New Version 0.6.1
...
- improve performance on linux systems by ~2%
You apply 2 queue per GPU, right? Make please option for switch between old/new modes for selected card. For example - like "dev=0,1,2" for old mode and "dev=0,0,1,2,2" for new mode (card 0 and 2 - 2 thread/GPU, card 1 - 1 thread/GPU)
Also, with v6.1:
gtx750 - SM5.0 - same performance, +1..2 sols/sec
gtx950 - SM5.2 - yes, +2% faster
gtx1050ti - SM6.1 - 1-2% SLOWER
Rollback to v6.0

Thx for reporting performance measurements.
+1-2 Sol/s on an gtx750 are about +2%.
gtx1050ti / sm6.1: There is nothing special about sm6.1 in respect to this optimization. That's not what I'm getting.
Could you pls provide the log files of your tests?
A run of 5min (for 0.6/0.6.1) should be enough on a previously cooled down system.