@karlhigley I benchmarked matrix multiplication with tf and interestingly, the 8 cores are also pretty fast so that the GPU was only about 10 times faster than the 8 cores (but about 100x than a single core).