Offloading to the GPU?

[ervler]

Apologies if this topic has already come up (it probably has, but I'm not getting any search hits with the keywords I've tried).

I'm dealing with a fairly large network (20k nodes) that is bringing my machine to a standstill (Core Duo E8500, 4gb RAM, WinXP, generic GPU).

My question: What aspects of the processing can I offload to my GPU? And, how do I do this?

Specific questions:
- Can I run statistics using my GPU? If so, how do I enable this?
- Do plugins and layouts need to support GPU processing on an individual basis or is it possible to farm everything out to the GPU?

A more general question: Any recommendations on hardware components that are going to give me the best bang for my buck in terms of beefing up Gephi's processing and size of networks that it can handle (e.g. GPU, RAM, CPU, SSD drive, etc.)?

Thanks!

[ervler]

Bump

[admin]

Hi,

Graph drawing only is made by the GPU.

The Parallel ForceAtlas Layout plugin uses GPU: https://gephi.org/plugins/parallel-force-atlas/
But the ForceAtlas2 running on CPU is now more efficient.

Hardware:
- NVidia graphic card (ATI drivers are far more buggy)
- RAM ~ number of nodes
- CPU ~ number of edges (algo complexity)

4gb RAM should be enough for 20k nodes. Don't forget to allow enough memory to the JVM (see this page).

[ervler]

Fantastic thanks. So basically, anything that can take advantage of the GPU and CUDA already is