[Rtk-users] No Performance Change After Modifying RTK_CUDA_PROJECTIONS_SLAB_SIZE

[Simon Rit]

Hi,
There is no clear rule on which is the best value. Doing it projection by
projection is slower than doing it a few projections at a time because this
is making a more optimal use of cache memory. However, cache memory is
limited so it is not surprising that the performances don't increase
indefinitely. The best is to test a few values to check which is the best
one.
Best regards,
Simon

On Tue, Mar 18, 2025 at 1:44 PM Anshika G <Anshika.G at panaceamedical.com>
wrote:

> Hi everyone,
>
> While going through the RTK installation documentation, I came across the
> RTK_CUDA_PROJECTIONS_SLAB_SIZE parameter, which can be configured during
> the CMake process. This parameter defines the number of projections
> processed simultaneously during CUDA processing, with the default value set
> to 16.
>
> My understanding is that increasing this value should, in theory, reduce
> the overall runtime of the FDK reconstruction when using CUDA. However, I
> tried increasing it to 32 and 64, but observed no noticeable improvement in
> runtime. I’m using an NVIDIA Quadro RTX 6000 GPU.
>
>
>
> Has anyone else experimented with this parameter or can provide insight
> into why changing the value doesn't seem to affect performance?
>
> Any suggestions or guidance would be appreciated!
>
>
>
> Thank you!
>
>
>
>
> _______________________________________________
> Rtk-users mailing list
> rtk-users at openrtk.org
> https://www.creatis.insa-lyon.fr/mailman/listinfo/rtk-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.creatis.insa-lyon.fr/pipermail/rtk-users/attachments/20250318/78caee9a/attachment.htm>