Ruby 3 JIT was optimized against CPU-intensive demo, which left Rails enthusiasts left out. But what about tasks in the middle of both? Here’s my latest benchmark on Ruby 3 JIT in generating PDF files with Prawn.
A little over a year ago, I created a PDF generation benchmark to help Takashi optimize the upcoming Ruby 3 JIT. I blogged about how to make such a benchmark, and my results for Ruby 2.5, 2.6, and 2.6 –jit.
The JIT wasn’t very successful for this task, but it was very early. So, what about now, that Ruby 3 is out?
My numbers are again for generating a PDF invoice with InvoicePrinter, this time for version 2.1 that supports Ruby 3.
Here are CPU numbers:
Comparison: render /home/strzibny/.rubies/ruby-2.7.0/bin/ruby: 64.6 i/s /home/strzibny/.rubies/ruby-3.0.0/bin/ruby: 61.8 i/s - 1.05x slower /home/strzibny/.rubies/ruby-3.0.0/bin/ruby --jit: 56.3 i/s - 1.15x slower /home/strzibny/.rubies/ruby-2.7.0/bin/ruby: 64.9 i/s /home/strzibny/.rubies/ruby-3.0.0/bin/ruby: 62.0 i/s - 1.05x slower /home/strzibny/.rubies/ruby-3.0.0/bin/ruby --jit: 54.5 i/s - 1.19x slower
As you can see, we are not there yet. Ruby 3.0 is up to 5% slower than 2.7, and Ruby 3 JIT up to 19% slower.
And numbers for memory:
Comparison: render /home/strzibny/.rubies/ruby-3.0.0/bin/ruby: 23896000.0 bytes /home/strzibny/.rubies/ruby-2.7.0/bin/ruby: 25920000.0 bytes - 1.08x larger /home/strzibny/.rubies/ruby-3.0.0/bin/ruby --jit: 79876000.0 bytes - 3.34x larger /home/strzibny/.rubies/ruby-3.0.0/bin/ruby: 24140000.0 bytes /home/strzibny/.rubies/ruby-2.7.0/bin/ruby: 26108000.0 bytes - 1.08x larger /home/strzibny/.rubies/ruby-3.0.0/bin/ruby --jit: 79916000.0 bytes - 3.31x larger
It seems we are at least getting better RAM consumption in Ruby 3.0, Ruby 2.7 takes up to 8% more memory.
As expected, Ruby JIT takes a lot more memory – more than three times. I wonder if somebody runs the Ruby 3.0 JIT in some real workloads because I cannot see myself using the JIT any time soon.
← IT'S OUT NOW
I wrote a complete guide on web application deployment. Ruby with Puma, Python with Gunicorn, NGINX, PostgreSQL, Redis, networking, processes, systemd, backups, and all your usual suspects.