Hi Robert,

I'm also having a problem with the massive performance drop with
multiple (in my case > 8) HDR RTT camera nodes .  I am using a Quadro FX
5500 G with 1 GB of memory, so I know it isn't my graphics card memory
(the same problem happens on my GeForce Go 7800).  I'm also running
under Linux with 2 dual core opeterons.  My RTT sizes are 512x512 with
nearest/nearest mipmapping and I'm using 32 bit HDR rendering.  

I have made some simple modifications to osgprerender.cpp (see attached)
to illustrate the problem I and others appear to have seen.  I also hard
coded the values I'm using in osgprerender and set up the proper 512x512
HDR settings for both texture 2D's and texture rectangles.   8 RTT
camera nodes run great with a small performance drop for each camera as
would be expected.  When I add the 9th camera node that is when things
break.  All that you need to do to run it is pass in a model and if you
want texture rectangle use --texture-rectangle.

Is this there a frame buffer object limitation of the driver/card?  I
thought the number of FBO's was only limited by the amount of texture
memory available on the card, but I'm using a card with 1GB of memory
and not overly large textures.  

Thanks,
Dennis

>> From: Robert Osfield <robert.osfield_at_gmail.com> 
>> Date: Tue Jan 23 2007 - 00:50:39 PST
>> 
>> Hi Brian, 
>> 
>> I'd check to see if you graphics card has sufficient memory for what 
>> you are trying to do 
>> 
>> Robert. 
>> 
>> On 22 Jan 2007 23:27:35 -0000, [EMAIL PROTECTED]
<[EMAIL PROTECTED]> wrote: 
>>> Hi, 
>>> 
>>> I'm using floating point textures as my targets for multipass RTT.
In my 
>>> tests I have noticed that if three or less camera nodes are set up
using 
>>> GL_RGBA16F_ARB internal format textures, the frame rate stays at a
constant 
>>> 75FPS when only the texture is rendered to the screen (i.e. the
texture is 
>>> rendered over the full window). 
>>> 
>>> If I add an additional camera node, the frame rate drops to 15 FPS.
Five 
>>> nodes yields 8FPS, and 5FPS are achieved with six nodes. Also, if I
set 
>>> the texture to render to 1/4 of the screen and position the main
camera so 
>>> that the actual scene being rendered to texture is visible, I'll get
25FPS 
>>> for the three camera node implementation. 
>>> 
>>> I'm setting up my textures with internal formats of GL_RGBA16F_ARB,
a 
>>> source type of GL_FLOAT, and a source format of GL_RGBA. 
>>> 
>>> The FBO is set up like: 
>>> fbo->setAttachment(GL_COLOR_ATTACHMENT0_EXT, 
>>> osg::FrameBufferAttachment(texture)); 
>>> 
>>> fbo->setAttachment(GL_DEPTH_ATTACHMENT_EXT, 
>>> osg::FrameBufferAttachment(new osg::RenderBuffer(512, 512, 
>>> GL_DEPTH_COMPONENT24))); 
>>> 
>>> My min/max filters are set up as NEAREST. 
>>> 
>>> If I change my textures to internal formats of GL_RGBA and the
source type 
>>> as GL_UNSIGNED_BYTE, my performance is great. 
>>> 
>>> Any thoughts? 
>>> 
>>> Thanks, 
>>> Brian 
>>> _______________________________________________ 
>>> osg-users mailing list 
>>> [email protected] 
>>> http://openscenegraph.net/mailman/listinfo/osg-users 
>>> http://www.openscenegraph.org/ 
>>>
>

Attachment: osgprerender.cpp
Description: osgprerender.cpp

_______________________________________________
osg-users mailing list
[email protected]
http://openscenegraph.net/mailman/listinfo/osg-users
http://www.openscenegraph.org/

Reply via email to