I typically run 8 concurrent pps sieve threads on my gtx 680 in a mac pro 3,1. I was getting run times of approx. 4200 seconds each. After upgrading from 10.10.4 to 10.10.5, these run times have gone to approx. 5200 seconds each. Switching to/from nvidia vs. mac drivers doesn't help, and upgrading cuda didn't help.
Anyone else notice this? Its a pretty big hit.
Box is a dedicated cruncher, so its not like anything else is/should be happening on it.