Forum

Author Topic: Building a machine for Photoscan/Cloudcompare, looking for feedback  (Read 4630 times)

gilbertsmith

  • Newbie
  • *
  • Posts: 4
    • View Profile
Hi there, I'm researching a machine for a client to run Photoscan and Cloudcompare. I'm not very familiar with either program myself so I'm not too sure what they require outside of lots of RAM, CPU, and OpenCL performance.

I read on one post (dated 2013) that Photoscan doesn't benefit from SLI or Crossfire. Is this still true?

My client is wanting to run a 3D monitor with NVIDIA 3DVision, but it also seems that AMD cards perform better with OpenCL and are a better choice for this setup. If I were to run a lower end NVIDIA card to power the display, and an AMD card for the processing, would that work? I want to make sure the AMD card won't just be ignored if it's not the primary video card.

That said, here's the parts we're looking at right now:

2 x Intel Xeon E5-2698
ASUS Z10PE-D8 motherboard
2 x Kingston DDR4 128GB RAM kits

Video card I'm not sure about, leaning towards either a Titan X if I can't run AMD and NVIDIA together for some reason, or a Fury X and a lower end NVIDIA card if I can.

Any feedback is greatly appreciated!

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #1 on: November 29, 2015, 01:30:42 PM »
Change the CPU for this one
http://ark.intel.com/products/81909/Intel-Xeon-Processor-E5-2687W-v3-25M-Cache-3_10-GHz.
its much faster as the E5-2698
----------------
www.mhb.sk

doomz

  • Newbie
  • *
  • Posts: 5
    • View Profile
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #2 on: November 29, 2015, 04:12:54 PM »
I saw somewhere comparing benchmark results between Xeon E5-2687W v3 and and Xeon E5-2690 v3. Overall performance goes to the 2690v3 with a wider margin. The 2687W v3 provides better performance I guess in mesh generation.

gilbertsmith

  • Newbie
  • *
  • Posts: 4
    • View Profile
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #3 on: November 30, 2015, 01:16:39 AM »
That seems odd.. The 2698 is a lot faster in the benchmarks I've looked at.. double the cache, double the cores, double the threads.. The only thing the 2687 beats it on is clock speed.. but it looks like the 2698 beats it on IPC as well, so what makes it faster?

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #4 on: November 30, 2015, 03:20:38 AM »
not everytime is better to have MANY cores not every sw CAN be programmed to use all cores  for full eficiency

the you prefered is good for server stuff = many  VM  stuff, and Rendering  but not for overall stuff

you need Workstation CPU to get max out of it...

Already many times discussed here, more cores SLOWER PC recon times...

----------------
www.mhb.sk

gilbertsmith

  • Newbie
  • *
  • Posts: 4
    • View Profile
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #5 on: November 30, 2015, 06:08:04 AM »
Ok, any thoughts on SLI/Crossfire? Is a Fury X the best card for Photoscan? How does a Titan X stack up to it, or Quadro cards?

gilbertsmith

  • Newbie
  • *
  • Posts: 4
    • View Profile
Re: Building a machine for Photoscan/Cloudcompare, looking for feedback
« Reply #6 on: December 03, 2015, 07:17:15 PM »
If anyone else is wondering about this, I went ahead and did some testing of my own with the Photoscan trial and the Doll set on the Agisoft site.

First up, there doesn't seem to be any problems running NVIDIA and AMD cards. Photoscan will use an AMD card to process while an NVIDIA card is running the display, so that's good.

Second, I didn't see any benefit from Crossfire, but I did see a benefit from having multiple GPUs. Here's some numbers I got for the Dense Point Cloud generation:

No GPU, 4 CPU cores: 503s, 500s
Radeon 7870, 3 CPU cores: 343s, 351s
Radeon 7870 x 2, No Crossfire, 2 CPU cores: 302s, 302s (~16% faster)
Radeon 7870 x 2 in Crossfire, 2 CPU cores: 308s, 307s

So it looks like Crossfire might actually hurt performance a bit, but doesn't seem to help at all.

Hopefully this helps anyone else with the same questions I had.