The pbo result was for the M2070 (not the Q). Here is the original M2070 result
>>>>>>>>>> PIXEL FORMAT: BGRA <<<<<<<<<< glDrawPixels(): 427.5 Mpixels/sec glReadPixels(): 175.5 Mpixels/sec (min = 167.1, max = 176.7, sdev = 0.5792) glReadPixels() accounted for 100.00% of total readback time The snippets below are for the M2070Q (top) and M2070 with pbo (below that). Paul ----- Original Message ----- > From: "Paul Melis" <paul.me...@surfsara.nl> > To: virtualgl-users@lists.sourceforge.net > Sent: Monday, 17 June, 2013 10:49:55 PM > Subject: Re: [VirtualGL-Users] Improving performance on very large displays > > On 06/17/13 11:23, Paul McIntosh wrote: > > A bit better... > Err, the PBO numbers seem worse? > > Pbuffer (BGRA): > > glDrawPixels(): 429.4 Mpixels/sec > glReadPixels(): 688.5 Mpixels/sec (min = 605.3, max = 697.0, sdev = > 4.143) > glReadPixels() accounted for 99.98% of total readback time > > PBO (BGRA): > > glDrawPixels(): 427.9 Mpixels/sec > glReadPixels(): 226.6 Mpixels/sec (min = 166.2, max = 227.5, sdev = > 2.838) > glReadPixels() accounted for 1.050% of total readback time > > Paul > > > > [paulmc@m1029 paulmc]$ glreadtest -pbo > > > > GLreadtest v2.3.2 (Build 20121221) > > > > Rendering to Pbuffer using GLX on display :1.0 > > Using PBO's for readback > > Xlib: extension "GLX" missing on display ":1.0". > > Xlib: extension "GLX" missing on display ":1.0". > > 256: Could not obtain Visual > > [paulmc@m1029 paulmc]$ export DISPLAY=:0 > > [paulmc@m1029 paulmc]$ glreadtest -pbo > > > > GLreadtest v2.3.2 (Build 20121221) > > > > Rendering to Pbuffer using GLX on display :0.0 > > Using PBO's for readback > > FB Config = 0xc5 > > Drawable size = 701 x 701 pixels > > Using 1-byte row alignment > > > >>>>>>>>>>> PIXEL FORMAT: LUM <<<<<<<<<< > > glDrawPixels(): 1160 Mpixels/sec > > glReadPixels(): 0.7835 Mpixels/sec (min = 0.7821, max = 0.7848, > > sdev = 0.00135 > > 2) > > glReadPixels() accounted for 99.97% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: RED <<<<<<<<<< > > glDrawPixels(): 425.3 Mpixels/sec > > glReadPixels(): 443.9 Mpixels/sec (min = 184.9, max = 447.6, sdev > > = 8.973) > > glReadPixels() accounted for 2.117% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: BGRA <<<<<<<<<< > > glDrawPixels(): 427.9 Mpixels/sec > > glReadPixels(): 226.6 Mpixels/sec (min = 166.2, max = 227.5, sdev > > = 2.838) > > glReadPixels() accounted for 1.050% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: ABGR <<<<<<<<<< > > glDrawPixels(): 427.8 Mpixels/sec > > glReadPixels(): 226.4 Mpixels/sec (min = 168.5, max = 227.5, sdev > > = 2.799) > > glReadPixels() accounted for 1.176% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: BGR <<<<<<<<<< > > glDrawPixels(): 424.9 Mpixels/sec > > glReadPixels(): 267.3 Mpixels/sec (min = 197.2, max = 268.1, sdev > > = 3.026) > > glReadPixels() accounted for 1.288% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: RGBA <<<<<<<<<< > > glDrawPixels(): 427.7 Mpixels/sec > > glReadPixels(): 226.7 Mpixels/sec (min = 169.7, max = 227.4, sdev > > = 2.671) > > glReadPixels() accounted for 1.042% of total readback time > > > >>>>>>>>>>> PIXEL FORMAT: RGB <<<<<<<<<< > > glDrawPixels(): 359.3 Mpixels/sec > > glReadPixels(): 267.2 Mpixels/sec (min = 199.1, max = 268.2, sdev > > = 2.971) > > glReadPixels() accounted for 1.285% of total readback time > > > > Paul > > > > ----- Original Message ----- > >> From: "DRC" <dcomman...@users.sourceforge.net> > >> To: virtualgl-users@lists.sourceforge.net > >> Sent: Monday, 17 June, 2013 5:43:52 PM > >> Subject: Re: [VirtualGL-Users] Improving performance on very large > >> displays > >> > >> Does enabling PBOs make any difference? > >> > >> > >> On 6/17/13 2:02 AM, Arthur HUILLET wrote: > >>> On Mon, Jun 17, 2013 at 11:30:34AM +1000, Paul McIntosh wrote: > >>>> Ok - here is the data for a Tesla M2070 and M2070Q (Quadro) > >>>> > >>>> There is a big difference between the Quadro and non-Quadro. > >>>> Looks > >>>> like the read rate is 1/4 for the non-Quadro. These are > >>>> virtually > >>>> same hardware, however I know that the Quadro has some cache > >>>> size > >>>> differences for transferring data (so it can't all be blamed on > >>>> the driver). > >>> Thanks for confirming my remark with actual figures. I do believe > >>> that it can all be blamed on the driver, and that it was designed > >>> that way, but that doesn't really matter - non Quadro readback > >>> performance may very well be too limited for certain use cases; I > >>> think this is the message to take home. > >>> > >>> Out of curiosity, can you elaborate as to the cache size > >>> differences? Where did you get this information and what > >>> differences are there exactly? > >>> > >> ------------------------------------------------------------------------------ > >> This SF.net email is sponsored by Windows: > >> > >> Build for Windows Store. > >> > >> http://p.sf.net/sfu/windows-dev2dev > >> _______________________________________________ > >> VirtualGL-Users mailing list > >> VirtualGL-Users@lists.sourceforge.net > >> https://lists.sourceforge.net/lists/listinfo/virtualgl-users > >> > > > ------------------------------------------------------------------------------ > This SF.net email is sponsored by Windows: > > Build for Windows Store. > > http://p.sf.net/sfu/windows-dev2dev > _______________________________________________ > VirtualGL-Users mailing list > VirtualGL-Users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/virtualgl-users > -- Dr Paul McIntosh Senior HPC Consultant, Technical Lead, Multi-modal Australian ScienceS Imaging and Visualisation Environment (www.massive.org.au) Monash University, Ph: 9902 0439 Mob: 0434 524935 ------------------------------------------------------------------------------ This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev _______________________________________________ VirtualGL-Users mailing list VirtualGL-Users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/virtualgl-users
