simon@macondo.sw.mcc.com (Simon John Gibbs) (02/08/89)
The "premiere" issue of MIPS magazine (Feb. 89) has a review of the Personal Iris. There are some numbers given for graphics performance: about 16,000 vectors/sec. and less than 1000 polygons/sec. (here numbers are given for 1 light source, 2 light sources, shaded, and different image sizes - 1/4 screen, 1/2 screen and full screen). Anyway, they don't give much more information about these measurements but I was surprised that they're so low. Do these numbers make sense? What about the 100,000 polygons/sec. figure that's so frequently mentioned? Simon Gibbs simon@mcc.COM
blbates@AERO4.LARC.NASA.GOV (Bates TAD/HRNAB ms294 x2601) (02/09/89)
I would say there are a number of possibilities:
1. SGI sets up some test cases that give the best possible
results. (Most companies do)
2. MIPS gave it some extremely difficult test cases.
3. MIPS did not code up the test cases in an optimal fashion.
(i.e. poor programming)
4. All or part of the above.
Personally I don't trust ANY companies computer specs. If I can
put one of my own programs on the machine and can see how fast it
works, then I can be convinced on how good the machine is. Different
people use the machines in different ways, so they may get different
performance results. It all depends on what you are looking for.
--
Brent L. Bates
NASA-Langley Research Center
M.S. 294
Hampton, Virginia 23665-5225
(804) 864-2854
E-mail: blbates@aero4.larc.nasa.gov or blbates@aero2.larc.nasa.govgavin@krypton.SGI.COM (Gavin Bell) (02/11/89)
We quote 5,900 Z-buffered, Gouraud shaded, 4 sided 100x100 independent polygons per second on the Personal Iris. The 100,000 polygons/second figure you heard is for the GTX products. Will you get 5,900 polygons per second in your application? Not if: 1) You spend any time computing the polygons, or spend any time re-organizing the vertex data to match the v() commands. 2) You spend any time clearing your window or z-buffer (remember, it takes ~10 microseconds to clear the screen, so at 30 frames/second ~20 percent of your time is spent just clearing the framebuffer). 3) You have big polygons (bigger than 10 by 10 pixels). 4) You use the old drawing commands. 5) You draw few polygons in double-buffered mode. Worst case is drawing one polygon, then swapping buffers-- the swapbuffers() command has to wait for the vertical retrace of the monitor you are using, so you will get only ~60 polygons/second. The benchmark used to get the 5,900 poly/sec number is, of course, nowhere close to a real application. It is single-buffered, never clears the framebuffer or z-buffer, and has almost no CPU overhead, and draws only 100 pixel polygons. But it does give you an idea of maximum drawing speed. --gavin (gavin@sgi.com)
thant@horus.SGI.COM (Thant Tessman) (02/11/89)
In article <26596@sgi.SGI.COM>, gavin@krypton.SGI.COM (Gavin Bell) writes: > We quote 5,900 Z-buffered, Gouraud shaded, 4 sided 100x100 independent > polygons per second on the Personal Iris. The 100,000 polygons/second > figure you heard is for the GTX products. > > Will you get 5,900 polygons per second in your application? Not if: > 1) You spend any time computing the polygons, or spend any time > re-organizing the vertex data to match the v() commands. > 2) You spend any time clearing your window or z-buffer (remember, it > takes ~10 microseconds to clear the screen, so at 30 frames/second > ~20 percent of your time is spent just clearing the framebuffer). > 3) You have big polygons (bigger than 10 by 10 pixels). > 4) You use the old drawing commands. > 5) You draw few polygons in double-buffered mode. Worst case is > drawing one polygon, then swapping buffers-- the swapbuffers() > command has to wait for the vertical retrace of the monitor you > are using, so you will get only ~60 polygons/second. > [stuff deleted] > --gavin (gavin@sgi.com) > However... The polygons are z-buffered, gouraud-shaded, independent, randomly oriented quadrilaterals. Not a screen-aligned t-mesh. thant (thant@sgi.com)
madd@adt.UUCP (jim frost) (02/13/89)
>We quote 5,900 Z-buffered, Gouraud shaded, 4 sided 100x100 independent >polygons per second on the Personal Iris. I assume you mean 10x10, which is what the brag sheet says. Our testing has shown that in flat mode the personal iris does 17,500, which is wonderful for our product. jim frost associative design technology madd@bu-it.bu.edu