This should be possible – you may have to tell the application what GPU you 
would like it to use by setting an environment variable.  I run on a machine 
with four GPUs and can individually select which one I want to use by calling 
‘CUDA_VISIBLE_DEVICES=0 rtkfdk ...’  or ‘CUDA_VISIBLE_DEVICES=1 rtkfdk …’ which 
exposes the target GPU to the running application.  This is a very manual, 
command line approach, however the CUDA_VISIBLE_DEVICES variable can also be 
set to a unique value for each user in their respective user profile (e.g. 
setting ‘CUDA_VISIBLE_DEVICES=1’ for USER1 in ~/.bashrc on a Linux machine).  
This second approach makes it so that each user has a dedicated GPU at their 
disposal whenever they need it.

Hope this helps.

— Matt

Matthew J. Riblett
Virginia Commonwealth University
Department of Radiation Oncology
Medical Physics Graduate Program

Office:  Sanger Hall, Room B1-013
401 College Street   |  P.O. Box 980058
Richmond, Virginia 23298

> On Feb 9, 2018, at 9:07 AM, Vincent Libertiaux <> wrote:
> Hello Simon,
> hello everyone.
> I have a computer that I would like to use as a CT reconstruction server for 
> several users, let's say 3.  I was wondering if it is possible to run 3 
> simultaneous reconstructions with RTK, one on each GPU ?
> I thank you very much in advance for any answer or clue.
> Best regards,
> Vincent
> _______________________________________________
> Rtk-users mailing list

Rtk-users mailing list

Reply via email to