Favorite Files

No favorite files added yet

Recent Posts

1. Comment - Before it can tackle Windows, Chrome must leave Safari in the dust

   (Jul 10, 2009 - 12:07 PM)

   You appear to be guessing wildly at how the V8 benchmark suite works.
   
   If you read the code, it's actually quite simple: Each part of the benchmark runs in a loop that counts how many times it can run in a set time (scaled by some arbitrary value). If you can execute the benchmark twice as fast, you get twice the score.
   
   Then the final score is counted as the geometric mean of the part scores, meaning that getting twice as fast on one part gives the same final score as getting twice as fast on another.
   If you are twice as fast on *every* test, you get twice the final score.
   
   Nothing exponential in that. Quite the contrary actually.
   
   I'm guessing the main reason for the huge difference is that the V8 benchmark suite is not micro-benchmarks that can finish in 5 ms, but actually test serious workloads. In particular the memory management benchmark (Splay) really tests something that I haven't seen any other benchmark test (and I guess V8 is scoring so well because they have actually cared to optimize their memory management and create a benchmark for it).