Re: [osg-users] Performace issue with nVidia 400 series
mrohn wrote: Hi, now we discovered that the fps drop down is related to the two sided lightning. If we disable the support for the two sided lightning we have the expected performance. But this is not a really good solution for us. Is there any assumption why this happens? ... Thank you! Cheers, Michael Greetings: Well, I checked the code and unless it is changed someware else by defalt the setting for dubble sided lighting in LightModel.ccp is set to false. Any arguments to the contrary? D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39669#39669 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Hi, now we discovered that the fps drop down is related to the two sided lightning. If we disable the support for the two sided lightning we have the expected performance. But this is not a really good solution for us. Is there any assumption why this happens? ... Thank you! Cheers, Michael -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39628#39628 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
dglenn wrote: Jason Daly wrote: On 05/18/2011 05:14 AM, Fred Smith wrote: Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. I think it's specifically GPU-host memory transfers that are slow. The other direction is OK, as is the computation speed. The thread on the OpenGL forums has the details, and someone posted a small program that does timing tests and illustrates the issue pretty well. http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflatNumber=284065page=1 --J I looked at the link you gave and downloaded the program that was the (the 4th revision of it) and tried on my GTX 460 and an old 8800 GT and compared it to the results that was posted and something does not add up with the test program! Most people there compare GTX2xx hardware and Fermi hardware. You are comparing G80 and Fermi hardware. -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39527#39527 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Hi, David, 4xx series slow specifically on readback from framebuffer\FBO to texture\PBO\host memory (in GL, dunno about cuda\opencl). That is what all people talking about. If you dont use this features much you will have no problems whatsoever. Cheers, Sergey. 19.05.2011, 03:27, David Glenn david.e.glenn@navy.mil: Jason Daly wrote: On 05/18/2011 05:14 AM, Fred Smith wrote: Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. I think it's specifically GPU-host memory transfers that are slow. The other direction is OK, as is the computation speed. The thread on the OpenGL forums has the details, and someone posted a small program that does timing tests and illustrates the issue pretty well. http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflatNumber=284065page=1 --J I looked at the link you gave and downloaded the program that was the (the 4th revision of it) and tried on my GTX 460 and an old 8800 GT and compared it to the results that was posted and something does not add up with the test program! Before I learned of all of this, I ran some other tests using a stress program and gDEBugger and found that the GTX 460 was much faster than 8800 GT (about 10 times faster under stress). I also tested it on most of my high end terrain and it ran much faster (about the biggest texture based thing I have) even though I don't have figures for it yet. I'm not saying that there is not a problem, but I'm a bit confused when one test I do contadicted by another and that the numbers for that other test and the conclusion seem spect to me! D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39519#39519 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
hybr wrote: Hi, David, 4xx series slow specifically on readback from framebuffer\FBO to texture\PBO\host memory (in GL, dunno about cuda\opencl). That is what all people talking about. If you dont use this features much you will have no problems whatsoever. Cheers, Sergey. 19.05.2011, 03:27, David Glenn : Jason Daly wrote: On 05/18/2011 05:14 AM, Fred Smith wrote: Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. I think it's specifically GPU-host memory transfers that are slow. The other direction is OK, as is the computation speed. The thread on the OpenGL forums has the details, and someone posted a small program that does timing tests and illustrates the issue pretty well. http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflatNumber=284065page=1 --J I looked at the link you gave and downloaded the program that was the (the 4th revision of it) and tried on my GTX 460 and an old 8800 GT and compared it to the results that was posted and something does not add up with the test program! Before I learned of all of this, I ran some other tests using a stress program and gDEBugger and found that the GTX 460 was much faster than 8800 GT (about 10 times faster under stress). I also tested it on most of my high end terrain and it ran much faster (about the biggest texture based thing I have) even though I don't have figures for it yet. I'm not saying that there is not a problem, but I'm a bit confused when one test I do contadicted by another and that the numbers for that other test and the conclusion seem spect to me! D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39519#39519 ___ osg-users mailing list http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org -- Post generated by Mail2Forum Well I'm using a far chunk of terrain that is very texture based and I haven't seen any issues. Maybe I'm making a mistake getting the 460's and sould try the GTX 280. D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39561#39561 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
On 05/19/2011 12:07 PM, David Glenn wrote Well I'm using a far chunk of terrain that is very texture based and I haven't seen any issues. Maybe I'm making a mistake getting the 460's and sould try the GTX 280. By texture-based do you mean you're reading the texture image from the GPU back into your application, or you're just using the texture data on the GPU for rendering. Unless you're reading the texture data back from the GPU (glReadPixels(), glGetTextureImage(), PBO in READ mode), you won't see the problem. --J ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Jason Daly wrote: On 05/19/2011 12:07 PM, David Glenn wrote Well I'm using a far chunk of terrain that is very texture based and I haven't seen any issues. Maybe I'm making a mistake getting the 460's and sould try the GTX 280. By texture-based do you mean you're reading the texture image from the GPU back into your application, or you're just using the texture data on the GPU for rendering. Unless you're reading the texture data back from the GPU (glReadPixels(), glGetTextureImage(), PBO in READ mode), you won't see the problem. --J ___ osg-users mailing list http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org -- Post generated by Mail2Forum Not unless OSG is doing that at some point! I'm not adding any drawables or changing much of the OSG code at this point. At least not yet! Grin! The only other stuff that I found that GTX460 have some issues with some of the AMD base motherboard stuff that can jam performance of the card, to the point that some textures might not appear in some OGL based games – exp: Mafia 2. But you might have already heard of that! D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39571#39571 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. Cheers, Fred -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39496#39496 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
