ed@zaphod.uchicago.edu (Ed Friedman) (02/07/89)
We have a 4D/70G which seems to run graphics very slowly compared to older Iris models. Is there some test we can do (benchmark?) to find out if the G option is in fact installed and working properly? Is it possible that the 4Sight windowing system is the cause of this slowdown? Thanks in advance, Ed Friedman
madd@adt.UUCP (jim frost) (02/07/89)
>We have a 4D/70G which seems to run graphics very slowly compared to >older Iris models. Is there some test we can do (benchmark?) to find >out if the G option is in fact installed and working properly? Is it >possible that the 4Sight windowing system is the cause of this slowdown? We ran into this same problem when porting our application from the 3000's to the 4D. The biggest slowdown we had was because the default shading model was different; changing it to flat (we use the machine as a blindingly fast 2D machine) was a substantial improvement. Also, check out the list suggested functions in the porting guide. They say that many of the 3000 compatibility functions run substantially slower (10X is the figure they quote) than the newer ones. We did find improvements when using these functions, especially on GT's. Be careful -- the newer functions were not implemented in some of the older gl libraries that were shipped; the original 3.01 release had quite a few that would print out "<func> is not implemented". 3.01C (I believe this is the current release) works fine. The flight simulator has many nice improvements, too. The 3000's still seem to outperform the 4D's in some things, at least with our product, but that may be a result of the design of our graphics routines. About 4Sight: it's not the greatest, but compared to mex it's fantastic. If it only supported a meta key.... jim frost associative design technology madd@bu-it.bu.edu
spike@adt.UUCP (Joe Ilacqua) (02/08/89)
<We have a 4D/70G which seems to run graphics very slowly compared to <older Iris models. [...] Is it possible that the 4Sight windowing system <is the cause of this slowdown? One thing to check is, if you are not using Gouraud shading, be sure you have a 'shademodel(FLAT)' near the start of your program. If you do not, everything will be drawn Gouraud shaded. This was causing a huge preformance hit on some code we ported to a 4D/70. Joe Ilacqua Associative Design Technology
zombie@voodoo.UUCP (Mike York) (02/09/89)
In article <8902071551.AA16241@adt.uucp> madd@adt.UUCP (jim frost) writes: > >The 3000's still seem to outperform the 4D's in some things, at least >with our product, but that may be a result of the design of our >graphics routines. On a similar note, we've found some things to run significantly slower on a 4D/70 GT than on our 4D/70 G's, particularly picking (which is all important in our application). With 3.0.1, picking was 6 times slower on the 4D/70 GT. with 3.1C, picking is only 2 times slower. Can hardly wait for 3.1D/3.2/WhateverTheyWantToCallIt :^). The explanation I got from SGI is that the architecture of the GT models does not lend itself well to operations that imply feedback (picking, popattributes, etc). Ironically, it's beginning to look like the best overall performer for our application (a technical illustration package) is the 4D/20 Super (or whatever they call the Eclipse with all the bit planes and the FPA). The vector graphics are only slightly slower than the 4D/70 G (for our application), but the raster capabilities are so much better. It's looking like a pretty nice little box. -- Mike York Boeing Computer Services, Renton, Washington (206) 234-7724 uw-beaver!ssc-vax!voodoo!zombie