Forum

Author Topic: Question about generating mesh  (Read 5750 times)

andyroo

  • Sr. Member
  • ****
  • Posts: 438
    • View Profile
Question about generating mesh
« on: September 09, 2013, 10:58:23 PM »
Mostly for my own curiousity and education, I am wondering why mesh generation doesn't take full advantage of either my CPUs or my GPUs, or in other words, what is the bottleneck on mesh generation. I am using the PS 1.0 beta, but this seems like it's always been the case for me, so posting in general forum.m

It seems like many other tasks are limited by CPU or GPU or both - where I can look at my performance and see 100% utilization or close to that. On mesh generation my CPU and GPU usage is only in the 15-20% range.

My system: Dell T7500 with two Intel Xeon X5647 2.93 GHz CPUs, 192GB ECC memory, two Gigabyte AMD 7970 GPUs.

Screenshot of mesh progress, CPU/mem use and PS windows attached.

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Question about generating mesh
« Reply #1 on: September 10, 2013, 12:20:42 AM »
jsut the PSU ( power source )model please.....
----------------
www.mhb.sk

andyroo

  • Sr. Member
  • ****
  • Posts: 438
    • View Profile
Re: Question about generating mesh
« Reply #2 on: September 10, 2013, 01:20:53 AM »
I should clarify, since I restarted generating the mesh, that it looks like it's the triangulation step that was slow on my machine (where it was in the screenshot). I restarted mesh generation after the crash (disabled all CPUs and primary 7970) and all CPUs are pegged right now. I am at bold step in mesh generation log below (copied logfile entries from last session)

Generating mesh...
generating 106392x27946 grid (0.00339475 resolution)
grid rasterized in 2.409 sec
filtering median... done in 13.429 sec
filtering idw... done in 45.749 sec
constructed triangulation from 5923401 vertices, 11846796 faces
grid interpolated in 14095.4 sec
triangulating...

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Question about generating mesh
« Reply #3 on: September 10, 2013, 10:35:04 AM »
Hello andyroo,

GPUs are not used during mesh generation process, they are only utilized on depth maps reconstruction step.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Porly

  • Jr. Member
  • **
  • Posts: 77
    • View Profile
Re: Question about generating mesh
« Reply #4 on: September 10, 2013, 12:26:35 PM »
Hello,

my suggestion is that for building mesh step, the RAM speed has the biggest influence. For the speed of the process it is not depending on how much RAM you have, but on the freuquency of your RAM.

The faster your RAM, the bigger is your CPU usage, since the CPU distributes the work in these process.

This is only my suggestion.


Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Question about generating mesh
« Reply #5 on: September 10, 2013, 01:28:55 PM »
Poorly, the ram influence on REAL app usage is 1-2 % - read carefully this link and its compare even the dual and Quad config
http://www.xbitlabs.com/articles/memory/display/lga2011-ddr3_3.html
----------------
www.mhb.sk

Porly

  • Jr. Member
  • **
  • Posts: 77
    • View Profile
Re: Question about generating mesh
« Reply #6 on: September 10, 2013, 01:47:55 PM »
Quote
The most important conclusion made from the graphs above is that reducing the number of memory channels in LGA 2011 platforms doesn’t lead to any catastrophic consequences. Yes, the performance drops, but only as little as by 1-2%, which you will barely feel in real applications.

For me it sounds like they are comparing the different perfomence depending on the number of memory channels, and not the usage of RAM in real APPs. So that 1-2% is the performence drop???

However it was only a suggestion :)

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Question about generating mesh
« Reply #7 on: September 10, 2013, 01:54:04 PM »
see better they use 1600 and 1800 Mhz RAm modules and 2 and 4 of them. Just theoretically 4 channels should be faster, and in SYNTHETIC it show niiiice results but in REAL app you get just so small benefits.... its just marketing stuff to sell expensive hifreq modules. in games thy can get few more frames more of that setup, but paying horrible price for that - extreme user.... is better to have 32-64 GB 1333 ram as 16GB 2400 Mhz or similar.... its about the deeper knowledge how its work inside PC......   
----------------
www.mhb.sk

andyroo

  • Sr. Member
  • ****
  • Posts: 438
    • View Profile
Re: Question about generating mesh
« Reply #8 on: September 10, 2013, 08:48:33 PM »
Hmm. looking back through my logfiles I think maybe I misspoke and this is more of an issue with v 1.0 beta, and I know a lot changed with the mesh generation step. maybe this should be moved to that thread?

Or maybe it's an issue with what's recorded in the logfile... either way I just went back through my logfiles and it looks like when I started playing w the beta the triangulation time reported
increased by a lot, like an order of magnitude. I made a pretty chart and was about to attach it, but then my computer crashed, so now I'm just going to end here. Nice of my browser or this forum to save the entered text while I rebooted.