zhang954527 / cpugpu_speedtest

Speed test benchmark for CPU (multi threads or processes) or GPU.

CPU/GPU Perforance Benchmarks

Times CPU/GPU performance according some computation benchmarks.

Basic Usage

Build:

make

Run:

./mm
./mm -n 200
./mm -n 2000 -double
./mm_omp -n 2000 -ntomp 4
mpirun -np <procs> ./mm

Usage Syntax

Usage:  mm / mm_omp [options] [operations]

Options:
  -n <size>      Square matrix size n*n (default 1000)
  -ntomp <nomp>  OpenMP threads number  (default 1)
  -float         Use single precision (default)
  -double        Use double precision

Bench Record

Test1 - 1 nodes, 1 procs, 1 threads (Intel(R) Xeon(R) Platinum 8336C CPU @ 2.30GHz)

./mm
Matrix size: 1000 Precision: float
Matrix multiplication time: 1.33137 second

./mm -double
Matrix size: 1000 Precision: double
Matrix multiplication time: 1.44273 seconds

./mm -n 500
Matrix size: 500 Precision: float
Matrix multiplication time: 0.157968 seconds

./mm -n 2000
Matrix size: 2000 Precision: float
Number of threads: 1
Matrix multiplication time: 48.4952 seconds

Test2 - 1 nodes, 1 procs, n threads (Intel(R) Xeon(R) Platinum 8336C CPU @ 2.30GHz)

The speed basically increases linearly with the number of threads.

./mm_omp -n 2000 -ntomp 2
Matrix size: 2000 Precision: float
Number of threads: 2
Matrix multiplication time: 22.976 seconds

./mm_omp -n 2000 -ntomp 2 -double
Matrix size: 2000 Precision: double
Number of threads: 2
Matrix multiplication time: 26.6082 seconds

./mm_omp -n 2000 -ntomp 4
Matrix size: 2000 Precision: float
Number of threads: 4
Matrix multiplication time: 11.5409 seconds

./mm_omp -n 2000 -ntomp 8
Matrix size: 2000 Precision: float
Number of threads: 8
Matrix multiplication time: 5.99805 seconds