So, after our chat earlier, I've just gone back through my email history on
build-times and found a sea of information that really closely matches
everything the team mentioned today. Anyway, enough whinging, on with the

Here are my build-timings from back when Kaarlo and I were evaluating the
benefit of getting every dev machine a SSD.  These are NUKE build times
from my Linux machine (the one I still have), using both spindle and an
SSD, from just after we brought in the 1st pass of the RIP code (judging by
my email history).  I don't have equivalent timing for windows because I
didn't have time to test the build-times scientifically (well I did, but
the variance was/is much higher on windows so I needed more runs and my
scripts failed during the one weekend I had to run them). However, not that
that's useful in this context, I do have equivalent results for the
Core/Curve module on Windows. The data just shows that there is/was only
minor benefit of SSDs over spindle HDDs for builds, but checkouts/syncs are
significantly faster with SSDs. The Core/Curve tests also showed that
release builds are *much* slower.

If memory serves builds on Windows at that time were roughly twice as slow
and, in general Linux was faster than Mac (now Mac is fastest). Full builds
on mac take something like 40+ minutes in debug and 1.5 hours or more in

Note that these graphs doesn't show the data for 3.5 minute full-builds
that I mentioned today, as this investigation was done in reaction to the
introduction of the RIP code. For a full build, at this time, it was
actually just over 12 mins.

Each bar-grouping shows a run, so you can get an *indication* of the
std-dev. It looks like I didn't do release-tests on Linux for some reason -
I probably had work to do.

Frank Russel Grant Harrison.

Frank Harrison
Senior Nuke Software Engineer
The Foundry
Tel: +44 (0)20 7968 6828 - Fax: +44 (0)20 7930 8906
Nuke-dev mailing list,

Reply via email to