[gpderetta]

Yes, I use [3] and I get 2.4s/iter on my 10 core machine. I was wondering if keras would give additional help here. I'll have to try I guess.