Faster and efficient parallel random generator for CUDA, alternatives to cuRand

I tried to check the speedup of cuRand by running a serial code 1 million times and then comparing it with cuda code with 8 blocks each with 1000 threads and running 125 times. I got speedup arounnd 10-11, ideally I should be getting around 1000 but I didn't even get a speedup of 100.

Is cuRand the best option to generate parallel random numbers? Are there any other libraries? Or is it possible that I'm using cuRand wrong(this seems unlikely because I referred to many other codes and found them similar to mine)

I read about GASPRNG but I couldn't find proper documentation for it.