Oops. Can you move my original post to that board ?
For the answer :
I got 32 Go of RAM. Calculator says 32 for a billion, but rendering a billion is just a pain. So i work with subset of 50% or 25% of particles. Those subsets render on 1.5.1 but crash max on 1.5.2 if i go over 10%.
I’ll try with a PRT volume to see if it’s the particle count or the file and post the result here.
Pierre.
I am saving 1 billion particles from a PRT Volume right now to test here.
I don’t know what could be killing 1.5.2 at higher counts - it would be great to confirm with a second set of particle files that it is a general reproducible problem and not limited to a specific particle file.
I was able to reproduce this with 40% of 1BP, we are looking into it.
Just for the fun of it, can you try opening the System Preferences rollout in the new Preferences floater (accessed via the buttons at the bottom of the Main Controls rollout or via the new Krakatoa Menu) and switching from 8 to 4 threads (assuming you have 8 cores).
software.primefocusworld.com/sof … loater.php
software.primefocusworld.com/sof … ollout.php
If 4 cores still crashes, try 2 threads.
It will render slower, but might work around the crash for now.
We know where the bug is (an integer number overflow) and we are looking into fixing it.
Thanks for catching that! The Open Beta is paying off! 
It was fixed internally and the fix will be in the Beta 6 build whenever we release it.
This bug affected only Beta 5 which had the multi-threaded drawing acceleration code rolled in.
Just for the fun of it, I rendered the half billion particles scene lit by one spot light in both the fixed 1.5.2 build and 1.5.1 to see the speed difference.
1.5.2 Beta 6 took 5 minutes and 45 seconds to load, light and render 500MP. The time was split approximately equally between loading, sorting and drawing (about 2 minutes each). Loading time was 1m 57 sec., sorting for light was 1m 07 sec. sorting for drawing was 1m 03 sec, the final pass drawing was 37.7 seconds and the rest was the attenuation calculations (which is now also multi-threaded internally).
1.5.1 took 13 minutes 30 seconds for exactly the same scene with the loading time pretty much the same (1m 55 sec), Sorting for lighting 1m 23 sec, sorting for drawing 1m 18 sec, attenuation calculations 5m 10 seconds and final pass drawing 3 m 17 seconds.
This is on an 8 core Xeon with 16 GB of RAM, using about 15 GB.