On 05/18/2011 05:14 AM, Fred Smith wrote: Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. I think it's specifically GPU-host memory transfers that are slow. The other direction is OK, as is the computation speed. The thread on the OpenGL forums has the details, and someone posted a small program that does timing tests and illustrates the issue pretty well. http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflatNumber=284065page=1 --J ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Greetings! Well, given what I've been told by NVIDIA, that there is a Quadro group that if you develop with that card series that they will be at my bacon call and there is the GForce Group that lives in the nether regions. Getting help for Quadro is easy since it built only by PNY and they work closely with them. Getting help for GForce is a pain as they refer you to the other companies that usually don't want to deal with you! Never mind the fact that I represent a navy lab that is buying 200 of their cards, you’re still a second class citizen to them. EVGA (for example), I tried to get help configuring there card for Linux Box and I got you have to go and check your bios on your EVGA motherboard to fix the problem. I replied that I had a graphics card and not a motherboard and get the same response. I guess that this is the punishment that I get for not going with Quadro! ... D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39509#39509 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
Jason Daly wrote: On 05/18/2011 05:14 AM, Fred Smith wrote: Hi, The only problem that I personally heard about is about the slow transfer speed to/from GPU memory on Fermi cards. DMA transfers are said to be slower than on GTX 2xx hardware. I haven't seen anything related to computing. I think it's specifically GPU-host memory transfers that are slow. The other direction is OK, as is the computation speed. The thread on the OpenGL forums has the details, and someone posted a small program that does timing tests and illustrates the issue pretty well. http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflatNumber=284065page=1 --J I looked at the link you gave and downloaded the program that was the (the 4th revision of it) and tried on my GTX 460 and an old 8800 GT and compared it to the results that was posted and something does not add up with the test program! Before I learned of all of this, I ran some other tests using a stress program and gDEBugger and found that the GTX 460 was much faster than 8800 GT (about 10 times faster under stress). I also tested it on most of my high end terrain and it ran much faster (about the biggest texture based thing I have) even though I don't have figures for it yet. I'm not saying that there is not a problem, but I'm a bit confused when one test I do contadicted by another and that the numbers for that other test and the conclusion seem spect to me! D Glenn D Glenn (a.k.a David Glenn) - Moving Heaven and Earth! -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39519#39519 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
[osg-users] Performace issue with nVidia 400 series
Hi, we have a performance problem with the nVidia 400 series. But first: Our software is a 3D physics simulation. We use ODE for the physics. Qt is responsible for the 2D GUI and OSG handles the 3D visualisation. I changed my graphics-card from a nVidia 9600GT to a nVidia GTX460 (768mb of RAM). After this the performance goes down. It is roughly half of the FPS. I found this statement in the wikipedia: OpenGL Problems It has been reported by users as well as developers [13] [14] [15] [16] [17] [18] [19] [20] that nVidia 400-series cards have severe performance problems with 3D content-creation applications such as Autodesk Maya and 3ds Max, Blender, Rhinoceros 3D—as well as some OpenGL games—to the extent that video cards two generations older routinely outperform 400-series in such applications and games. The problem, which affects any OpenGL application using textures, involves accessing framebuffer contents or storing data on the GPU. So far, one customer using an OpenGL based application got a response from nVidia support indicating that the behavior is expected in the GeForce 400 line of cards, and no software update is available to improve the performance of the hardware.[21]. The problem can be worked around with a hack by using a CUDA memory copy to access the buffer object. Here is the link: http://en.wikipedia.org/wiki/GeForce_400_Series#OpenGL_Problems Is there a known issue? I searched the board but I had no success. The problem is also present on a nVidia GTX480. Some details from our software: OSG: 2.8.3 / 2.8.4 and 2.9.14 (I tested all these versions) Qt: 4.6.3 (SDK 2010.04) ODE: 0.11 build with minGW (from the Qt SDK) running under Win7 64bit nVidia driver: 270.61 Did I miss anything? ... Thank you! Cheers, Michael[/url] -- Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=39474#39474 ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Performace issue with nVidia 400 series
On 05/17/2011 11:36 AM, Michael Rohn wrote: Is there a known issue? I searched the board but I had no success. The problem is also present on a nVidia GTX480. Look at the links on the Wikipedia page (especially the link to the OpenGL forum discussion). I haven't seen any official statement on the issue from nVidia, but the conjecture on the forums seems to be that it's a deliberate throttling of performance to create a product line separation between the GeForce gamer cards and the Quadro professional cards (Quadro cards are priced many times higher than the comparable GeForce product). Traditionally, games haven't needed quick pixel readback as much as other applications (GPU-heaving computational or scientific applications), so nVidia seems to be making it a professional feature and charging a premium for it. As the article mentions, you can get around it using CUDA code (and possibly OpenCL code as well), although that's obviously not a real solution. There seems to have been a bit of an uproar about this, so it remains to be seen if nVidia will address this issue or not. I suspect if enough people jump ship and start buying AMD cards, they might give it more attention. --J ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
