Hi Having an out of memory error - we have had larger data sets in with no problem but wondering if i can get someone to have a look at the mini dump to see what the actual cause is.
its an urgent request as its a live job - any help would be greatly appreciated to nail down the route cause, memory is most likely, but cache is also very low.
Tried High, Normal, cleared cache, reset to default.
Also to note the Photosets are running off a network Drive
CPU 1950x Threadripper
64gb ram
2 x 1080ti
400gb available ssd Cache
1023 images taken with 5DSR
contents of dump
RealityCapture 1.0.3.6310 RC (c) Capturing Reality s.r.o.
Using 32 CPU cores.
Added 1023 images.
Loading Project completed in 5.326 seconds.
Processing failed: Operation aborted.
Reconstruction in High Detail aborted after 26.014 seconds.
Processing failed: The application has run out of memory.
[0x1030\0x2030\0x2000\0x3704\0x4999\0x5051]
[0x1030\0x2030\0x2000\0x3704\0x4999\0x5051]
Reconstruction in Normal Detail failed after 6126.401 seconds.
Hi jack,
I guess you re-booted in between tries? Just saying because I only do sleep mode for weeks and sometime that helps.
Also, you could try adjusting the reconstruction region a bit.
Next I would run another alignment.
If none of that helps I would start a new project from scratch.
This seems to be happening more and more lately and it’s pretty certainly not your system specs. Although I’m not quite sure why you say cache might be the issue and then say theres 400 GB available?
Hi Gotz
I am trying the Reconstruction region solution now. With a reduced data set of 512 - out of memory error dissapears and a new error occurs - Cant build X number of parts - varies from 40 to 45 parts.
I have dropped from default 5 million per part to 2 million per part.
I have started from scratch twice.
I only say that cache might be an issue as i have seen people commenting on filling up up to 1tb on a build.
We are using Jpegs
Currently the mesh now builds but i get 45 parts could not be built and then the remaining parts are visible but obviously not useable.
We have another machine running with a 1500 data set using the same settings and still get out of memory error this machine is identical but with a larger cache SSD.
After Further testing
It builds the 500 set on Normal with 5 million parts
Still waiting to see if the 1500 set will build on Normal.
I have noticed that during the Depth map cal pass it hit 70% then goes incredibly slow on the high settings - cpu around 30-40% GPU around 35 - 30%
Normal Depth map pass hits 70% and continues at a relatively steady pace.
I have seen the Mesh built from the High settings albeit only a few parts but the detail is night and day from the Downscaled Normal settings.
Hi jack,
thanks for further feedback.
That’s good news, relatively speaking!
My point was that maybe you can identify problem areas with doing smaller batches.
Sometimes RC can also patch those up if you spoon feed it, post problematic set last.