Agisoft Metashape
Agisoft Metashape => General => Topic started by: mkrech on November 29, 2017, 11:55:29 PM
-
Is there a good solution for a GPU optimized cloud based virtual workstation that would process large photo sets quickly?
Also, does the machine need to be OpenCL based or can CUDA machines work?
Thanks
-
I've processed 11,000 photos as one chunk using MS Azure NC24 (24 vCPUs, 224 GB RAM, 4 Nvidia K80 cards). They also have the NC24 v2 now. They can be Windows or Linux.
https://azure.microsoft.com/en-us/pricing/details/virtual-machines/windows/
-
how long did it take to calculate, and how much did you pay? Where can I find the details about the GPU?
-
There's some info here, but a google search will find you much more :)
https://azure.microsoft.com/en-us/blog/azure-n-series-general-availability-on-december-1/
I have used the NV machines extensively. Same performance as the NC machines, not sure the actual technical differences. Price varies depending on the size of VM you select from around $1 to $5 per hour.
They work great to process models, but not so great if you need to edit on screen, as you'll need a really good bandwidth to have a decent user experience.
Just remember to switch off when you are done, otherwise you keep paying.
I think you can open a trial account and get $70 free usage, to try it out.
-
Hi everyone !
Really interesting topic !
But I have some questions about.
Firstly, how does it work for the software ? Do we need to have PhotoScan installed on the virtual machine ?
Regards
-
On MS Azure NC24
Project 1:
Agisoft v.1.3.2
8500 acres (34.5 sq km)
Photos: 10622 (Sony QX1)
Photo alignment (high; 40,000; 4,000)
-matching time: 2h 2m
-alignment time: 45m
Optimization: 4m
Depth maps: 7h 48m
Dense Cloud: 5h 17m (medium, aggresive)
Model: 3m
DEM: 48m
Ortho: 2h 59m
Project 2:
Agisoft v.1.3.2
7880 acres (31.9 sq km)
Photos: 10409 (Sony QX1)
Photo alignment (highest; 40,000; 4,000)
-matching time: 3h 54m
-alignment time: 46m
Optimization: 1m
Depth maps: 8h 10m
Dense Cloud: 5h 13m (medium, moderate)
Model: 2m
DEM: 42m
Ortho: 1h 55m
-
Firstly, how does it work for the software ? Do we need to have PhotoScan installed on the virtual machine ?
Yes
-
To expand on what jinjamu pointed out, simply "powering off" in Windows does not stop the clock. You have to "Stop" the VM via the Azure portal to stop charges from being incurred. I don't know if AWS is the same or not.
-
Hi TXPE !
Firstly, how does it work for the software ? Do we need to have PhotoScan installed on the virtual machine ?
Yes
Thanks for your detailed answer, and the processing time !
Is it easy to install any software on virtual machine ? I mean there is no restriction for that ?
Regards
-
I'm running on a Windows server OS. It is like running any other Windows OS. You will connect via Remote Desktop, VNC, TeamView, etc. You will also have to download and install the Nvidia drivers. You can also set up Linux VMs.
https://azure.microsoft.com/en-us/pricing/details/virtual-machines/windows/
You will have to move your Photoscan license to the VM.
I think the bigger hurdle was figuring out the storage. I think MS finally changed their website, it now reads as 1.44TB as "Temporary Storage". This is mapped as the D: drive. The data on the D drive can be lost. I lost the data on the D drive when converting from an NC24 to an NC12 once. The C: drive should be safe and is where I store the active project I'm working on. Within Azure, you will want to set up a storage account. I set up a "files" storage. This is mappable. It is basically a NAS. I have this storage space mapped on the VM and my personal PC. Use this storage to upload and download data. You can do this with the VM turned off, thus saving money. When you have all of your data/photos uploaded, then fire up the VM.
-
Hi TXPE. Thanks for sharing your experience.
I would like to know if the payment for the use of cloud based virtual machine services (per hours) requieres the extra payment for the storage of the data's proccessing.
