Benchmark RadarAI BENCHMARK PROFILE
DataKernelBench
Evaluates LLM-generated CUDA or Triton kernels for database query operations on GPUs, measuring execution speedup over torch.compile on TPC-H workloads.
- Released
- 2026-08-25
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Addresses the lack of benchmarks for database-style operators in LLM kernel generation, enabling comparison of models on real query optimization tasks.
Motivation
GPUs increasingly accelerate database systems, but query-specific peak performance still often relies on hand-written kernels.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.