[Gate-users] Methods speed up GATE simulation

[Zhengzhi Liu]

Dear Gate users,

I am simulating a  NEMA body phantom
<http://www.spect.com/pub/NEMA_IEC_Body_Phantom_Set.pdf> in a PETscaner
system in GATE. The scan time is 1600 seconds (4 passes, 40 couch stations,
10 seconds per station). I run this simulation on 56 cores cluster using
"gjs -numberofsplit 50 -clusterplatform condor -condorscript main.mac". It
has been about 2 days and looks like it has not finished half of the
simulation yet. Thus, I looked up online trying to find ways to speed up
the simulation.

Here are the common methods I found:

   1. use multicores cluster
   2. run GATE on GPU
   3. use back-to-back gamma source rather than ion sources
   4. apply cuts
   5. reduce the amount of data to be written and prefer root over ASCII

I have already run 50 copies Gate simulation simultaneously, used
back-to-back gamma source, and been collecting coincidence data only.  I
feel it is still very slow. For my application to simulate a real PET scan
of 25 minutes, it may take 3 to 4 days, which is quite hard for me to wait
for so long to get the data.

Here are my questions, I hope to hear an answer/suggestion from GATE
experts.

   1. I have not tried to run GATE on GPU yet, generally speaking, how many
   times faster is GATE GPU version than CPU version?
   2. How much speed gain can cuts achieve? A few percentages, tens of
   percentage, or a couple of times faster? I hesitate to apply cuts and worry
   that it may mess up the result.
   3. Is there any other way to significantly boost the speed on top of
   these mentioned above?

Thank you very much for your reading and any suggestion.
Have a nice/peaceful day.

Sincere wishes,
Zhengzhi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opengatecollaboration.org/pipermail/gate-users/attachments/20200406/5f8f89dc/attachment.html>