Response
Here are some general tips regarding optimal hardware configuration for PhotoScan processing:
CPU: effects on the processing speed on every processing step. Higher frequency and bigger number of cores will result in faster processing. Usually we recommend the following six-core CPUs as optimal desktop solutions for PhotoScan processing, like Intel i7 4960X (3.60 GHz). In case the processing tasks require more than 64 GB RAM we can suggest to use single-socket Xeon-based configuration with E5-1660 v2 (3.50 GHz). For jobs that require 128+ GB RAM double-socket Xeon workstations should be considered, for example, based on E5-2667v2 processor.
You can check the following benchmark page for some comparative information on different CPUs:
http://www.anandtech.com/bench/CPU/1057 GPU: is used only for depth maps estimation (sub-step of Build Dense Cloud stage) but improves the speed of this process significantly. The recommended cards are the following: nVidia GeForce GTX 580, GTX 680, GTX 780 or GTX TITAN (GTX 780 Ti and GTX TITAN Black edition would be the fastest, however, we haven't yet tested GTX 980). Several GPUs can be used for additional speed-up. Also some of PhotoScan users successfully utilize ATI Radeon HD 7970 and Radeon R9 290x cards and report that the processing performance could be even better than nVidia cards on some datasets, but due to instable driver support usually we do not recommend ATI cards. Also note that professional class graphic cards like Quadro are not that effective in terms of PhotoScan performance and also may have visualization problems when dealing with high-poly meshes.
And here is another thread regarding the GPU performance benchmarks based on users experience:
http://www.agisoft.com/forum/index.php?topic=651.msg10384#msg10384Please also note that nVidia drivers 340.52 and later cause OpenCL computations failure, so we recommend to use the previous driver version. The problem will be fixed in the next PhotoScan version (1.1.0).
RAM: allows to process bigger number of photos in the single chunk and/or to achieve higher reconstruction quality, since the model generation stage is very resource demanding. Usually we recommend to install 64 GB RAM (maximal possible on the desktops), if the processing tasks requires high reconstruction quality or large amount of images. Approximate memory requirements for each processing stage depending on the number/resolution of images and reconstruction parameters used can be checked in the following document:
http://www.agisoft.com/pdf/tips_and_tricks/PhotoScan_Memory_Requirements.pdf--
Best regards,
AgiSoft team.
Topic: Computer Build (Read 3693 times)
