Dense Point Cloud calculation taking suspiciously long time on AWS EC2 Instance

[farza]

Hey all!

I've been using Photoscan for a while now on my laptop which runs a NVIDIA 960 and i7. I decided to try and use an EC2 instance (p2.xlarge) with much better specs in hopes that it would result in faster render times.  I'll include a screenshot to some of the specs at the bottom of the post. My dataset features around 150 images of some toy dinosaurs of varying shapes and sizes. On my laptop, I was able to create the dense point cloud for this (after aligning the photos) within 4-5 hours. 

The issue is, this EC2 Instance takes a very similar amount of time when building the dense point cloud and I don't see why that is considering the specs of this instance far surpass my laptop. Shouldn't the calculation be much faster?


Any assistance would be fantastic as I've been banging my head against the wall for a while with this.

EDIT: I have updated the NVIDIA drivers as well, no change.

[Alexey Pasumansky]

Hello farza,

Dense cloud generation consists of two parts - depth maps generation (GPU-supported task) and depth filtering (CPU-only), so the second one depends on CPU and on a processor with low frequency it may take considerably longer time.
Do you have a log related to the dense cloud reconstruction from EC2 instance? Also do you see the utilization of all the CPU cores?

[jinjamu]

Hi Farza,
By any chance are you running the identical version of Photoscan on the EC2 machine as on your other test machine, or are you running a different version, if so, what versions?
Thanks
John

[farza]

Hello farza,

Dense cloud generation consists of two parts - depth maps generation (GPU-supported task) and depth filtering (CPU-only), so the second one depends on CPU and on a processor with low frequency it may take considerably longer time.
Do you have a log related to the dense cloud reconstruction from EC2 instance? Also do you see the utilization of all the CPU cores?

Hi there Alexey, thanks for the reply! After your reply I decided to go for the c4.2xlarge which runs Intel Xeon E5-2666 v3 and has a vCPU count of 8. These are instances focused more on CPU Compute and don't have GPUs. During the calculation of the dense point cloud, the CPU was being fully used. When using this instance, the depth-map generation took very long while the generation of the point cloud was quicker then the instance used in my original post.

I guess this sort of helps alleviate some of my confusion! But, what instance would work best for Photoscan? Trying out every instance is a bit tough, but if if you have knowledge on it and could take a look at the specs for different EC2 instances that'd be awesome!

Any other tips to speed things up would be helpful as well, thanks

Hi Farza,
By any chance are you running the identical version of Photoscan on the EC2 machine as on your other test machine, or are you running a different version, if so, what versions?
Thanks
John

Yup, everything is almost the same! The main difference being my laptop runs Win. 10 while the EC2 instance uses Windows Datacenter 2016.