About Crosssim trainning

[neryang]

Why am I so slow to train the mnist data set with training functions, even though I've trained it with Gpu

[ptxiao]

What memory device settings are you training with? If you're training with device look-up tables and including device-to-device variability, then it can get quite slow because it is randomly sampling from a different conductance update lookup table for every device in the entire system. This is inherently quite complicated to model accurately and those operations are not easily accelerated with a GPU (though there should still be some speed-up).