The new random number generators in 3.2rc2 are a great addition. Thanks Nvidia.
The documentation states that “The XORWOW generator was proposed by Marsaglia  and has been tested using the TestU01 “Crush” framework of tests . The full suite of NIST pseudorandomness tests  has also been run.” So … how did it do? Should we read “has been tested using” to mean “has passed all of the tests from”?  doesn’t include XORWOW in its results (Brent-xor4096s is similar but uses 4 xorshifts rather than 3) so presumably XORWOW has been independently tested. Did it pass all of the tests from TestU01 “Crush” or just do pretty well? Has it been tested on TestU01 “BigCrush”? How did it do on the full NIST suite?
PS: I note from  that the Mersenne twister (MT19937) failed two tests from TestU01 “Crush”. And of course it’s not well suited to GPGPU. So thanks Nvidia for taking the trouble to find something better.