Using hardware acceleration on a single node

[jameskress]

Hi All,

I am trying to understand how hardware acceleration currently works in VisIt. It looks like when I run a serial engine on a localhost that it makes use of the GPU. THough when I have more than one GPU, how does it pick, or does it use both?

Now let's say I want to run a parallel engine on a localhost, will VisIt still be able to use one or more GPU's? Do I now need to specify a special flag? Will this work with the binary release, or it is something I need to compile to work?

I have some users that are pushing to make sure that the GPU is doing their rendering of their batch jobs, but wanted to understand how VisIt used GPU's, and to make sure we don't need to launch only the number of mpui procs == number of GPU on the system.

Appreciate any insight!

Cheers,

[JustinPrivitera]

There are multiple ways VisIt can use hardware acceleration. Using GPUs to do rendering usually does not pay off in a multi-node situation. It can also use vtk-m (Viskores) for a limited set of operations. In the future rendering with ANARI may use GPUs.