Jacket GBENCH allows users to gauge the GPU performance of their computer relative to equivalent benchmarks obtained from a variety of other computers, including the CPU of the same computer. Benchmarks include six different tasks, common to the technical computing community:
1. LU: LU decomposition of 1024 x 1024 matrix
2. FFT: Fast Fourier Transform of a 2^20 x 1 vector
3. BLAS: Matrix multiplication of two 1024x1024 matrices
4. 3D Conv: Convolution of 64x64x64 array with 3x3x3 kernel
5. FOR/GFOR: Matrix-vector multiplication of 1024x1024x32 array
6. Equations: Solution of a system of 1024 equations
GBENCH is a practical application benchmark measured in real seconds and is not meant to be a scientific or theoretical benchmark measured in GFLOPs. Also note that for fairness, arithmetic precisions (e.g. double, single) have been matched on the CPU and GPU. Finally, the data sizes used in these computations are large enough to exploit data parallelism (e.g. no scalar arithmetic was attempted). This benchmark assumes a data parallel problem.
Note: GBENCH v1.0 was built using Jacket v1.2.1 and MATLAB R2009b. GBENCH requires
CUDA 2.3 (driver and toolkit).
Note:
will not work with older CUDA versions!
Here is my result:
