Microsoft and Tsinghua find most GPU kernel library failures stem from environment, not kernel code
Overview
Microsoft and Tsinghua University analyzed 300 closed GitHub issues from seven GPU kernel libraries, including FlashAttention, Triton, and CUTLASS. They report that 60.33% of failures traced to the compiler, build system, or runtime environment, and 39.67% to the kernel library itself.
Software incompatibility among drivers, CUDA, frameworks, and packages was the largest single cause, at 76 issues or 25.33% of the sample. The findings were reported by Rohan Paul on X, citing the paper.
Written by AI from the articles below · updated Nov 11, 12:37 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
Rohan Paul@rohanpaul_aiXMicrosoft and Tsinghua study finds most GPU kernel library failures stem from compilers and environmentsAIMicrosoft and Tsinghua University analyzed 300 closed GitHub issues from 7 GPU kernel libraries, including FlashAttention, Triton, and CUTLASS. They found 60.33% of failures traced to the compiler, build system, or runtime environment, and 39.67% to the kernel library itself. Software incompatibility among drivers, CUDA, frameworks, and packages was the largest single cause, at 76 issues or 25.33%.

Heat Trend
- Current comparable heat
- 9
- Peak comparable heat
- 10Nov 11, 1:00 AM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.