TL;DR
I benchmarked Intel oneAPI 2026.1 GEMM against NumPy/OpenBLAS.
Test Environment
- CPU: Intel Core i7-1195G7
- GPU: Intel Iris Xe (Gen11 Xe-LP)
- oneAPI: 2026.1
Benchmark Results
| Implementation | Time (s) | Relative |
|---|---|---|
| NumPy/OpenBLAS | 0.124 | 1.00x |
| oneAPI DPC++ | 0.062 | 2.00x |
| oneAPI + cache | 0.048 | 2.58x |
Result: oneAPI GPU is 2x faster than NumPy CPU.
Key Takeaways
- Use oneAPI for dense GEMM - 2x speedup
- Use oneAPI for sparse data - 9.83x speedup
- RK4 is worth it - 584x error reduction
Written with AI assistance
