 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | Rules | You are not logged in.
hmm a little bit disapointing. no change in new 2.5 B1.
extrem example: adding 4410 images to a new group (294 images a 15 brackets.)
than CPU monitor of windows shows that one CPU is working 100% and 7 CPUs are empty on 0% load.
result expecting 8 times of waiting...
Georg
Offline
currently running detection of the group containing 4410 images.
SIFT controlpoint extraction: 8 CPUs used 100%load
Coordination of controlpoints 4-8 CPU used up to 100% load - windows is partly freezing some minutes but get reanimated again, up to 8gig of 8gig ram used.
phase is ongoing since one hour currently, preditction on group status info says 2h 34m...
Georg
Offline
thanks for sharing this georg
its a bit odd, how are your hard drives configured? maybe we do need more memory :-) I max out 8GB too.
Henrik
Offline
I've sadly on my DELL PC on my C:\ windows system partition installed an (software) raid 1 of two 1GB diskc.
after 10 hours in my test the adjustment of controllpoints phase is still ongoing. 7.6GB mem used and 16GB mem subscribed by autopano... so clear that this swapping causes delays I expect...
CPU usage is 5-10% currently in average with 25% peaks at 4 of the 8 CPU.
currently status in my test ... screenshot from 9am...
Georg
Offline
gkaefer wrote:
currently running detection of the group containing 4410 images.
SIFT controlpoint extraction: 8 CPUs used 100%load
Coordination of controlpoints 4-8 CPU used up to 100% load - windows is partly freezing some minutes but get reanimated again, up to 8gig of 8gig ram used.
phase is ongoing since one hour currently, preditction on group status info says 2h 34m...
Georg
That time - 2h 34m - is the time elapsed between shooting the first and last shots in the image set and not an indication of predicted time to detect the pano. You'll note it hasn't chnaged in all those many hours your system has been working on this image set.
Last edited by mediavets (2010-09-21 10:18:06)
Offline
mediavets wrote:
gkaefer wrote:
currently running detection of the group containing 4410 images.
SIFT controlpoint extraction: 8 CPUs used 100%load
Coordination of controlpoints 4-8 CPU used up to 100% load - windows is partly freezing some minutes but get reanimated again, up to 8gig of 8gig ram used.
phase is ongoing since one hour currently, preditction on group status info says 2h 34m...
GeorgThat time - 2h 34m - is the time elapsed between shooting the first and last shots in the image set and not an indication of predicted time to detect the pano. You'll note it hasn't chnaged in all those many hours your system has been working on this image set.
thanks for enlightment ;-) yes it did not change ;-))
but on other side it took exact 5:30 hours to shoot 882 images live on location - I had an massive sunburn because I forgot the suncream....
perhaps the 2:34 hours is the time were I did take the 884 images to produce fake AEB images (882*5)
Liebe Gruesse,
Georg
Offline
and finally preview of group got finished after 10-18hours (was not at home...)
so 900.000 controllpoints and around 20000 links between images for the 4410 images.
opening the group editor takes 5 minutes...
zooming from default 0.8% to 100% and finishing textzure reloading takes 3 minutes each move to right/left or down/up takes 1 minute.
so great to see that all this is working without bluescreen or bugsplash - perfect work done!
and also clear such amount of images for one pano your need far more memory than the 8gig ram I've, I would suggest the double of memory if you sum up the number of images * imagesize so 46gig RAM or more I would say and some big SSD for temp dirs and I would also install the APG itself offsite the windows partition on a SSD and additional the windows swap also on extra SSD...
Liebe Gruesse,
Georg
Offline
mediavets wrote:
gkaefer wrote:
currently running detection of the group containing 4410 images.
SIFT controlpoint extraction: 8 CPUs used 100%load
Coordination of controlpoints 4-8 CPU used up to 100% load - windows is partly freezing some minutes but get reanimated again, up to 8gig of 8gig ram used.
phase is ongoing since one hour currently, preditction on group status info says 2h 34m...
GeorgThat time - 2h 34m - is the time elapsed between shooting the first and last shots in the image set and not an indication of predicted time to detect the pano. You'll note it hasn't chnaged in all those many hours your system has been working on this image set.
Hi Andrew, you really know how take the wind out a man's sails :-) here was Georg hoping! now he will be doing so for a whole lot longer :-)
thanks for clearing this up
Henrik
PS: It would be nice to have an ETA (at least as a guide of what to expect) (ETA estimated Time of Arrival - so make that ETTF, Estimated Time To Finish) :-)
Offline
Hi Georg et al...
When you look at those numbers, you have to admit its a pretty amazing program, though I have my own difficulties and often switch to PTGui for help, I still can't get over how well this program sometimes comes out and totally surprise me.
Interesting part here, that you break all this down.
QUESTION!! Alexandre and Kolor, in order to speed the process up, how can we improve on the hardware to make this go faster, I would think and correct me if I am wrong, that that Viewing of the image is GPU related and if it is would multiple GPU's help here?
In terms of rendering, it comes down to hard drive speed, so a really large (large = many disks) array RAID-0 would make a difference.
can anyone confirm this, please
Henrik
PS: I find the waiting business, such a waste of my time
gkaefer wrote:
and finally preview of group got finished after 10-18hours (was not at home...)
so 900.000 controllpoints and around 20000 links between images for the 4410 images.
opening the group editor takes 5 minutes...
zooming from default 0.8% to 100% and finishing textzure reloading takes 3 minutes each move to right/left or down/up takes 1 minute.
so great to see that all this is working without bluescreen or bugsplash - perfect work done!
and also clear such amount of images for one pano your need far more memory than the 8gig ram I've, I would suggest the double of memory if you sum up the number of images * imagesize so 46gig RAM or more I would say and some big SSD for temp dirs and I would also install the APG itself offsite the windows partition on a SSD and additional the windows swap also on extra SSD...
Liebe Gruesse,
Georg
Offline
To gkaefer, if you have a 15 brackets, just use the stack feature, it's here to help you in such big case and it really improves the results.
Offline
For detection, RAM is needed as well as fast CPU. Everything else won't really help, we don't do much I/O during detection.
Offline
Ok detection is RAM and CPU bound,
viewing is GPU bound
Now rendering is ???
Henrik
Offline
Rendering is CPU bound ( if enough RAM ), or harddrive IO speed limited if big panorama.
Offline
Thanks Alexandre
OK, so it is taxing on all cylinders, if one is doing big pano's.
What would be interesting is when it becomes I/O limited vs quantity of ram/ CPU speed.
I am still a few weeks away from my monster build, which hopefully will eat multi-gigabyte images for breakfast :-) without breaking out in a sweat
thanks
Henrik
Offline
tived wrote:
What would be interesting is when it becomes I/O limited vs quantity of ram/ CPU speed.
Henrik
It seems no-one knows yet:
http://www.autopano.net/forum/p69973-20 … -57#p69973
Offline
thanks for that Andrew,
I have been away for a week and a bit, shooting pano's on a salt lake.
Henrik
Offline
I agree. This kind of relation could be measured with the release version, but we are still optimizing this now. Between beta 1 and beta 2, you'll get another boost in performance and lower IO needs.
Offline