Also, I would like to know if necessary to pay for a base cost to be able use this service.
Best regards.
-
For the NC24 VM, there is no minimum monthly fee. Only pay for time used.
The storage acct does cost based on how much data you have saved. 110GB is running about $6.00 USD/month.
-
Hi. Thank for your answer.
In the case of Agisoft licensing, there is limit by the number of times that the license is moved?.
I apologies for as many questions.
-
Good question. Would have to ask Agisoft.
-
The Free Account of Azur has a limitation of 4vCPU's per machine.
From my point of view the Free Account is useless to test high performance computations.
You have to upgrade to a "Normal" Account, but you cannot test for free.
Kind regards
Alain
-
In the case of Agisoft licensing, there is limit by the number of times that the license is moved?.
There's no limit for the license transferring. So you can deactivate/activate the license unlimited number of times.
-
Are you doing the 30 day trial? If I recall correctly, I called them during the 30 day trial period and they gave me access to the NC24.
-
Hey TXPE,
That looks really great. Its very huge project. Must have experience difficulty while processing. Can you share your processing experience here? How you managed to pull off such a huge project?
-
Very useful information. Thanks for sharing, TXPE :D
Regards,
SAV
I'm running on a Windows server OS. It is like running any other Windows OS. You will connect via Remote Desktop, VNC, TeamView, etc. You will also have to download and install the Nvidia drivers. You can also set up Linux VMs.
https://azure.microsoft.com/en-us/pricing/details/virtual-machines/windows/
You will have to move your Photoscan license to the VM.
I think the bigger hurdle was figuring out the storage. I think MS finally changed their website, it now reads as 1.44TB as "Temporary Storage". This is mapped as the D: drive. The data on the D drive can be lost. I lost the data on the D drive when converting from an NC24 to an NC12 once. The C: drive should be safe and is where I store the active project I'm working on. Within Azure, you will want to set up a storage account. I set up a "files" storage. This is mappable. It is basically a NAS. I have this storage space mapped on the VM and my personal PC. Use this storage to upload and download data. You can do this with the VM turned off, thus saving money. When you have all of your data/photos uploaded, then fire up the VM.
-
Are you doing the 30 day trial? If I recall correctly, I called them during the 30 day trial period and they gave me access to the NC24.
Yes, I'm using the 30 day trial and I asked the Support if I can use for example the NC24. They said that they have limited the trial account to the "low" performance machines.
-
@TXPE
How did you manage to upload such a large amount of data to the Virtual Machine ?
I have about 11 GB to upload and I guess this takes longer than computing the Project on my local machine ;-)
-
That price seems cheaper than AWS with a gpu vm. They now have elastic gpu but I think it's still expensive.
And, yes, you have to turn off the AWS machine to stop the clock for billing . Some vms like Lightsail or Digital Ocean are a set price per month whether on or off, but these are usage based.
-
That price seems cheaper than AWS with a gpu vm.
"That" what ?
What means AWS ? Azur Workstation ?
-
There is also Paperspace (https://www.paperspace.com/&R=UJX7WI), online VM service. Offers a Win7/10 operating system and a Quadro P5000 graphic, 8 CPU, 30 GB Ram for 0.65 US$/h or a Quadro P6000 for 0.90 US$/h. Their bandwidth is 1GB/s and you can add storage up to 2 TB. Access easily by webbrowser or desktop app.
-
@TXPE
How did you manage to upload such a large amount of data to the Virtual Machine ?
I have about 11 GB to upload and I guess this takes longer than computing the Project on my local machine ;-)
It is slow, as MS will throttle the connection. It took about 5 hrs to upload 41GB (4256 photos). It should have taken less than an 1 hr. This is one of my biggest complaints with this approach. I am currently communicating w/ MS now on how I can increase the bandwidth and quotas, even if I have to pay for it.
-
And how long did it take to compute the dense cloud and what quality did you choose and what hardware did you choose and what did all in all cost ? :-)