YJIT Benchmarks

YJIT metrics from the yjit-bench suite as of 2024-10-11
Using the geomean of the headline benchmarks for x86 YJIT 3.4.0dev is
  • 74.2% faster than CRuby 3.4.0dev
  • 4.5% faster than YJIT 3.3.4
On railsbench it is
  • 79.4% faster than CRuby 3.4.0dev
  • 2.8% faster than YJIT 3.3.4
Latest Full Details:
YJIT 3.4.0dev build configuration
Ruby
ruby 3.4.0dev (2024-10-10T20:03:40Z :detached: 09100508e6) +YJIT +PRISM [x86_64-linux]
Configure args
--disable-install-doc --disable-install-rdoc --enable-yjit
Compiler
gcc (Ubuntu 13.1.0-8ubuntu1~20.04.2) 13.1.0

Latest Headlining Results vs CRuby

These "headlining" benchmarks are used for the above measurements. There are more benchmarks on the Full Details Page.

Performance on Headline Benchmarks

0.0 0.5 1.0 1.5 2.0 2.5 CRuby 3.3.4 CRuby 3.4.0dev YJIT 3.3.4 YJIT 3.4.0dev activerecord chunky-png erubi-rails hexapdf liquid-c liquid-compile liquid-render lobsters mail psych-load railsbench rubocop ruby-lsp sequel geomean* 1.04 2.35 2.48 1.00 1.67 1.82 0.98 1.88 1.80 1.01 1.66 1.84 1.01 1.40 1.48 1.02 1.45 1.54 0.99 2.46 2.59 0.99 1.53 1.53 1.00 1.41 1.42 1.01 1.52 1.78 0.99 1.74 1.79 1.00 1.75 1.82 1.04 1.66 1.68 0.99 1.35 1.36 1.00 1.67 1.75 Speed of each Ruby implementation relative to the baseline CRuby measurement. Higher is better.

Memory Usage on Headline Benchmarks

0.0 0.5 1.0 1.5 2.0 CRuby 3.3.4 CRuby 3.4.0dev YJIT 3.3.4 YJIT 3.4.0dev activerecord chunky-png erubi-rails hexapdf liquid-c liquid-compile liquid-render lobsters mail psych-load railsbench rubocop ruby-lsp sequel geomean* 1.04 1.17 1.12 0.89 1.20 0.87 0.98 1.08 1.05 1.03 1.94 2.09 1.04 1.09 1.16 1.04 1.10 1.21 1.02 1.17 1.27 0.90 1.09 1.06 1.07 1.06 1.17 1.01 1.12 1.20 0.98 1.14 1.20 1.00 1.35 1.41 1.04 1.09 1.16 1.00 1.09 1.13 1.00 1.18 1.20 Memory usage of each Ruby implementation relative to the baseline CRuby measurement. Lower is better.

YJIT Results Over Time

  •    x86_64 railsbench   
  •    x86_64 optcarrot   
  •    x86_64 liquid-render   
  •    x86_64 activerecord   
Y axis values are the total number of seconds to run the benchmark one time - lower is better.