TensorFlow.js (WebGL) based NxN matrix multiplication C = A x B
benchmark. Random A, B are generated for calculations.
FLOPS = 2 N3 / Tmin.
You can set new N value (note that execution time ~N3).
The first run initializes A,B and is the slowest.
Test results are strange
(note, that FP16 is 2 times faster than FP32 in OpenCL CLBlast tests).
As you can see in console, in tfjs-core fragment shaders "highp" precision
for floats is set therefore drivers use fp32 math.
In my fast HGEMM with RGBA16F textures script
"mediump" precision for floats are used!
I don't see any acceleration on RTX 2070 (~0.8 TFLOPS) but
HGEMM with RGBA16F textures is 2 times faster
(~1.6 TFLOPS, D3D11 and OpenGL backends).
Intel Atom z3735f (Chrome, WebGL2, D3D11, Win10 32 bit) ~10 GFLOPS
was ~5 GFLOPS with float textures.
But there is not any acceleration on Intel I5 + Win10 64 bit.
Intel team is working on D3D11 drivers.
Vega AMD GPU supports FP16 math but I don't see any acceleration
(it may be drivers feature, if e.g. FP32 are used for HALF_FLOAT).