I also had problems with running out of memory on mesh generation. I eventually figured out how to increase the amount of virtual memory available (SWAP space). Here's how to do it in Windows 7 64-bit:
1. Hit start button
2. Right click on "Computer" and click "properties"
3. On the left side of the newly opened window click "Advanced System Settings"
4. In the "Advanced" tab of the newly opened "System Properties" Window, click "settings under "Performance"
5. Click the "Advanced tab of the newly opened "Performance Options" window
6. Click "Change" under "Virtual Memory".
7. You may have to uncheck "Automatically manage paging file size for all drives" if it isn't already.
8. Select the "Custom size" radio button and increase the initial size and maximum size as you see fit.
Here's a quick screenshot I took of the virtual memory configuration (from steps 7-8) on one of my machines:
Agisoft will (as far as I've been able to determine) use as much virtual memory as is allocated, preventing any crashes caused by meshing.
Beware, using virtual memory may prevent crashes but it will slow down mesh generation considerably.
I've been doing some testing with very high resolution historic aerial photos on different machines and on using network processing with slightly different settings in order to figure out how much hardware (and VRAM) affect processing speed/quality. The most significant findings so far (still looking into it, so these results are quite preliminary) concern two machines:
1. i7: (Custom high end gaming pc I built to do GIS and 3D modeling)OS: Windows 7 Enterprise 64-bit
Motherboard: Gigabyte X79-UD5
CPU: i7-3930K overclocked to 4.00GHz (6 core/12 threads)
GPU: EVGA GeForce GTX 570
RAM: 64 GB DDR3 G-Skill RAM (at I think 1600+, maybe 1800+ can't recall)
Drive: 250GB G-Skill SSD
2. Xeon (Prebuilt Dell Precision Workstation 7910):OS: Windows 7 Enterprise 64-bit
Motherboard: Dell Inc. 0215PR (SOCKET 1) (generic dell motherboard, ugh)
CPU: Intel Xeon E5 v3 @ 3.10GHz (10 cores/20 threads)
GPU: EVGA GeForce GTX 980
RAM: 128 GB DDR4 @ 2133MHz
Drive: 335 GB Intel SSD
The i7 absolutely blows away the Xeon at aligning photos (about 40% faster). If I turn off all CPU cores on the Xeon and let it just use the GTX 980 I can beat the i7 at building the dense point cloud. However, When it comes to building meshing with large enough chunks (think even 10+ 300MB high resolution scanned aerials) the xeon slowly but surely works through it while the i7 pretty much stops processing due to having to utilize virtual memory. I've had to hard shutdown the i7 machine before as all the RAM is pretty much juiced so I can hardly even move the mouse.
Now, I just got the new Xeon machine a couple weeks ago and am still working out all the kinks of network processing, so my observations here may be influenced by my inexperience. However, if you need to work with lots of aerial photos I would highly suggest getting a workstation pc with a Xeon cpu with few cores and the highest possible clock speed you can get (based on observations from
this thread) as well as a really good GPU to speed up the dense point cloud generation (I've been pretty happy with the GTX 980. If you can't afford the Xeon machines (it can be a pretty large jump in price from the i7s) you can go with a super high quality gaming PC (I recall the i7-5960X is one of the best these days) but you won't be able to use more than 64 GB RAM.
Topic: 16GB or 24 GB RAM for Photoscan (Read 21746 times)
