brsmith@cs.umn.edu (Brian R. Smith) (11/12/89)
Many people have responded to my original posting; thanks to all of you. Unfortunately, I still don't have an explanation. SparcStations can have 1 of 2 different floating-point coprocessors (or accelerators; I'm not sure what the difference is); either a Texas Instruments chip (which occupies a daughterboard), or a Weitek chip. I found out after I posted that one of our Sparcs did indeed have a TI fpu. The difference between it and the eight "fast" Weitek's wasn't detectable by my crude benchmark. The three "slow" Weitek's took 50% longer than either the "fast" Weitek's or the TI. A couple of days ago, I decided to run the "benchmark" again, as one of the machines had been swapped. I'd deleted the benchmark, so I hacked up a new one, and ran it on all the machines to get a comparison. Of the slow three, "kilo" is down (power supply problems), "deca" was still slow, and "giga" appears to be back with the "fast" Weitek's. Two of the other "fast" Weitek's ran my benchmark at the same speed as "deca"... As you can guess, I'm thoroughly confused by now. Time for more testing. Anyone have a good, generic set of floating-point benchmarks? Brian brsmith@umn-cs.cs.umn.edu