1. 4
  1.  

  2. 2

    The comparison table is at best misleading.
    The author compares the “PCG family” against other RNGs in an unrealistic configuration.
    For example, Xorshift128 was used in Safari, Firefox, and Chrome (as of 2015).
    Also, which family member and what is the state space size?

    1. 1

      For example, Xorshift128 was used in Safari, Firefox, and Chrome (as of 2015).

      Which makes the comparison table misleading how?

      1. 2

        Here are the reasons I feel this comparison table is misleading.

        • Nonspecific
          PCG has implementations for 8bit, 16bit, 32bit, 64bit, and 128bit integers.
          They won’t all perform the same; narrower widths have a smaller range of possible values.
          Good comparisons are specific, like-to-like, concrete, and falsifiable.
          Compare the PCG table to the xoshino/xoroshiro table.
          The latter is a good comparison and the former is marketing fluff.
        • Realistic
          Most systems have strengths and weakness.
          It’s often possible to make another system look worse by comparing your best case to their worst case.
          Good comparisons reflect how the system is used in the real world.
          This is why the SPEC benchmarks, for example, include gcc, perl, and x264.
          XorShift has variants that improve the quality of the output.
          I mentioned XorShift128+ (my post had a typo) because I feel that it is more representative of a real world XorShift generator.

    Stories with similar links:

    1. PCG, A Family of Better Random Number Generators via friendlysock 1 year ago | 19 points | 12 comments
    2. PCG, A Family of Better Random Number Generators via pushcx 5 years ago | 12 points | 2 comments