Windows Server 2008R2 x64

[matknaak]

Hi,

we tried to install Cloudcompare via Citrix in a Windows Server 2008R2 x64 environment - CC 2.7 and 2.8beta did not work, but we managed to install CC 2.6.2. However, the older version still does not seem to start. Is there a problem betwenn CC and Windows server?

Our IT gave me the following error report (hope that may help; please don't mind the German):

Name der fehlerhaften Anwendung: CloudCompare.exe, Version: 0.0.0.0, Zeitstempel: 0x571e6670
Name des fehlerhaften Moduls: Qt5Gui.dll, Version: 5.6.0.0, Zeitstempel: 0x56d82c20
Ausnahmecode: 0xc0000005
Fehleroffset: 0x000000000005c600
ID des fehlerhaften Prozesses: 0x1dd4
Startzeit der fehlerhaften Anwendung: 0x01d2520fa93cde7c
Pfad der fehlerhaften Anwendung: D:\Program Files\CloudCompare\CloudCompare.exe
Pfad des fehlerhaften Moduls: D:\Program Files\CloudCompare\Qt5Gui.dll
Berichtskennung: e808d317-be02-11e6-81c7-005056810007

cheers

mat

[daniel]

I never tried to install CC on a Windows server, and I never had any comment about this from users.

1) Does your server even have a graphic card? (if yes, what model is it?)

2) The error is strange because it implies that the Qt DLL version is 5.6 while the old 2.6.2 version was relying on Qt 4.8 ... Are you sure the files of the 2.7 version have been cleaned properly before installing the 2.6.2 version? (they should be uninstalled automatically, but as it seems you had an issue even when installing the newer versions, something bad must have happened). Simply deleted the CloudCompare folder and retry installing the 2.6.2 version

[matknaak]

Hi Daniel,
I run CC on a desktop pc with 16GB RAM and am experiencing trouble to work with pointclouds larger than 250 million points. Therefore, we tried to install CC on our server, but failed as mentioned before.
No, our server has no graphic card yet, but we are about to get one. recently I tested CC 2.7 remotely on a server with an envidia grid environment, which I believe is a graphic card on such a server - CC worked fine. The triel environment was unfortunately set up with only 16GB RAM too.
So I hope that once we manage to get the envidia grid solution plus CC on the server, I should be able to load huge pointclouds and be able to navigate in 3D without any trouble.
I will try to get our IT guys to try to re-install CC, though I prefer CC 2.7.

cheers

mat

[daniel]

First I would say that obtaining a low framerate or relatively slow algorithms with 250 million points is quite normal (especially if you have a mid-range graphic card). With recent versions, the LOD mechanism should help with the slow framerate (you'll just need to wait for the LOD mechanism to be ready, see the Console for some feedback about the construction process). Make sure also to hide normals (if you have any) because they are awfully slow to display. Of course the better the graphic card is, the faster the display will be.

And for algorithms, you can't expect them to run as fast as for 10 million points. Depending on the algorithm and the cloud structure, the efficiency can be severely downgraded (pay a great attention to the scales with such big clouds). It can be a good idea sometimes to subsample the cloud first and test the algorithm first on the downsampled version (such as to tune the parameters).

To finish, I would say that 16 Gb of memory should be more than enough for 250 million points (and up to 750 maybe). Just check the free memory once the cloud has been loaded and the octree and LOD structures have been computed.

P.S.: if your server doesn't have a graphic card, then CC can't start indeed.

[matknaak]

Hi Daniel,

thanks for your reply.

re server - at this moment our server has no graphic card, however, after a restart, CC 2.6.2 started and works (though graphic is slow). I do have the problem now that if I like to access data from my computer or different parts of the server, only parts of a point cloud dataset are recognized (I had this problem before with other servers - it is a problem not related to CC). I will now have to wait for the server graphic card before continuing in this direction.

re my desktop PC - you are right, large point clouds may result in low framerates and slow algorithms, though I am allways amazed how CC handles all this. As I mentioned, I am using Windows 7 Professional, 16GB RAM and a NVIDIA GeForce GTX 750 Ti graphic card. There should not be any problems from the hardware side to load point clouds exceeding 230 million points. However CC 2.8beta crashes before the points are finally loaded and displayed. CC 2.7.0 loads and displays the points. The last line in the Console says: VBO(s) (re)initialized for cloud 'xxx' (-641.81 Mb = 100.00% of points could be loaded). However, the moment I even hit file name in the DB Tree CC crashes with the message that the application requested more GPU memory than is available.

The data set is a text file with x, y, z, R, G, B. The coordinates seem to have a high precision with 8 decimal places, though the last 4 are just zero. Might this be causing trouble - is there a way to truncate those digits?

cheers

mat

[daniel]

No there's no reason for CC 2.8 to crash (and the number of digits has no influence).

What version of CC 2.8 beta are you using exactly? Is it the most recent one? (it changes quite often).

You can try to disable the LOD mechanism (see the display options regarding the reduction of point clouds above a given size). And if it doesn't fix your issue, you can also try to disable the usage of VBOs. I'm interested in your feedback.

[matknaak]

Hi Daniel,

I have been using a version of CC 2.8beta that I downloaded 7th October.
After changing the settings in the display options, I managed to load the point cloud and can rotate the scene.
It seems that "Try to load clouds on GPU for faster display" needs to be disabled when I load the data. Once it is loaded it can be enabled again. I am a geologist, so don't think I fully understand this, but I am happy for now. Thanks a bunch.

cheers

mat

[daniel]

Hopefully the latest beta version should fix your issue. I would be very interested to know if that's the case, or if your bug is still here.

[Dimitri]

Hi matknaak,

I'll be interested to know if you're able to run CC on remote desktop through your nvidia grid.

Regarding graphic power:
There's a post hidden somewhere in the forum where users have compared various graphics cards (quadro vs geforce). Here's the link a user (JedFrechette) set up to register the results of using a reference dataset:

Here is a link to a form where you can enter your own results:

http://goo.gl/forms/IiC8yuunx9

And here are the results for a couple of machines I happen to have in front of me today:

https://docs.google.com/spreadsheets/d/ ... sp=sharing

The gefore 750ti, though a honest graphic card, is relatively old. I've recently updated to a Geforce GTX1070 with 8 Gb memory, and it's just fantastic and not so expensive...Using the LOD feature (activated by default) is actually detrimental on these very big graphic cards (unless you have really, really massive point clouds). To check if you're reaching the max graphic card memory, you can duplicate large points clouds until CC says that it cannot load them anymore. There you'll know roughly how large can be the point cloud you can display efficiently with the graphic cards, knowing that you can still do operations on larger clouds if they are not displayed. With my previous gtx970 with 4 gb memory, I think I was able to load up to 700 millions points. I've not done the test with the new gtx1070 and also need to compute the framerate with the test data.

[matknaak]

Hi,

I am currently out of office, but will get back to you on Friday. Thams for your efforts

cheers

mat