+1
These benchmarks seem to test the performance of specific algorithms, implemented in different languages. Including start up or compile time doesn’t make any sense to me in that case. From: [email protected] [mailto:[email protected]] On Behalf Of Stefan Karpinski Sent: Wednesday, January 27, 2016 2:39 PM To: Julia Users <[email protected]> Subject: Re: [julia-users] Re: Julia Benchmarks Feedback It's true. However interpreters don't take much time to start up – an interpreter is the lowest latency way to get from source code to running program (compared to AOT or JIT compilation). There's an argument to be made for including all of the things that take time when benchmarking. However, we don't time benchmarks because we're interested in how fast language implementations can run the benchmarks – we're trying to get some sense of how fast they'll run real programs. And if you care about speed, you probably have a program that's going to take a non-trivial amount of time to run – which means that runtime startup and JIT are going to become negligible. But we don't really want benchmarks that run for minutes. So instead we eliminate fixed overheads like runtime startup and JIT time when measuring benchmarks – not because they doesn't happen, but because they're a fixed cost that ends up being insignificant for real life programs. The only scenario where I can imagine really caring about runtime startup and JIT is if you were evaluation how good a language implementation is for running short-lived command-line programs. Then it would absolutely make sense to measure runtime startup and JIT. But I don't think that's what these benchmarks are trying to measure. On Wed, Jan 27, 2016 at 4:48 PM, Ismael Venegas Castelló <[email protected] <mailto:[email protected]> > wrote: but then again the benchmarks are flawed because other implementations are also timing the interpreters startup time. El martes, 26 de enero de 2016, 22:07:54 (UTC-6), George escribió: I was surprised to se the results on the following benchmarks: <https://github.com/kostya/benchmarks> https://github.com/kostya/benchmarks Some benchmarks of different languages Brainfuck <https://github.com/kostya/benchmarks/tree/master/brainfuck> Brainfuck bench.b Language Time,s Memory, Mb Nim Clang 3.21 0.7 Felix 4.07 1.3 Nim Gcc 4.52 0.6 Java 4.94 147.6 C++ 5.08 1.1 Rust 5.46 4.9 Scala 5.90 116.3 D 6.57 1.0 D Ldc 6.61 0.9 Crystal 6.97 1.3 Go 7.29 1.3 Javascript Node 8.74 15.0 D Gdc 8.87 1.0 Julia 9.25 59.0 Javascript V8 9.41 8.1 Go Gcc 13.60 10.0 Python Pypy 13.94 55.4 Javascript Jx 17.14 11.0 C# Mono 18.08 15.4 OOC 48.86 1.3 Ruby JRuby 87.05 124.1 Ruby Topaz 112.91 36.0 Ruby JRuby9K 160.15 297.2 Ruby 226.86 8.0 Tcl 262.20 2.7 Python 452.44 4.9 Ruby Rbx 472.08 45.0 Python3 480.78 5.5 mandel.b <https://github.com/kostya/benchmarks/blob/master/brainfuck/mandel.b> Mandel in Brainfuck Language Time,s Memory, Mb Nim Clang 28.96 1.0 Felix 40.06 3.7 D Ldc 43.30 0.9 D 45.29 1.2 Rust 46.34 4.9 Crystal 48.62 1.3 Nim Gcc 50.45 0.9 Go 52.56 7.6 Java 55.14 69.9 Cpp 56.63 1.1 Scala 64.37 126.4 D Gdc 70.12 1.5 Go Gcc 85.67 10.7 Javascript Node 92.65 15.8 Julia 94.33 56.9 ...
