daves@martha (Dave Springer) (04/25/91)
I would like to know just exactly how fast a PI is, and all the other machines SGI makes. When I say "how fast" I mean how many 10-pixel marketing lines can you _really_ get from a client app to the screen; how many 100-pixel marketing triangles can you _really_ get, etc. (this includes GL overhead, transformations, clipping, and so on). Please include price information, like "On a PI that costs $15,000, you can get 2000 smooth-shaded polygons per sec, with 3 lights". Obviously I can get some numbers from the marketing literature, but I get enough lies as it is. Please respond directly to daves@next.com. Thanks. der Springer.
tarolli@westcoast.esd.sgi.com (Gary Tarolli) (04/25/91)
In article <558@rosie.NeXT.COM>, daves@martha (Dave Springer) writes: >I would like to know just exactly how fast a PI is, and all the other machines >SGI makes. When I say "how fast" I mean how many 10-pixel marketing lines can >you _really_ get from a client app to the screen; how many 100-pixel marketing >triangles can you _really_ get, etc. (this includes GL overhead, >transformations, clipping, and so on). Please include price information, like The numbers we report in our marketting literature are actually attainable in a real application. When we say the VGX draws 1 million triangles per second (or whatever the number is), you can write a C pgm to get that performance. The numbers all come from actual C benchmark programs. Now this doesn't guarantee your application can achieve these numbers, but if you write your code carefully and operate in the same mode(s) as the benchmark, you can. -------------------- Gary Tarolli
drb@eecg.toronto.edu (David R. Blythe) (04/29/91)
In article <558@rosie.NeXT.COM> daves@martha (Dave Springer) writes: >I would like to know just exactly how fast a PI is, and all the other machines >SGI makes. When I say "how fast" I mean how many 10-pixel marketing lines can >you _really_ get from a client app to the screen; how many 100-pixel marketing >triangles can you _really_ get, etc. (this includes GL overhead, >transformations, clipping, and so on). Please include price information, like >"On a PI that costs $15,000, you can get 2000 smooth-shaded polygons per sec, >with 3 lights". Obviously I can get some numbers from the marketing >literature, but I get enough lies as it is. Please respond directly to From my experience the "marketing" numbers are pretty accurate, at least I can get the claimed performance if I sweat over my code for a few hours. Oddly enough, I noted the literature now rates the performance of a 4D/70GT at 60K polygons/sec up from 40K polygons/sec. The result of crafty microcoding and new gl primitives? -drb
paquin@kahua.esd.sgi.com (Tom Paquin) (05/04/91)
In article <558@rosie.NeXT.COM>, daves@martha (Dave Springer) writes: |> When I say "how fast" I mean how many 10-pixel marketing lines can |> you _really_ get from a client app to the screen ... |> Obviously I can get some numbers from the marketing |> literature, but I get enough lies as it is. My experience is that numbers which come out of SGI marketing are reproducible on a typical machine. Of course, running makes and several GL apps simultaneously will trash a benchmark. But it doesn't take odd configurations, special software, or eye of newt in the dead of night to get published numbers from an Iris 4D. Carefully writing a simple C test should do it. -- -Tom ***** Opinions are mine, etc.