Hi Chirs et. al. On Wed, 3 Apr 2019 at 13:08, Chris Hanson <xe...@alphapixel.com> wrote: > He's GPU side. I tech-reviewed for the OpenGL SuperBible that he was one of > the authors of. > > He might still be able to advise us on troubleshooting if we can't pin it > down. He's been very helpful before.
Thanks for the info. For now I've got the new system working well enough that I can get on and use it as my main dev system. With the changes of the set up today I've got the VSG performing inline with my expectations. On the OSG side, ouch it's struggling far more than I expected. I've tried different CPU affinity tests using taskset and different compilers and AMD zen compile options and haven't see the gap close between the Intel and AMD. From the CPU scene graph tests I've done it really looks like the OSG heavy-weight node approach shows a far bigger penalty on AMD vs Intel. Credit to Intel for having a CPU architecture that can handle such an awkward beast. AMD clearly haven't optimized the CPU architecture for software like the OSG. As may "day job" is now pushing on with VSG and OSG is in maintenance mode I am inclined to take note of this finding on the OSG side, but to not spend further time trying to find improvements for the OSG on AMD. For users hitting up against CPU scene graph bottlenecks, especially on AMD, then I would suggest that VSG will be the most effective way to resolve them. However, the VSG is still a long way from being ready to be used in serious graphics application, it's under heavy development and large API changes each month. Another 6 months things will become more viable on the VSG side as we close in on 1.0, it'll still be quite a different API to the OSG so not a straight forward drop in replacement. At this point knowing that the VSG is going to be available and will radically improve performance should make it something to watch and plan for, especially for new projects. Cheers, Robert. _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org