NCCL EP: Towards a Unified Expert Parallel Communication API for NCCL

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Goldman, Amos, Boker, Nimrod, Sheraizin, Maayan, Admoni, Nimrod, Polyakov, Artem, Bhattacharya, Subhadeep, Yu, Fan, Sun, Kai, Theodorakis, Georgios, Yin, Hsin-Chun, Gootzen, Peter-Jan, Shafi, Aamir, Ravid, Assaf, Di Girolamo, Salvatore, Dinan, James, Li, Xiaofan, Venkata, Manjunath Gorentla, Bloch, Gil
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908932446879744
author Goldman, Amos
Boker, Nimrod
Sheraizin, Maayan
Admoni, Nimrod
Polyakov, Artem
Bhattacharya, Subhadeep
Yu, Fan
Sun, Kai
Theodorakis, Georgios
Yin, Hsin-Chun
Gootzen, Peter-Jan
Shafi, Aamir
Ravid, Assaf
Di Girolamo, Salvatore
Dinan, James
Li, Xiaofan
Venkata, Manjunath Gorentla
Bloch, Gil
author_facet Goldman, Amos
Boker, Nimrod
Sheraizin, Maayan
Admoni, Nimrod
Polyakov, Artem
Bhattacharya, Subhadeep
Yu, Fan
Sun, Kai
Theodorakis, Georgios
Yin, Hsin-Chun
Gootzen, Peter-Jan
Shafi, Aamir
Ravid, Assaf
Di Girolamo, Salvatore
Dinan, James
Li, Xiaofan
Venkata, Manjunath Gorentla
Bloch, Gil
contents Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, driving the development of specialized device-initiated communication libraries such as DeepEP, Hybrid-EP, and others. These libraries demonstrate the performance benefits of GPU-initiated RDMA for MoE dispatch and combine operations. This paper presents NCCL EP (Expert Parallelism), a ground-up MoE communication library built entirely on NCCL's Device API. NCCL EP provides unified ncclEpDispatch and ncclEpCombine primitives with both C and Python interfaces, supporting Low-Latency (LL) mode for inference decoding and High-Throughput (HT) mode for training and inference prefill. LL targets small batch sizes (1-128 tokens) using direct all-to-all RDMA+NVLink mesh connectivity with double-buffered communication for overlapping dispatch and combine phases. HT targets large batches (4096+ tokens) using hierarchical communication that aggregates tokens within NVLink domains before inter-node RDMA transmission. Both modes leverage Device API for both intra- and inter-node communications, taking advantage of its topology awareness and optimized GPU-initiated implementation. We evaluate NCCL EP on an H100-based cluster across multi-node configurations, demonstrating competitive LL kernel performance and presenting end-to-end results with vLLM integration. By building MoE communication natively within NCCL, NCCL EP provides a supported path for expert parallelism on current and emerging NVIDIA platforms.
format Preprint
id arxiv_https___arxiv_org_abs_2603_13606
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle NCCL EP: Towards a Unified Expert Parallel Communication API for NCCL
Goldman, Amos
Boker, Nimrod
Sheraizin, Maayan
Admoni, Nimrod
Polyakov, Artem
Bhattacharya, Subhadeep
Yu, Fan
Sun, Kai
Theodorakis, Georgios
Yin, Hsin-Chun
Gootzen, Peter-Jan
Shafi, Aamir
Ravid, Assaf
Di Girolamo, Salvatore
Dinan, James
Li, Xiaofan
Venkata, Manjunath Gorentla
Bloch, Gil
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Hardware Architecture
Machine Learning
68W10, 68M10, 65Y05
C.2.1; C.2.4; C.1.2; C.1.4
Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, driving the development of specialized device-initiated communication libraries such as DeepEP, Hybrid-EP, and others. These libraries demonstrate the performance benefits of GPU-initiated RDMA for MoE dispatch and combine operations. This paper presents NCCL EP (Expert Parallelism), a ground-up MoE communication library built entirely on NCCL's Device API. NCCL EP provides unified ncclEpDispatch and ncclEpCombine primitives with both C and Python interfaces, supporting Low-Latency (LL) mode for inference decoding and High-Throughput (HT) mode for training and inference prefill. LL targets small batch sizes (1-128 tokens) using direct all-to-all RDMA+NVLink mesh connectivity with double-buffered communication for overlapping dispatch and combine phases. HT targets large batches (4096+ tokens) using hierarchical communication that aggregates tokens within NVLink domains before inter-node RDMA transmission. Both modes leverage Device API for both intra- and inter-node communications, taking advantage of its topology awareness and optimized GPU-initiated implementation. We evaluate NCCL EP on an H100-based cluster across multi-node configurations, demonstrating competitive LL kernel performance and presenting end-to-end results with vLLM integration. By building MoE communication natively within NCCL, NCCL EP provides a supported path for expert parallelism on current and emerging NVIDIA platforms.
title NCCL EP: Towards a Unified Expert Parallel Communication API for NCCL
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Hardware Architecture
Machine Learning
68W10, 68M10, 65Y05
C.2.1; C.2.4; C.1.2; C.1.4
url https://arxiv.org/abs/2603.13606