AI-native GPU kernel language paper published, claims performance exceeds vendor libraries
Understand this story
Researcher Heng Shi and other authors published a paper proposing an AI-native GPU kernel programming language co-designed with AI tuning, with the goal of enabling AI to autonomously complete kernel tuning. The paper claims that, on an H800, AI-tuned kernels for 95 production shapes are on average 1.23 times faster than cuBLAS, CUTLASS, and cuSPARSELt. The authors claim this is the first AI-native GPU kernel programming language, and that in comparisons with six DSLs it is the only approach that consistently outperforms vendor libraries. The above performance data come from the paper's own claims and have not yet been independently reproduced.
AI overview based on source reports · Updated 13 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIA Novel AI-Native GPU Kernel Programming Language for Autonomous Kernel Tuning
This SIGOPS ATC 2026 paper presents an AI-native GPU kernel programming language co-designed with AI tuning, arguing that existing LLM-driven tuners are limited because they only search parameters within human-supplied skeletons and rely on substrates with large token footprints and no compile-time feedback. In a six-DSL comparison using the same agent, H800 GPU, and iteration budget, its AI-tuned kernels averaged 1.23× over cuBLAS, CUTLASS, and cuSPARSELt across 95 production shapes, with trajectory analysis attributing gains to structural discoveries such as warp specialisation, pipeline staging, and TMA. The authors report compile pass@1 of 85.7% (95.6% pass@5), 82.1% pass@1 on dynamic-shape kernels versus Triton's 48%, and 82.3% under moderate-capacity models versus Triton's 65.8%.
Activity for this story
- Current comparable activity
- 7
- Peak comparable activity
- 10Nov 20 07:00
- Change in the past 24 hours
- –
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.