In a quick test the extraction of points is much better in version 1.3, What I do not understand is because the processing speed values are so different between versions.
v1.2.6
Device 1 performance: 141.395 million samples/sec (CPU)
Device 2 performance: 423.937 million samples/sec (GeForce GTX TITAN X)
Total performance: 565.332 million samples/sec
Finished processing in 60.611 sec
v1.3.0
Device 1: 17.0262 million samples/sec (CPU)
Device 2: 274.403 million samples/sec (GeForce GTX TITAN X)
Total performance: 291.43 million samples/sec
Finished processing in 42.773 sec
v1.2.6 = 565.332 million samples/sec vs v1.3.0 = 291.43 million samples/sec
v1.2.6 real time = 60.611 sec vs v1.3.0 real time = 42.773 sec
Apparently better in version 1.2.6 but the truth is that version 1.3.0 is much faster, but apparently numerically slower
Because now the performance numerically shows slower, but in time is faster?