Hi everyone,
Originally I used Photoscan, but because of the much better quality I have changed some days ago to CR.
If I reconstruct the model in normal quality, everything goes fine and requires about 4 hours to process, but when I try to reconstruct it in high quality, the calculation time increases extremely after the RAM is full.
I use pagefiles on a 250GB NVMe SSD (Samsung’s 960 EVO, which is pretty fast, I think). The cache of CR is also on this SSD.
The model is divided into 83 parts. After approximately 70 parts were calculated, which needed about 8 hours, the RAM is completely full and the SSD shows 100% activity in the resource monitor, but only a small (3200MB/s would be possible) data I/O of up to 500MB/s. The CPU has almost nothing to do, and the process progress of 81% does not increase over 10 hours.
My question:
Do I have too little RAM?
If so, what can I do to get the model in high quality without having to buy additional RAM?
Is it normal that the computing time is so much higher?
My specs: 450 Pics with 20 megapixels shot with my Phantom 4 Pro in RAW (DNG-file) altogether about 17GB.
Computer: i7 6800k@4,3 GHz, GTX 1060 3GB, 32GB RAM, Windows 7 Professional,
250GB NVMe SSD Samsung 960 EVO for cache and pagefile, a 500GB harddrive for the system. The pagefile is on the SSD only!
Reconstruction time for high quality can take several times longer than normal quality.
What slows it down is the constant swapping since the RAM is full.
Even the fastest SSD will not be fast enough once that happens.
So, yes. More RAM will definitely help in this case.
Patience will reward you too 
I found the problem.
It was really noobish :? … I did not know that there was a difference between reconstruction region and clipping box.
So it has reconstructed the whole area, which is, since I have made aerial shoots, a few square kilometers large. :roll:
TimN wrote:
Hi everyone,
Originally I used Photoscan, but because of the much better quality I have changed some days ago to CR.
If I reconstruct the model in normal quality, everything goes fine and requires about 4 hours to process, but when I try to reconstruct it in high quality, the calculation time increases extremely after the RAM is full.
I use pagefiles on a 250GB NVMe SSD (Samsung’s 960 EVO, which is pretty fast, I think). The cache of CR is also on this SSD.
The model is divided into 83 parts. After approximately 70 parts were calculated, which needed about 8 hours, the RAM is completely full and the SSD shows 100% activity in the resource monitor, but only a small (3200MB/s would be possible) data I/O of up to 500MB/s. The CPU has almost nothing to do, and the process progress of 81% does not increase over 10 hours.
My question:
Do I have too little RAM?
If so, what can I do to get the model in high quality without having to buy additional RAM?
Is it normal that the computing time is so much higher?
My specs: 450 Pics with 20 megapixels shot with my Phantom 4 Pro in RAW (DNG-file) altogether about 17GB.
Computer: i7 6800k@4,3 GHz, GTX 1060 3GB, 32GB RAM, Windows 7 Professional,
250GB NVMe SSD Samsung 960 EVO for cache and pagefile, a 500GB harddrive for the system. The pagefile is on the SSD only!
How long is the computation time?
Hi Tim,
there are a few posts with the same problem.
I am sure you’ll find them. 
Bottom line is that RAM should not be filled and virtual memory not be used.
It does that sometimes but really should not.
If you realize that it’s doing it, you can abort and start fresh.
Thanks to the cache, it won’t take very long to catch up - the loss usually outweighs the time it takes to finish in “turtle-mode” easily, if you catch it early on.
So if I hear that right, if I wanted fast processing, I should avoid using pagefiles?
I use cache, but as it seems, all depth maps are stored, but only the first two to three computed parts. If I abort after 90%, and then start again, then the first three parts come immediately, but the rest does not. Is that normal?
I’m not too firm on the details.
But I did read repeatedly that RC is supposed to only use “real” RAM.
It has methods to cut everything up into small enough chunks for any setup (within limits).
What do you mean by computed parts?
I remember when it happened to me, I canceled it, then started again.
First it went quick for a tiny bit, then seemed to take for ages.
But eventually, it went through in a fraction of the time.
Hope that helps!
BTW have you tried in normal mode with downsampling at 2?
I find that that is very often quite enough if you have high enough resolution and its A LOT quicker!