Matrix multiplication. Shader v.4_4 benchmark

WebGL2-compute based NxN matrix multiplication C = A x B Wider data-types v.4_4 benchmark. Fixed random A, B 4096x4096 buffers are used for calculations, therefore You can set new N value (up to N=4096, multiple of TS). Tile size (local_size_x,y) TS = 32. it - number of iterations. FLOPS = 2 N3 / time. Random CGPU and CGPU - CCPU are shown.


GeForce GT 710     ~60 GFLOPS (N=4096, TS=32).
Intel Atom Z3735F   1.9 GFLOPS (N=1024, TS=32).

Nvidia GPUs have 256 bits register file. Therefore 8 x 32-bits words data (vec8) have the best performance in OpenCL. Unfortunately in WebGL the script with mat2x4 data is a bit slower than vec4 based one (53 and 60 GFLOPS).

GEMM in WebGL2-compute     updated 26 Mar 2019