Multithreading in matrix factorisation gives no performance boost

[evdoks]

I've notices that matrix factorisation is using multithreading with threads that are called one after each other. I also did some time measurements and seems that the number of threads (1 or multiple) has no impact on the performance. Am I missing something or is the multithreading just an old code that remained due to the attempt to parallelise mini-batch SGD?

[AbeyHurtis]

I do not understand fully what is your issue but from as far as i understood it ,
If the threads are executed sequentially(one after the other) then it will be the same as executing it without using threads ,you should try to initialize the matrices in different cores of (specific gpu)