# best way to use paraview for AWS EC2?

**URL:** https://discourse.paraview.org/t/best-way-to-use-paraview-for-aws-ec2/7710
**Category:** ParaView Support
**Created:** [July 28, 2021, 8:53am UTC](https://discourse.paraview.org/t/best-way-to-use-paraview-for-aws-ec2/7710 "2021-07-28T08:53:20Z")
**Posts on this page:** 2
**Page:** 1

<div class="post-metadata">

### Author: ![Sina](https://discourse.paraview.org/letter_avatar_proxy/v4/letter/s/ba8739/32.png) [@Sina](https://discourse.paraview.org/u/Sina)
#### Post date: [July 28, 2021, 8:53am UTC](https://discourse.paraview.org/t/best-way-to-use-paraview-for-aws-ec2/7710/1 "2021-07-28T08:53:20Z")

</div>

Hello,

I am using AWS EC2 graphic instance (x32 CPU, 128GB RAM, 1 GPU) regularly for post processing of CFD (OpenFOAM) cases. We usually use NICE DCV to connect to the machine with GUI and from there open up a Paraview session. Despite having 32 cpus, 128GB RAM and a gpu, we sometimes struggle to load and/or work/render openfoam cases as it gets very slow. We usually work with 30-50 million cells cases.

So my question is what is the best setup for my case?

- Enabling multicore support from setting?
- Using Paraview Server and Client both on AWS EC2?
- Using Paraview Server on AWS EC2 and Client on my laptop (x12 CPUS, 32GB RAM, 1 GPU)? Is there a way that my laptop gnpu/cpu can be shared to increase the efficiency of interacting with paraview?
- Or if the current workflow is as efficient as we can get with this instance type.

Cheers,  
Sina

---

<div class="post-metadata">

### Author: ![Will\_Schroeder](https://discourse.paraview.org/user_avatar/discourse.paraview.org/will_schroeder/32/7258_2.png) [@Will\_Schroeder](https://discourse.paraview.org/u/Will_Schroeder)
#### Post date: [July 30, 2021, 9:54am UTC](https://discourse.paraview.org/t/best-way-to-use-paraview-for-aws-ec2/7710/2 "2021-07-30T09:54:58Z")

</div>

Sina, I routinely work with 0.5 gigacell (tets,hexes) unstructured grid datasets on Ubuntu 20 core / 64gbyte machines - so a 30-50 million cell case looks modest in size to me. I enable SMP using TBB (VTK\_SMP\_IMPLEMENTATION\_TYPE=TBB). It could be that a filter in your workflow is not behaving well - there are have been dramatic performance improvements made to VTK/PV in the last couple of years - but there are still filters that need improvement. Can you elaborate on your workflow, what filters are you using, and what looks to be the bottlenecks?