GPU and video card memory

Hardware
1

I was wondering if there is a way to use GPU and video memory to speed up the analyses with FLAC3D.
Thank you in advance.

1 Like
2

Hello,
At this time FLAC3D does not utilize GPU acceleration (or similar) to speed up cycling.

3

Will this feature be supported in future?

4

In 3DEC is there this possibility?
Thank you

5

Thanks for everyone’s questions concerning GPU acceleration. Itasca has spent a lot of time thinking about what the best approach is for speeding up modeling and we have investigated using GPUs, but there are basically three reasons that we aren’t currently utilizing them:

  1. Itasca would need to do rewrite a lot of the software because GPUs use their own programming language, and this would be a major undertaking.

  2. Itasca programs are very interactive – you can visualize results as they are being calculated and even have FISH functions performing calculations during cycling as modeling is ongoing. This sort of communication overhead during cycling is not conducive to the GPU approach, which is more ideal more for quickly crunching a calculation that gets sent up to the GPU and sends back as some result. If we need to do a lot of communication back and forth, we lose some of that speed advantage.

  3. Finally, there is a limit on how big you can go using GPUs and Itasca’s made a strategic decision to move towards building a cluster version using MPI, which is theoretically unlimited since you can just keep adding nodes to the system make bigger and bigger models. Creating a Linux version and web licenses is moving Itasca software towards a cluster version.

This is not saying that Itasca will never utilize GPUs; it’s certainly something we’re still thinking about. But in terms of ”bang for the buck”, we have decided to go for the cluster route rather than the GPU route – at least for now.

2 Likes
6

Thanks for the comprehensive answer.