Hi Ronan,
Carsten Neumann wrote:
> Hello Ronan,
>
> ronan wrote:
>> Hi,
>>
>> I am trying to figure out the most efficient way of drawing some lines
>> to represent vectors. The issue i have is that i have got alot of
>> vectors to be drawn (approx. 8.5 million). Current i have tried using a
>> geometry core attached to a node with a certain number of lines
>> associated with each geometry. I have tried using the simple scene
>> manager and the show statistics i get the following frame rates
>>
>> All off one node - 1.4 fps
>> Split into 128 nodes - 2.65 fps
>> Split into 512 nodes - 1.64 fps
>> Split into 2048 nodes - 2.26 fps
>> Split into 8192 nodes - 1.12 fps
>> Split into 32768 node - 1.42 fps
>>
>> I am running these tests on a dual core 3GHz AMD 64 with a Nvidia Quatro
>> 3500 and OpenSG-1.8. So my question is what can i do to improve the
>> performance ? change graph structure, culling ???? i would have thought
>> that the performance would improve as you spread the geometries over a
>> number of nodes ?
>
> well, that depends on whether they are all on screen or not. If they
> are, more nodes actually hurt, as that means more nodes to traverse, if
> some nodes are off screen more nodes can help, if they are arranged such
> that frustum culling can remove them, i.e. the nodes should be in some
> hierarchical spacial grouping.
That is a core question. I'm assuming right now all of them are in the middle
of
the screen, but is that the application scenario, too, or are you going to be
inside the data?
> One other thing you might try is using a different type of primitive to
> draw your vectors, i.e. use thin polygons instead of lines. Since
> hardware and drivers are likely to be more optimized for poly drawing
> than for lines this might give you a bit better performance.
That's true for game cards, but given that he's running a Quadro that shouldn't
be a problem.
One other question I have is: are you using a single GL_LINES primitive or many
GL_LINE_STRIPS resp. GL_LINES? I recommend using as few as possible.
If you have a simple test program you could send it to us to see if there is
something else obvious that we're missing.
Yours
Dirk
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Opensg-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensg-users
