Measuring the speed of the GPU model operation

[kelam_goutam]

I guess this link will provide you relative answer

Measuring GPU tensor operation speed

Hi, I would like to illustrate the speed of tensor operations on GPU for a course. The following piece of code: x = torch.cuda.FloatTensor(10000, 500).normal_() w = torch.cuda.FloatTensor(200, 500).normal_() a = time.time() y = x.mm(w.t()) b = time.time() print('batch GPU {:.02e}s'.format(b - a)) a = time.time() y = x.mm(w.t()) b = time.time() print('batch GPU {:.02e}s'.format(b - a)) prints batch GPU 1.06e-01s batch GPU 3.43e-04s so I presume that there is some “lazy operations” delayed…