# GPU offload

**URL:** https://discourse.vtk.org/t/gpu-offload/5898
**Category:** Support
**Created:** [June 1, 2021, 8:42am UTC](https://discourse.vtk.org/t/gpu-offload/5898 "2021-06-01T08:42:08Z")
**Posts on this page:** 3
**Page:** 1

<div class="post-metadata">

### Author: ![pourarab](https://discourse.vtk.org/user_avatar/discourse.vtk.org/pourarab/32/1720_2.png) [@pourarab](https://discourse.vtk.org/u/pourarab)
#### Post date: [June 1, 2021, 8:42am UTC](https://discourse.vtk.org/t/gpu-offload/5898/1 "2021-06-01T08:42:08Z")

</div>

Hello,

I’m investigating using OpenMP GPU offload to run vtkCellTreeLocator-\>FindCellsWithinBounds on GPU. Is it possible to map it via OpenMP ? If not, is there any solution to do it as parallel execution on GPUs ?

Thanks,  
Danial

---

<div class="post-metadata">

### Author: ![Paulo\_Carvalho](https://discourse.vtk.org/user_avatar/discourse.vtk.org/paulo_carvalho/32/370_2.png) [@Paulo\_Carvalho](https://discourse.vtk.org/u/Paulo_Carvalho)
#### Post date: [June 1, 2021, 10:00pm UTC](https://discourse.vtk.org/t/gpu-offload/5898/2 "2021-06-01T22:00:23Z")

</div>

Hello,

Benefits from GPU processing come from certain types of computations such as matrix/kernel-oriented ones (e.g.: tensor processing in machine learning), mass vectorized operations that can be implemented as texture operations via shaders, etc… I don’t think, if doable, spatial searches can perform better in the GPU, considering transporting the data back and forth CPU memory \<–\> GPU memory is major bottleneck that will likely offset or even worse any performance gain for that case.

regards,

Paulo

---

<div class="post-metadata">

### Author: ![pourarab](https://discourse.vtk.org/user_avatar/discourse.vtk.org/pourarab/32/1720_2.png) [@pourarab](https://discourse.vtk.org/u/pourarab)
#### Post date: [June 2, 2021, 8:45am UTC](https://discourse.vtk.org/t/gpu-offload/5898/3 "2021-06-02T08:45:53Z")

</div>

Hello Paulo,

Thank you for your answer.

Best regards,  
Danial