AI & ML Paradigm Shift

Reframes GPU kernel optimization by benchmarking against hardware 'Speed-of-Light' limits rather than software baselines.

March 20, 2026

Original Paper

SOL-ExecBench: Speed-of-Light Benchmarking for Real-World GPU Kernels Against Hardware Limits

Edward Lin, Sahil Modi, Siva Kumar Sastry Hari, Qijing Huang, Zhifan Ye, Nestor Qin, Fengzhe Zhou, Yuan Zhang, Jingquan Wang, Sana Damani, Dheeraj Peri, Ouye Xie, Aditya Kane, Moshe Maor, Michael Behar, Triston Cao, Rishabh Mehta, Vartika Singh, Vikram Sharma Mailthody, Terry Chen, Zihao Ye, Hanfeng Chen, Tianqi Chen, Vinod Grover, Wei Chen, Wei Liu, Eric Chung, Luis Ceze, Roger Bringmann, Cyril Zeller, Michael Lightstone, Christos Kozyrakis, Humphrey Shi

arXiv · 2603.19173

The Takeaway

As AI agents begin to write CUDA kernels, measuring speedup over existing code is a moving target; this benchmark provides absolute hardware-grounded goals. It includes a sandboxed harness and clock-locking to prevent reward-hacking, essential for training next-generation LLM-based kernel optimizers.

From the abstract

As agentic AI systems become increasingly capable of generating and optimizing GPU kernels, progress is constrained by benchmarks that reward speedup over software baselines rather than proximity to hardware-efficient execution. We present SOL-ExecBench, a benchmark of 235 CUDA kernel optimization problems extracted from 124 production and emerging AI models spanning language, diffusion, vision, audio, video, and hybrid architectures, targeting NVIDIA Blackwell GPUs. The benchmark covers forward

Read the original paper →

← Back to today's papers