kenb@amc.com (Ken Birdwell) (06/18/91)
I'm looking for some sort of performance figures for shaded Z-buffer (either 3D triangles, bilinear surfaces or quadrilaterals) engines. I realise that it all depends of number of pixels per polygon and such but I would like to know what to expect in a controlled environment. I'm looking for a board to do 3D graphics for a IBM-PC bus system and need an estimate on the pixel or polygon per second rate. Silicon Graphics has a PC based system claims 14k/sec, but seemed much slower when I saw it in real life. i860 people are always claiming 30k/sec but I've never seen one driving a display. I've seen ads saying 20k polygons/second with a TI34020/DSP32C system, but I'm not sure how they got their numbers and no one in town has one. As far as non-PC systems go, I've seen the stuff from SG on their VGX series system (1 triangles million/sec), but I suspect it's too expensive (I need to keep the complete system under $10k). Kurt Akeley from SG sent out a simple test showing from 100k to 1086k lighted triangles per second a while back (Apr 1st) and I was wondering if anyone has run a version on another platform. (see kurt@cashew.asd.sgi.com for more info; no association with me) I have a client with a specialized data-acquisition board whose data (5k-20k polygone mesh) needs to be rendered "real-fast" and "real-often". Also, they would like to be able to rotate a 20k line Z-buffer wire mesh version at at least 15 fps. They would also like to stick with a IBM-PC based system (because of the data-acquisition board) and I need to tell them what they can expect, and what they could expect if they go to either a SG or HP or R6000 or whatever. This is for resale and they'll need several hundred systems. If you have one of these systems or work for someone who builds them, please send me whatever you can about its performance. PS: The data on the acquisition board is in dual-ported memory and they're not too adversed to putting on a high speed data transfer bus. --