Hardware for Agisoft

[GrinGEO]

Im checking out the possibilities to do a PC upgrade for Agisoft & Autodesk Civil 3D

My idea is

Intel Core i7-5820K Hexa-Core
8x8GB DDR4 Kit (total 64GB)
Radeon R390 8GB GPU


Does it make sense to go for 128 GB of RAM or is it better to have a second Radeon R390 8 GB GPU installed? Are AMD Radeon Multi GPUs supportet in Agisoft, since all reccomandations are only talking about NVIDIA GPUs? How is the performance impact? What are your suggests?

Should the pictures all be on a SSD or is a SSHD good enough?
Thanks for helping

[shaav@hotmail.de]

If you don't use your pictures on your SSHD regulearly, then your SSHD handles the workload as fast as a normal HDD. If that matters? I don't know.

A second GPU would make the progress of creating a dense cloud much faster:
https://www.pugetsystems.com/labs/articles/Agisoft-PhotoScan-GPU-Acceleration-710/

On the other hand, if you run out of memory, you will have a hard time, too.


If I would be you, I would go for several cheap GPUs, but without knowing how many photos/megapixel per photo you have it is hard to give you the optimal awnser.

[Alexey Pasumansky]

Hello GrinGEO,

According to our experience there's not much difference in performance for R9 290 and R9 390 cards, so it might be reasonable investing in R9 290x that outperforms regular cards. Also you can consider older Radeon cards like HD7970 that has similar performance to GTX TITAN and GTX 980.

The amount of RAM depend on the size of your projects and desired quality (also depends on the mesh reconstruction method you are planning to use).

[GrinGEO]

Really thanks for your great comments

you are right, mostly those GPUs are just relabelt from 290 to 390 series

But the 390 series has 8 GB of DDR5 RAM which should improve a lot the memory performance. And a 290X costs almost the same as a new 390X or even higher.

390 & 390x have now very similar price points

on pugetsystems they compare only NVIDIA, so Im not sure how it works for Radeons, but probably it should also give 35% more power. On the other hand, you need to disable one CPU core for each GPU, so you lose one more. What do you think?

we have decided to go for 4x 16 GB DDR4 ram with the option to upgrade to 128 GB later

regarding SSD, does anyone of you use it? I have them on my video editing workstation and really happy since I have multiple streams in 4K, but Im not sure if it is needed for Agisoft

[shaav@hotmail.de]

On the other hand, you need to disable one CPU core for each GPU, so you lose one more. What do you think?

The GPU is by far faster than the CPU. In my PC, my GTX 970 is ten times faster than a single core of my core i5-3570k @3,8Ghz. I did some tests to find out how I can achieve the best performence:

Dense Cloud medium Quality:
4/4 CPU cores disabled + GTX 970: 650 mil.pixels/sec
3/4 CPU cores disabled + GTX 970: 676 mil.pixels/sec
3/4 CPU cores disabled + GTX 970: 72 mil.pixels/sec
2/4 CPU cores disabled + GTX 970: 709 mil.pixels/sec
1/4 CPU cores disabled + GTX 970: 687 mil.pixels/sec
0/4 CPU cores disabled + GTX 970: 619 mil.pixels/sec

Results:
- sometimes you have the best performence when you disable two cores instead of one.
- The GPU is a hell lot faster than the CPU

[GrinGEO]

thanks for helping. than with the option of 2x R9 390 with a Hexa Core CPU & 128GB RAM I should have a great performance...

[GrinGEO]

Im wondering if the GPUs needs to be connected via SLI or CrossFire Adapter or if they are just inserted and running as 2 separate GPUs in Agisoft...

can anyone help regarding this?
Thanks

[MeHoo]

Two separate.  No SLI needed,  Actually.. disable it.

My issue, currently, is in the fact that only PART of the dense cloud construction uses the GPU.. a very small part.. I am finding that 90% of my processing time with 3x dedicated GPUs is still cPU bound... this is really upsetting.

60 hour scans really need to get faster because Reality Capture is dominating the speeds...