usenet@cps3xx.UUCP (Usenet file owner) (02/19/89)
We all realize that the NeXT's role in computing is not to be a number-cruncher. But since all other computers are subjected to performance benchmarks, I thought it only fair that ask the same of the Next. I therefore requested benchmark info from comp.sys.next readers a while ago. Two people responded to my request for NeXT benchmark results, giving mostly comparisons rather than hard benchmark data. (The comparisons were really what I was after, anyway.) I correlated their results with non-NeXT benchmark info of my own and from reviews in Unix Review magazine, and it all seems pretty consistent. I will summarize the comparisons in a table to spare you the effort of reading long English paragraphs, but be aware that the results are less precise than the numbers would imply. Disclaimers apply. All numbers are normalized indices of performance, with larger numbers being better. NeXT 12 Mhz 20 Mhz Sun 3/50 Sun 3/260 Sun 4/250 AT clone Decent 386 CPU (no 1 1.25 16bit; 0.5 1.1 3.2 float) 1.5 32bit Sequential 1 hard; 0.2 1 (DOS) 0.3 1 1.3 disk I/O 0.5 od The optical disk (as mentioned by other posters) is half as fast writing compared with reading, whereas with most hard disks (including the NeXT's), writing is somewhat greater than half the reading speed. (I'm talking file system I/O.) I find the sequential I/O performance of the NeXT machine (both hard disk and optical) rather impressive. I'm also impressed at the speed of the 80386 machines under DOS, especially with 32-bit compilers. (Note: Xenix and System V ports on a 386 seem to have much slower disk I/O than DOS.) This information is all very imprecise and, of course, the NeXT is not yet a mature machine. Still, I would not expect the NeXT's integer benchmark performance to improve significantly. I assume that the reviewer's benchmarks were compiled with with GNU GCC, and word (from Winter 89 Usenix) is that the 68020 version of GCC is quite decent. Apparently, the 68030 is not much more than a 68020 with more on-board memory management, so I would assume that the NeXT GCC compiler is also good. Improvements in the NeXT OS would probably not affect integer benchmark performance. Thanks to the two respondants (who apparently prefer to remain anonymous). I will post a followup when our NeXT machine arrives. Meanwhile, I would certainly appreciate any corrections or additions. Mark Riordan Michigan State Univ. riordanmr@clvax1.cl.msu.edu
helman@isl.Stanford.EDU (Jim Helman) (02/21/89)
In article <1887@cps3xx.UUCP> riordan@frith.UUCP (Mark Riordan) writes: >All numbers are normalized indices of performance, with larger numbers >being better. > > > NeXT 12 Mhz 20 Mhz Sun 3/50 Sun 3/260 Sun 4/250 > AT clone Decent 386 ... >Sequential 1 hard; 0.2 1 (DOS) 0.3 1 1.3 >disk I/O 0.5 od > >I find the sequential I/O performance of the NeXT machine (both hard disk and >optical) rather impressive. I'm also impressed at the speed of the 80386 >machines under DOS, especially with 32-bit compilers. (Note: Xenix and >System V ports on a 386 seem to have much slower disk I/O than DOS.) > It's very hard to come to any conclusions at all from the disk I/O comparions with the Suns without knowing what kind of drive and controller were used in the benchmark, e.g., did the 3/260 have a SCSI drive or a 3.0MB/sec SMD drive on a 753 controller. Such missing factors could change things by a factor of 2 or more. Does anyone know the characteristics of the NeXT's hard drive? Jim Helman Department of Applied Physics P.O. Box 10494 Stanford University Stanford, CA 94309 (jim@thrush.stanford.edu) (415) 723-4940 Jim Helman Department of Applied Physics P.O. Box 10494 Stanford University Stanford, CA 94309 (jim@thrush.stanford.edu) (415) 723-4940