Matrix multiplication AMD A6-5200 APU benchmark


Shader 7
TFjs
RGBA32F textures
N = 128
1.9 ms
3 ms
2.2 ms
N=512
48 (5.6 ms)
18.2 (15 ms)
14.2 (19 ms)
N=1024
79
25.2
14.8
N=2048
89
26.8
14.8
N=4096
89 GFLOPS
19.6 GFLOPS
-
For N = 128 overheads time is measured.

Tested on Windows 10, 64-bit, OpenGL backend.
AMD A6-5200 APU with Radeon HD8400
128 GPU cores (600MHz), memory 512MB DDR3 (800MHz) bandwidth ~5.2 GB/s (unfortunately one SODIMM slot is used).
SiSoftware Sandra OpenCL FP32 GPU test
Peak GPU
153.6 GFLOPS
GEMM
35.2 GFLOPS
FFT
6.2 GFLOPS
NBody
86.2 GFLOPS
CPU GEMM
30 GFLOPS


SGEMM in WebGL2-compute     updated 26 May 2019