Skip to content

MaxFlops

Jump to bottom
spaffy edited this page Feb 29, 2012 · 1 revision

MaxFlops

Description: Measures the maximum floating-point throughput for different combinations of floating point operations. Results are reported in GFLOPS (GFlop/s).

Problem Sizes: Problem size scales with device performance.

Precision: single precision and double precision.

  • benchmark name suffix, '-SP' or '-DP' indicates precision.

Includes PCIe Transfer Time: No

Specific Tests

  • Add1 - Reports floating point throughput using a stream of floating point additions

  • Add2 - Reports floating point throughput using two independent streams of floating point additions

  • Add4 - Reports floating point throughput using four independent streams of floating point additions

  • Add8 - Reports floating point throughput using eight independent streams of floating point additions

  • MAdd1 - Reports floating point throughput using a stream of floating point multiply-adds

  • MAdd2 - Reports floating point throughput using two independent streams of floating point multiply-adds

  • MAdd4 - Reports floating point throughput using four independent streams of floating point multiply-adds

  • MAdd8 - Reports floating point throughput using eight independent streams of floating point multiply-adds

  • MAddU - Reports floating point throughput for a hand-tuned kernel computing many floating point multiply-adds

  • Mul1 - Reports floating point throughput using a stream of floating point multiplications

  • Mul2 - Reports floating point throughput using two independent streams of floating point multiplications

  • Mul4 - Reports floating point throughput using four independent streams of floating point multiplications

  • Mul8 - Reports floating point throughput using eight independent streams of floating point multiplications

  • MulMadd1 - Reports floating point throughput using a stream of floating point multiplications and multiply-adds

  • MulMadd2 - Reports floating point throughput using two independent streams of floating point multiplications and multiply-adds

  • MulMadd4 - Reports floating point throughput using four independent streams of floating point multiplications and multiply-adds

  • MulMadd8 - Reports floating point throughput using eight independent streams of floating point multiplications and multiply-adds

  • MulMAddU - Reports floating point throughput for a hand-tuned kernel using a mix of floating point multiplications and multiply-adds

Clone this wiki locally