-
Notifications
You must be signed in to change notification settings - Fork 102
FFT
spaffy edited this page Feb 29, 2012
·
1 revision
Description: Measures performance of a single and double precision fast fourier transform (FFT). For optimal GPU performance, the size of the transform computed is fixed at 512 complex elements; varying the problem size thus changes the number of 512-element transforms computed in parallel.
Problem Sizes: (MB) 1, 8, 96, 256
Options
- --MB [N] - Specify a specific problem size of N MB, overrides normal -s sizes
Precision: Both
Includes PCIe Transfer Time: Yes (in *_PCIe results).
Specific Tests
- SP-FFT - Reports performance of single precision forward transform in GFLOPS/s
- SP-FFT_PCIe - Includes PCIe Transfer time in GFLOPS/s calculation.
- SP-FFT-INV - Reports performance of single precision inverse transform and normalization step in GFLOPs/s
- SP-FFT-INV_PCIe - Includes PCIe Transfer time in GFLOPS/s calculation.
- DP-FFT - Reports performance of double precision forward transform in GFLOPS/s
- DP-FFT_PCIe - Includes PCIe Transfer time in GFLOPS/s calculation.
- DP-FFT-INV - Reports performance of double precision inverse transform and normalization step in GFLOPs/s
- DP-FFT-INV_PCIe - Includes PCIe Transfer time in GFLOPS/s calculation.