
See Tests_results.txt for information on how good the PractRand standard 
battery of tests is at finding bias in PRNGs.  

Speed of single-threaded testing on my 3.3 GHrz Core i5, using a very fast 32 
bit PRNG, compiled on MSVC 2013 Express:

normal test set, no folding: 9.6 seconds per GB
normal test set, standard folding: 13.5 seconds per GB (recommended)
normal test set, extra folding: 23 seconds per GB
expanded test set, no folding: 25 seconds per GB
expanded test set, standard folding: 32 seconds per GB
expanded test set, extra folding: 54 seconds per GB

Note that these are not counting the time to evaluate the results.  That is 
because the time to evaluate results has very little relationship to the 
amount of data tested.  Usually the time to evaluate results is in the 
neighborhood of 1 or 2 seconds, but under some circumstances (usually 
involving extra folding) it can be significantly slower.  

Individual tests on my machine, no folding:

no tests at all    1.5 s/GB

DC6-9x1B-1         2.7 s/GB  (3.2 s/GB in 0.90)
DC6-6x2B-1         2.5 s/GB
DC6-5x4B-1         2.1 s/GB
DC6-5x8B-101       2.0 s/GB
DC6-4x8B-100       2.0 s/GB

FPF-14+6/32        2.2 s/GB
FPF-14+6/16        3.6 s/GB (5? s/GB in 0.90)
FPF-14+6/8         5.8 s/GB (8? s/GB in 0.90)
FPF-14+6/4        10.5 s/GB (14? s/GB in 0.90)

Gap16              3.1 s/GB  (3.4 s/GB in 0.90)

BCFN params   Original   Unbalanced   FullFreqs   UFF        MultiThreshold
(0,13)        9.9 s/GB   6.8 s/GB     12.0 s/GB   9.5 s/GB   ??? s/GB
(1,13)	      5.8 s/GB   4.1 s/GB     6.9 s/GB    5.7 s/GB   ??? s/GB
(2,13)        3.4 s/GB   2.8 s/GB     4.0 s/GB    3.5 s/GB   ??? s/GB
(3,13)        ??? s/GB   ??? s/GB     ??? s/GB    ??? s/GB   ??? s/GB
