Skip to content
spaffy edited this page Feb 29, 2012 · 1 revision

GEMM

Description: Measures performance for device versions of the GEMM BLAS routine (general matrix multiply), on a square matrix

Options

  • --KiB [N]: Specify the problem size is KB, overrides normal -s sizes

Problem Sizes: Matrix Size (KB) - 1, 4, 8, 16

Precision: Single

Includes PCIe Transfer Time: No

Specific Tests

  • CUBLAS-N - Reports the performance of the GEMM kernel without transpose in GFLOPS/s
  • CUBLAS-T - Reports the performance of the GEMM kernel with transposing input in GFLOPS/s
Clone this wiki locally