Hello,
At my office for the next 3 years we are planning to make the textured photo realistic model of about 30 monuments through photography (drone and ground) and 3D laser scanner Leica P40, here in Greece and we want to finally buy CaptureReality.
Here’s the deal.
We are set up a server to do the heavy computation in captureReality and vray render in max. We have a budget for the server, ranging from 15000 to 20000 euros.
I would like the community and the admins to suggest me the most optimized server rig for its purpose and budget range. Don’t include peripherals. Just the server rig. e.g. Should we wait for AMD Naples? Should we use game-oriented cards like 1080TI (3 stack) or go for professional P6000 and a graphics accelerator? How much RAM is enough? All those things.
As for storage backup we’ll use a NAS. That’s considered a peripheral device.
The monuments are HUGE in scale and the .ptx and images data A LOT.
is there really any point looking at epyc for rc?
not much point having a 128 threads if it will only use 32 threads?
you’ll have to pay for 4 x license costs for one PC, and which point you might as well have 4 pc’s running it.
just keep in mind the limits per license. 32 cores/threads. 3 x gpu. and then lots of ssd’s for storage.
I’m not sure how numa nodes effect’s performance, it would be nice to see some benchmarks between intel/amd. and dual/single cpu’s etc…
gpu’s don’t need much ram, doesn’t ever go over 2gb while processing. but new view port uses about 8gb for 40mil tri’s.
1080 ti’s are probably best to go for.
If you have enough network bandwidth, You may want to consider using AWS/Azure/etc gpu instances for the heavy lifting rather than invest in so much hardware.
AWS offer their g3 series with 2xm60 gpu for ~$4/hr (g3.8xlarge on demand, less if you can go spot priced)
So basic gamer class pc in the office (fast i7+gtx1080ti, NVMe drives) for project setup and initial test runs, then throw the lot up into the cloud for high density runs. Easy to scale as required - just pick the right size for the job and go 
Jennifer
I think this would make more sense if CLI version had a floating license.
steam version easy to run anywhere. but with only 2500 photos, its not hard to get a local pc to run out high detail models in a reasonable time.
the times when I really would have liked the extra power temporarily is doing large models with cli.
I think my internet speed would make it slower to use aws vs local steam.
but then i guess it could make sense if you have fast internet.