MoEntwine: Unleashing the Potential of Wafer-scale Chips for Large-scale Expert Parallel Inference

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Tang, Xinru, Hou, Jingxiang, Jiang, Dingcheng, Wei, Taiquan, Liu, Jiaxin, Deng, Jinyi, Wang, Huizheng, Yang, Qize, Shang, Haoran, Li, Chao, Hu, Yang, Yin, Shouyi
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914121830629376
author Tang, Xinru
Hou, Jingxiang
Jiang, Dingcheng
Wei, Taiquan
Liu, Jiaxin
Deng, Jinyi
Wang, Huizheng
Yang, Qize
Shang, Haoran
Li, Chao
Hu, Yang
Yin, Shouyi
author_facet Tang, Xinru
Hou, Jingxiang
Jiang, Dingcheng
Wei, Taiquan
Liu, Jiaxin
Deng, Jinyi
Wang, Huizheng
Yang, Qize
Shang, Haoran
Li, Chao
Hu, Yang
Yin, Shouyi
contents As large language models (LLMs) continue to scale up, mixture-of-experts (MoE) has become a common technology in SOTA models. MoE models rely on expert parallelism (EP) to alleviate memory bottleneck, which introduces all-to-all communication to dispatch and combine tokens across devices. However, in widely-adopted GPU clusters, high-overhead cross-node communication makes all-to-all expensive, hindering the adoption of EP. Recently, wafer-scale chips (WSCs) have emerged as a platform integrating numerous devices on a wafer-sized interposer. WSCs provide a unified high-performance network connecting all devices, presenting a promising potential for hosting MoE models. Yet, their network is restricted to a mesh topology, causing imbalanced communication pressure and performance loss. Moreover, the lack of on-wafer disk leads to high-overhead expert migration on the critical path. To fully unleash this potential, we first propose Entwined Ring Mapping (ER-Mapping), which co-designs the mapping of attention and MoE layers to balance communication pressure and achieve better performance. We find that under ER-Mapping, the distribution of cold and hot links in the attention and MoE layers is complementary. Therefore, to hide the migration overhead, we propose the Non-invasive Balancer (NI-Balancer), which splits a complete expert migration into multiple steps and alternately utilizes the cold links of both layers. Evaluation shows ER-Mapping achieves communication reduction up to 62%. NI-Balancer further delivers 54% and 22% improvements in MoE computation and communication, respectively. Compared with the SOTA NVL72 supernode, the WSC platform delivers an average 39% higher per-device MoE performance owing to its scalability to larger EP.
format Preprint
id arxiv_https___arxiv_org_abs_2510_25258
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MoEntwine: Unleashing the Potential of Wafer-scale Chips for Large-scale Expert Parallel Inference
Tang, Xinru
Hou, Jingxiang
Jiang, Dingcheng
Wei, Taiquan
Liu, Jiaxin
Deng, Jinyi
Wang, Huizheng
Yang, Qize
Shang, Haoran
Li, Chao
Hu, Yang
Yin, Shouyi
Distributed, Parallel, and Cluster Computing
As large language models (LLMs) continue to scale up, mixture-of-experts (MoE) has become a common technology in SOTA models. MoE models rely on expert parallelism (EP) to alleviate memory bottleneck, which introduces all-to-all communication to dispatch and combine tokens across devices. However, in widely-adopted GPU clusters, high-overhead cross-node communication makes all-to-all expensive, hindering the adoption of EP. Recently, wafer-scale chips (WSCs) have emerged as a platform integrating numerous devices on a wafer-sized interposer. WSCs provide a unified high-performance network connecting all devices, presenting a promising potential for hosting MoE models. Yet, their network is restricted to a mesh topology, causing imbalanced communication pressure and performance loss. Moreover, the lack of on-wafer disk leads to high-overhead expert migration on the critical path. To fully unleash this potential, we first propose Entwined Ring Mapping (ER-Mapping), which co-designs the mapping of attention and MoE layers to balance communication pressure and achieve better performance. We find that under ER-Mapping, the distribution of cold and hot links in the attention and MoE layers is complementary. Therefore, to hide the migration overhead, we propose the Non-invasive Balancer (NI-Balancer), which splits a complete expert migration into multiple steps and alternately utilizes the cold links of both layers. Evaluation shows ER-Mapping achieves communication reduction up to 62%. NI-Balancer further delivers 54% and 22% improvements in MoE computation and communication, respectively. Compared with the SOTA NVL72 supernode, the WSC platform delivers an average 39% higher per-device MoE performance owing to its scalability to larger EP.
title MoEntwine: Unleashing the Potential of Wafer-scale Chips for Large-scale Expert Parallel Inference
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2510.25258