NCCL
E209960
GPU communication library
collective communications library
communication library
parallel computing library
NCCL (NVIDIA Collective Communications Library) is a high-performance library that optimizes multi-GPU and multi-node communication for deep learning and HPC applications.
All labels observed (4)
| Label | Occurrences |
|---|---|
| NCCL canonical | 6 |
| NVIDIA NCCL | 4 |
| NVIDIA Collective Communications Library | 1 |
| NVIDIA Communication Libraries | 1 |
How this entity was disambiguated
This entity first appeared as the object of triple T1893382 — resolving that mention is where its identity was fixed. The disambiguator weighed these candidate entities and picked the highlighted one (or “None”, minting a new entity). This is how homonymy is resolved: the same surface form can point to different entities.
NED1
Entity disambiguation (via context triple)
gpt-5-mini-2025-08-07
Target entity: NCCL Context triple: [NVIDIA CUDA, includes, NCCL]
-
A.
NVIDIA CUDA
NVIDIA CUDA is a parallel computing platform and programming model that enables developers to use NVIDIA GPUs for general-purpose high-performance computing.
-
B.
NVIDIA AI Enterprise software suite
NVIDIA AI Enterprise software suite is a comprehensive, enterprise-grade collection of AI tools, frameworks, and optimized software designed to accelerate the development and deployment of AI and data analytics workloads across modern data centers and clouds.
-
C.
GPU
The GPU (State Political Directorate) was the Soviet Union’s early secret police and intelligence agency that operated in the 1920s, overseeing political repression and internal security before later reorganizations.
-
D.
GPU
GPU is the vehicle registration code used on license plates for cars registered in Poland’s Pomeranian Voivodeship.
-
E.
PlaidML
PlaidML is an open-source, hardware-agnostic deep learning engine designed to accelerate neural network computation on a wide range of GPUs and other devices.
- F. None of above. chosen
- G. Unsure - the case is ambiguous/there is not enough information to decide.
NED2
Entity disambiguation (via description)
gpt-5-mini-2025-08-07
Target entity: NCCL Target entity description: NCCL (NVIDIA Collective Communications Library) is a high-performance library that optimizes multi-GPU and multi-node communication for deep learning and HPC applications.
-
A.
NVIDIA CUDA
NVIDIA CUDA is a parallel computing platform and programming model that enables developers to use NVIDIA GPUs for general-purpose high-performance computing.
-
B.
NVIDIA AI Enterprise software suite
NVIDIA AI Enterprise software suite is a comprehensive, enterprise-grade collection of AI tools, frameworks, and optimized software designed to accelerate the development and deployment of AI and data analytics workloads across modern data centers and clouds.
-
C.
GPU
The GPU (State Political Directorate) was the Soviet Union’s early secret police and intelligence agency that operated in the 1920s, overseeing political repression and internal security before later reorganizations.
-
D.
GPU
GPU is the vehicle registration code used on license plates for cars registered in Poland’s Pomeranian Voivodeship.
-
E.
PlaidML
PlaidML is an open-source, hardware-agnostic deep learning engine designed to accelerate neural network computation on a wide range of GPUs and other devices.
- F. None of above. chosen
Statements (66)
| Predicate | Object |
|---|---|
| instanceOf |
GPU communication library
ⓘ
collective communications library ⓘ communication library ⓘ parallel computing library ⓘ |
| developer |
NVIDIA
ⓘ
linked to:
NVIDIA Corporation
|
| domain |
deep learning
ⓘ
distributed training ⓘ high-performance computing ⓘ multi-GPU computing ⓘ multi-node computing ⓘ |
| feature |
GPUDirect RDMA support
ⓘ
asynchronous communication API ⓘ grouped collective operations ⓘ hierarchical collectives ⓘ multi-process communication support ⓘ ring-based collectives ⓘ topology-aware communication algorithms ⓘ tree-based collectives ⓘ |
| fullName |
NVIDIA Collective Communications Library
ⓘ
linked to:
NCCL
|
| goal |
high-bandwidth GPU communication
ⓘ
low-latency collective operations ⓘ scalable multi-GPU training ⓘ |
| integratesWith |
CUDA
ⓘ
linked to:
NVIDIA CUDA
|
| license | BSD-style license ⓘ |
| optimizedFor |
GPU interconnects
ⓘ
NVIDIA GPUs ⓘ
linked to:
Nvidia Maxwell GPU
|
| primaryAPIStyle | C API ⓘ |
| programmingLanguage | C ⓘ |
| provider |
NVIDIA
ⓘ
linked to:
NVIDIA Corporation
|
| requires | CUDA-enabled GPU ⓘ |
| supportsDataType |
bfloat16
ⓘ
float16 ⓘ float32 ⓘ float64 ⓘ int32 ⓘ int64 ⓘ uint8 ⓘ |
| supportsEnvironment |
cloud GPU clusters
ⓘ
on-premise clusters ⓘ |
| supportsHardware |
Ethernet
ⓘ
InfiniBand ⓘ NVLink ⓘ
linked to:
NVLink interconnect
NVSwitch ⓘ
linked to:
NVLink interconnect
PCIe ⓘ
linked to:
PCI Express
|
| supportsLanguageBinding |
C++
ⓘ
Python ⓘ |
| supportsOperation |
all-gather
ⓘ
all-reduce ⓘ broadcast ⓘ gather ⓘ reduce ⓘ reduce-scatter ⓘ scatter ⓘ |
| supportsReductionOp |
max
ⓘ
min ⓘ product ⓘ sum ⓘ |
| supportsTopology |
multi-node multi-GPU
ⓘ
single-node multi-GPU ⓘ |
| targetUser |
HPC developers
ⓘ
deep learning framework developers ⓘ |
| usedBy |
DeepSpeed
ⓘ
Horovod ⓘ Megatron-LM ⓘ PyTorch distributed ⓘ
linked to:
PyTorch
TensorFlow distributed training ⓘ |
How these facts were elicited
The pipeline generated the facts above by prompting gpt-5.1 with this entity's name + description and the instruction below.
Instruction
You are a knowledge base construction expert. Given a subject entity and a description of it, return factual statements that you know for the subject as a JSON list of dictionaries(triples), where keys must be "subject", "predicate" and "object". The number of facts may be very high, between 25 to 50 or more, for very popular subjects. For less popular subjects, the number of facts can be very low, like 5 or 10. # Requirements - If you don't know the subject at all, return an empty list. - If the subject is not a named entity, return an empty list. - Include at least one triple where predicate is "instanceOf". - Do not get too wordy. - Separate several objects into multiple triples with one object.
Input
Subject: NCCL Description of subject: NCCL (NVIDIA Collective Communications Library) is a high-performance library that optimizes multi-GPU and multi-node communication for deep learning and HPC applications.
Referenced by (12)
Full triples — surface form annotated when it differs from this entity's canonical label.
linked to: NCCL
subject linked to:
Compute Unified Device Architecture
linked to: NCCL
linked to: NCCL
linked to: NCCL
linked to: NCCL