[Gate-users] Gate on GPU

Ezzat Elmoujarkach ezzat.elmo at gmail.com
Tue Aug 28 09:12:50 CEST 2018


Hello Tom,

For the first question I would assume so. I never looked in the RT
applications since all my work so far are in imaging. However, I believe
you also should add the GPU lines for the GPU ID and buffer size.

For the second question. By theory yes, it should speed up the simulation
due to the nature of how the GPU runs calculations vs CPU where GPU can
have thousands of CUDA cores to finish the job quicker. I run a test
simulation ( did not have the time yet to run a full simulation) and was
using a GTX 920M card on a laptop with i7 intel core ( sorry cannot
remember the full specs). The results showed that the GPU was 12 times
faster than the CPU simulation.

Sorry I couldn't be of much help and I hope others can share the experience
in this area.

Best Regards,

Ezzat

On Tue, Aug 28, 2018 at 1:20 AM G Tom <gregthom992 at gmail.com> wrote:

> HI all,
>
> I see that some of you have tried to install GAte with GPU support and I
> have some questions:
>
> First of all,  does gate support Radiotherapy simulations on the GPU?
> I found this actor GateGPUPhotRadTheraActor but didn't find any examples;
> could I simply replace my current actor with "/gate/actor/addActor
> GateGPUPhotRadTheraActor   " ?
>
> Secondly, is there expected to be a speed up when using GateGPU compared
> to when using GateCPU ? Specifically, for a small local cluster of 40
> cores  (5 nodes Xeon(R) CPUs , E5420  @ 2.50GHz, with 8 cores each)
> compared to using GPU enabled gate ( GeForce GTX 1080) on one node ?
>
> Appreciate any opinions as I embark on this journey.
>
> Thanks
>
> GT
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opengatecollaboration.org/pipermail/gate-users/attachments/20180828/96d4c332/attachment-0001.html>


More information about the Gate-users mailing list