Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

I tested my well-optimised R code and saw only 3x to 10x performance gain. That's still not substantial enough currently to migrate a whole code base, in particular given that the libraries are also still not mature enough. The research group I'm working with also have no interest in adopting anything new, In fact most of our code is still in FORTRAN so that is something I would be more interested in migrating to Julia but I don't think that is happening anytime in the next 10 years


view as:

>only 3x to 10x performance gain

That sounds like a lot.

I can see why maturity might be an issue, but after the word only I'd expect something like 5-10%, not integer multiples.


It does sound like a lot but it depends on the actual wall clock time. If your run time goes from 10 days down to 1 day then yes, it matters a lot. If it goes from 1s down to 0.1s it might not matter so much.

Of course, but if you observe 3x to 10x performance gains across the board, you will have some programs that run in more than 1 second where it may be worthwhile.

I would call that a huge performance gain! Perhaps Julia can be chosen for new projects; it is rarely worth rewriting old ones.

I guess whether it's effective to migrate a code base depends a lot on where your costs are.

If you pay 500k$ for compute, it might become worthwhile to invest time into rewriting hot paths.


Seems like it would still be useful use Julia as the backend for the R package instead of Fortran. I've been showcasing a lot of that lately with good success:

https://www.stochasticlifestyle.com/juliacall-update-automat...

https://www.stochasticlifestyle.com/gpu-accelerated-ode-solv...


Realistically, how much did you expect over mature R libraries or Fortran? 1000x or 10_000x? I'd consider 3-10x a major gain.

Legal | privacy