Forum

Author Topic: Dual GPU performance increase  (Read 1169 times)

DroneInspect

  • Newbie
  • *
  • Posts: 6
    • View Profile
Dual GPU performance increase
« on: February 20, 2025, 01:52:10 AM »
Hi all!

I'm currently running a machine with X470 main board, Ryzen 9, 128 GB RAM and a single RTX 3070.
I am considering to add a second GPU - maybe a 4070 or something above...

Does soemone have any experience on the performance boost that will add to my system?
Is it worth the money? is it good to have 2 different GPUs or would it be better to have 2 GPUs of the same type?

Thanks in advance!

Regards

olihar

  • Sr. Member
  • ****
  • Posts: 296
    • View Profile
Re: Dual GPU performance increase
« Reply #1 on: February 21, 2025, 03:54:04 PM »
It will almost double some parts of the processing.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15382
    • View Profile
Re: Dual GPU performance increase
« Reply #2 on: February 21, 2025, 08:00:49 PM »
Hello DroneInspect,

It is not obligatory to have same GPU models in  multi-GPU setup, however, we usually recommend to have them of more or less comparable performance. So RTX 3070 and RTX 4070 would be fine, but RTX 4080 and RTX 2060 would be likely suboptimal, as higher performance GPU would regularly wait for the sub-jobs completed by lower performance one.
Best regards,
Alexey Pasumansky,
Agisoft LLC

CheeseAndJamSandwich

  • Full Member
  • ***
  • Posts: 194
    • View Profile
    • Sketchfab Models
Re: Dual GPU performance increase
« Reply #3 on: February 21, 2025, 09:35:49 PM »
Hello DroneInspect,

It is not obligatory to have same GPU models in  multi-GPU setup, however, we usually recommend to have them of more or less comparable performance. So RTX 3070 and RTX 4070 would be fine, but RTX 4080 and RTX 2060 would be likely suboptimal, as higher performance GPU would regularly wait for the sub-jobs completed by lower performance one.

Alexey,
Does MS know which GPU is 'the best'?
And could MS divide the tasks up such that they are smaller, allowing to for Any GPU, fast or slow, to just do what it can, but not cause any delays?
Same with network processing.  Does MS know which node has the fastest CPU, GPU?

As the industry is running out of performance to be had, without creating furnaces, everything is going even further towards huge numbers of cores, huge numbers of CPUs, and huge numbers of GPUs.
Can MS move even further towards these realities?

It'd be very useful if Metashape could work better with asymmetric GPUs, asymmetric nodes... No matter the performances, they all do as much as they can.  And MS chooses only the best for the stuff that is stuck single-threaded.

So it just seems that the sub-jobs could be chopped up made smaller.
My 'little' scan of our dive site, 'Manta Point'.  Mantas & divers photoshopped in for scale!
https://postimg.cc/K1sXypzs
Sketchfab Models:
https://sketchfab.com/cheeseandjamsandwich/models