Fix NVLS graph search when we have more than one NIC per GPU.
Fix crash with collnetDirect
* Add separate graph search for collnetDirect, testing alltoall paths
and working similarly to the NVLS search.
Fix hang when nodes have different CPU types
* Add the CPU type to the rank peer info.
* Align all ranks on the CPU type after the first allgather.
* Only use the aligned CPU type for all tuning operations.
Fixes issue #1136
Fixes issue #1184
Fix performance of registered send/recv operations
* Allow for single full size operations
* Add INFO to confirm the registration of send/recv buffers.
Move all sync ops to finalize stage
* Ensure ncclCommDestroy is non-blocking if ncclCommFinalize has
been called.
Improve error reporting during SHM segment creation
Improve support of various compilers
Merges PR #1177
Merges PR #1228
Allow net and tuner plugins to be statically linked
* Search for ncclNet or ncclTuner symbols in the main binary.
Merges PR #979
Plugin examples includes cleanup
* Harmonize err.h and common.h usage.
* Add mixed plugin with both net and tuner.
Add support for IB SHARP to NVLS (NVLink SHARP algorithm).
Add NVLS+Tree algorithm.
Add support for memory management using cuMem* functions.
Use all NICs for Send/Receive operations on systems with more than
one NIC per GPU (#804).
Add ncclCommSplit primitive, with resource sharing option in config.
Fix alltoallv hang (#788)
Increase number of channels on H100 when we're not limited by NVLink.
Improve error reporting in case of IB failure, printing local and
remote ID (#779).
Add build option to allow compilation against RDMA includes instead
of dynamically loading IB verbs symbols (#802).
Fix context creation for progress thread (#803).
NET/IB: add option to use multiple QPs in round-robin mode.
Fix tree performance issue when NVB is disabled on HCM topologies.
Optimize CUDA graph launch; avoid launching a CPU callback for
intra-node operations.
Simplify kernel common code to improve the latency of send/recv
operations.
Strengthen CUDA streams semantics.
Change NET API to v6, to add dmabuf support.
Add ncclGetLastError() function.
Add ncclRemoteError code and use it for remote network errors.
Support the use of a different NCCL_NET parameter per communicator.
Add support for SHM and P2P transfers using cudaMemcpy.
Add network communication through another GPU connected with NVLink
(PXN).
Add aggregation of messages coming from different local GPUs through
PXN and going to the same destination.
Add new v5 plugin API with grouped receives and tags.
Add compat for v4 plugins.
Add naming of NCCL threads to help debugging.
Fix NVLink detection and avoid data corruption when some NVLinks are
down.
Add support for Relaxed Ordering for IB.
Add profiling and timing infrastructure.
Add support for CUDA graphs.
Fuse BCM Gen4 switches to avoid suboptimal performance on some platforms. Issue #439.
Fix bootstrap issue caused by connection reordering.
Fix CPU locking block.
Improve CollNet algorithm.
Improve performance on DGX A100 for communicators with only one GPU per node.
Optimization for Tree allreduce on A100.
Improve aggregation performance.
Use shared buffers for inter-node send/recv.
Add NVTX profiling hooks.
Accelerate alltoall connections by merging communication for all
channels.
Add support for one hop communication through NVLink, for faster
send/recv communication on cubemesh topologies like DGX-1.
Improve alltoall scheduling to better balance intra/inter node
communication.
Increase send/recv parallelism by 8x, each warp sending or
receiving to a different peer.
Net: move to v4.
Net: make flush operation asynchronous to accelerate alltoall.
Net: define maximum number of requests.
Fix hang when using LL128 protocol after 2^31 steps.
Fix#379 : topology injection failing when using less GPUs than
described in the XML.
Fix#394 : protocol mismatch causing hangs or crashes when using
one GPU per node.
Add support for network collectives.
Add support for XML topology dump/injection.
Add text values for GDR and P2P Levels, including "NVL".
Add speed detection for PCI, Infiniband and Ethernet cards.
Add CPU detection for ARM and AMD CPUs.
Add support for adaptive routing on Infiniband.
Change NET plugin API to v3 : merge PCI path and GPU pointer
capability into a single structure and add other properties.
Add LL128 Protocol.
Rewrite the topology detection and tree/ring creation (#179). Improve
tree performance by sending/receiving from different GPUs. Add
model-based tuning to switch between the different algorithms and
protocols.
Rework P2P/SHM detection in containers (#155, #248).
Detect duplicated devices and return an error (#231).
Add tuning for GCP
Fix#209: improve socket transport performance
Split transfers over multiple sockets
Launch multiple threads to drive sockets
Detect AWS NICs and set nsockets/nthreads accordingly
Added detection of IBM/Power NVLink bridge device.
Add NUMA support to PCI distance calculations.
Added NCCL_IGNORE_CPU_AFFINITY env var.
Fix memory leaks; GithubIssue#180
Compiler warning fix; GithubIssue#178
Replace non-standard variable length arrays. GithubIssue#171
Fix Tree+Shared Memory crash. GithubPR#185
Fix LL cleanup hang during long running DL jobs.
Fix NCCL_RINGS environment variable handling.
Added extra checks to catch repeat calls to ncclCommDestroy() GithubIssue#191
Improve bootstrap socket connection reliability at scale.
Fix hostname hashing issue. GithubIssue#187
Code cleanup to rename all non device files from *.cu to *.cc
Add tree algorithms for allreduce to improve performance at scale.
Add ncclCommAbort() and ncclCommGetAsyncError() to properly handle
network errors and be permit recover.
Detect initial CPU affinity and no longer escape it.
Dynamically load external network from libnccl-net.so.
Add init function in networks.
Move PCI scoring to net.cu, only ask transport to provide a path.
Simplify CUDA PCI path detection.
Add dummy external network
Add support for inter-node communication using sockets and InfiniBand/RoCE.
Improve latency.
Add support for aggregation.
Improve LL/regular tuning.
Remove tests as those are now at github.com/nvidia/nccl-tests .
Kamil Iskra
Sylvain Jeaugey
Ke Wen
David Addison