Daisen: New Framework for Visualizing GPU Execution Traces to Improve Hardware Performance Analysis

Summary

Researchers have introduced Daisen, an open-source framework designed to help GPU hardware designers visualize and analyze detailed GPU execution traces generated by simulators. The framework addresses a critical challenge in GPU hardware design: making sense of massive volumes of simulator data to identify performance bottlenecks and optimization opportunities. Daisen combines a specialized data abstraction format for capturing GPU execution traces with a web-based visualization tool that helps designers examine hardware behavior, detect performance issues, and verify improvements. The research, conducted through interviews and surveys with GPU hardware designers, demonstrates that Daisen effectively supports the typical workflows of GPU optimization teams. Qualitative evaluation showed that users could successfully identify potential performance bottlenecks and opportunities for hardware improvement using the visualization interface.

• Qualitative evaluation with GPU designers confirms the framework aligns with real-world hardware optimization processes

Editorial Opinion

Daisen represents a thoughtful engineering solution to a genuine pain point in GPU hardware design—turning unwieldy simulator data into actionable insights through smart visualization. By open-sourcing the framework, the researchers enable the broader GPU community to benefit from improved analysis tools, potentially accelerating GPU optimization cycles across the industry. However, the framework's impact will ultimately depend on adoption by major GPU manufacturers and its extensibility to emerging GPU architectures.