jrg@Apple.COM (John R. Galloway) (03/22/89)
In light of the current MIPS/MFLOPS/Marketing series of postings I thought this might be an iluminating example. I was reading the latest copy of AT&T's "DSP Review" (vol 2 issue 1) on the DSP32C. One of the articles is about using the evaluation board (with 1 DSP32C) to calculate graphics trnaformations with the example of a line drawing of a VW bug. This particular drawing looked mighty familiar so I fired up my X 10.4 3d toy (its called geomout and I think it was pulled off the net a year or more ago) and indeed there it was (with 1078 polygons instead of the 1084 as stated in the article, but clearly very close). The article states that a picture is calculated in .85 seconds (using an assembly language matrix 4x4 multiply) or 1.28 seconds with just the standard C DSP32C library and then drawn in 4 seconds on the PC6300. For comparison I am running an Opus Systms PM200 which is a National 16 Mhz 32332, a 32081 floating point unit and 8MB of memory, in an 8Mhz AT (which the PM200 uses as an I/O processor) with a Viking 1 graphics board (Hitachi gpu). The cpu is at most 2 MIPS and I doubt even the most hype-ized of marketeers would claim more than 1 MFLOP for the 32081. My systems draws the VW in 5 secs, divided as 2 for calculatin and 3 for drawing (the article shows the use of float, but just to check I changed to doubles and the time didn't chnage enough to round up to the next second). I do not know what speed the DSP32C evaluation board was running at, but the article describes it as a 25 MFLOP part. The numbers however show only a 2.5 to 1 speed up, not the 25 to 1 the MFLOPS would suggest by themselves. This of course is comparing 2 different programs (they might have a common ancestry given the common database being used) and run on two very different systems. This in NO way is a claim that the DSP32C is anything but a wonderful part, I would love to have a handfull (a Mac-II combination signal processor and graphics board with say 8 of them and a A/D D/A seems like it wold be a fun project :-), I really know nothing about the chip, this was just an extreme example of how misleading isolated MIPS/MFLOPS claims can be. -jrg apple!jrg John R. Galloway, Jr. contract programmer, San Jose, Ca These are my views, NOT Apple's, I am a GUEST here, not an employee!!