Forum

Author Topic: GPU vs CPU  (Read 4445 times)

zhycever

  • Newbie
  • *
  • Posts: 5
    • View Profile
GPU vs CPU
« on: February 20, 2017, 09:04:34 AM »
Currently, I'm running Agisoft on a high performance cluster, It was designed for Bio-macromolecules Analysis which only has CPUs. The admin assigned a node that has no limits on CPU number, 80 at most and the ram is up to 1 T.

My question is that, in which step the agisoft photoscan needs GPU?  My goal is to get dense point cloud.
For now, the program is working fine. It takes 32 hours to process a 900 pictures' project to get dense point cloud. Is this fast comparing to GPU nodes? My future project usually has more than 1500 pictures.

I'm thinking about another GPU cluster which for now doesn't have available node for me. It has: 32 cores CPU, 64G RAM, 2*Tesla K20. How much can improve the performance like reduce processing time etc. Meanwhile, I'm considering install my own workstation using following configuration:
Intel Xeon E5-2683V3 3.2GHz, 12cores
128G RAM
Titan X PASCAL *4
Linux Ubuntu. Is this enough for 1500 or more pictures' project? Can I reduce some?

Setting is
alignment:high, 40000 points,pair selection disabled
dense point cloud: high, aggressive
OpenCL set 80 CPU cores