Eidola: Modeling Multi-GPU Network Communication Traffic in Distributed AI Workloads

2026-06-10

AI Summary

Existing tools struggle to accurately model the irregular and transient multi-GPU communication traffic in distributed AI training caused by kernel fusion and communication-computation overlap. This work proposes Eidolon, a lightweight GPU proxy model built upon the gem5 simulation framework, which leverages annotated timing profiles from real applications to simulate inter-GPU point-to-point write operations with cycle-level accuracy. Eidolon supports configurable synchronization mechanisms, including heuristics such as SyncMon. The approach enables high-fidelity and scalable modeling of multi-GPU communication, successfully reproducing the execution variability inherent in fused kernels. It further demonstrates that synchronization optimizations can substantially reduce polling-related memory traffic, offering an effective simulation platform for architectural exploration of large-scale distributed GPU systems.

Abstract

As distributed AI workloads grow in scale, multi-GPU systems have become essential for training large models. Although techniques like kernel fusion and overlapping communication with computation help reduce delays, they also introduce irregular and transient traffic patterns that are difficult to model using existing tools. These techniques rely heavily on fine-grained synchronization and peer-to-peer communication, which place significant pressure on interconnect bandwidth and latency. In this work, we introduce Eidola, a scalable extension to the gem5 simulation framework that enables detailed modeling of inter-GPU communication traffic. The extension is scalable as our GPU model serves as a succinct eidolon, emulating the minimal characteristics needed for traffic modeling. Eidola uses annotated timing profiles from real applications to emulate peer-to-peer GPU writes with cycle-level precision. This allows researchers to simulate and analyze synchronization behavior across large multi-GPU configurations. The simulator supports configurable per-GPU traffic patterns and enables isolated performance analysis under different communication scenarios. We demonstrate Eidola's effectiveness by reproducing variability in fused kernel execution and by implementing a SyncMon-inspired synchronization mechanism, confirming reductions in polling-related memory traffic. Our results show that Eidola provides a flexible and scalable platform for studying inter-GPU communication and supports architectural exploration in modern distributed GPU systems.

Problem

• multi-GPU communication
• traffic modeling
• distributed AI workloads
• interconnect bandwidth
• synchronization

Innovation

• multi-GPU communication
• cycle-accurate simulation
• distributed AI workloads
• traffic modeling
• gem5 extension