Hi J.P. I think one can get more thatn 1GB/s. Of course it strongly depends on used system memory, motherboard and CPU. There was discussion on GPGPU forum (http://www.gpgpu.org/forums/viewtopic.php?t=4798). And it seems that it is also a matter of how to write the correct benchmark. So, maybe you have to check if your benchmark code is fair enough to be measured.
cheers, art J.P. Delport wrote: > Hi Art, > > do you know what the fastest for a single channel transfer is? Or should > I make a shader that dumps single channel data into BGRA? Or can I cheat > and render to a luminance and transfer as BGRA? I'm getting +-1GB/s for > readpixels and luminance, so maybe I should stop trying to get it faster. > > I'm curious what speeds other people are getting. NVidia claims that > with certain HW (NVidia M/B) one can get up to 3GB/s with "pinned" memory. > > jp > > Art Tevs wrote: > > > Hi J.P. , Robert, > > > > as far as I remember PBO is indeed slower for formats not natively > > supported by the VRAM. On nVidia GPU's it seems that the PBO transfer of > > GL_BGRA is imho the fastest one. Yes, it is even faster than GL_RGBA. > > I think that has something todo with alignment of the data in the memory. > > If I am wrong, then please correct me. > > > > cheers, > > art > > > > > > J.P. Delport wrote: > > > > > Hi Robert, > > > > > > Robert Osfield wrote: > > > > > > > > > > Hi J.P, > > > > > > > > I found that the GL driver support for properly accelerating the > > > > reading from the GPU memory using PBO to be very sensitive for pixel > > > > format, and if you fell off the past path it would indeed be slower. > > > > It's a while since I implemented osgscreencapture so can't remember > > > > all the details off hand, but have a look into forcing the use of the > > > > RGBA frame buffer. > > > > > > > > > > > OK, I'll have a look at the format. I was actually triggered by one of > > > our own apps that reads GL_LUMINANCE textures back (at half the speed > > > with pbo than without) and then saw osgscreencapture exhibit the same > > > trend. For some reason I thought I remembered osgscreencapture being > > > faster with pbo (dual pbo is the default e.g.), but maybe my memory is > > > wrong. > > > > > > rgds > > > jp > > > > > > > > > > > > > Robert. > > > > > > > > On Mon, Dec 7, 2009 at 9:39 AM, J.P. Delport <> wrote: > > > > > > > > > > > > > Hi all, > > > > > > > > > > running with latest svn OSG, the osgscreencapture example produces > > > > > the best > > > > > speed for me (tested on a few machines) when run with the --no-pbo > > > > > command > > > > > line parameter. All the pbo options (--single-pbo to --triple-pbo) are > > > > > slower. Is this expected? I've thought that the pbo versions would be > > > > > faster? > > > > > > > > > > regards > > > > > jp > > > > > > > > > > > > > > > > > -- > > > This message is subject to the CSIR's copyright terms and conditions, > > > e-mail legal notice, and implemented Open Document Format (ODF) standard. > > > The full disclaimer details can be found at > > > http://www.csir.co.za/disclaimer.html. > > > > > > This message has been scanned for viruses and dangerous content by > > > MailScanner, > > > and is believed to be clean. MailScanner thanks Transtec Computers for > > > their support. > > > > > > _______________________________________________ > > > osg-users mailing list > > > > > > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > > > > > ------------------ > > > Post generated by Mail2Forum > > > > > > > > > ------------------ > > Read this topic online here: > > http://forum.openscenegraph.org/viewtopic.php?p=21012#21012 > > > > > > > > > > > > _______________________________________________ > > osg-users mailing list > > > > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > > > -- > This message is subject to the CSIR's copyright terms and conditions, e-mail > legal notice, and implemented Open Document Format (ODF) standard. > The full disclaimer details can be found at > http://www.csir.co.za/disclaimer.html. > > This message has been scanned for viruses and dangerous content by > MailScanner, > and is believed to be clean. MailScanner thanks Transtec Computers for their > support. > > _______________________________________________ > osg-users mailing list > > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > ------------------ > Post generated by Mail2Forum ------------------ Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=21017#21017 _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
