dhoyman@vms.macc.wisc.edu (02/18/91)
An earlier posting asked for the MIPS rating of a Mac and later postings have discussed the appropriateness of MIPS and FLOPS (unfortunately I have lost those, so I cannot post a followup). While I would agree that one cannot compare MIPS directly, the Dhystone benchmark is often used as a measure of MIPS with the conversion factor of 1750 Dhrystone/s = 1 MIPS. However, Dhrystone is actually measuring the number of passes thru a loop, not actual MIPS. Thus, one can compare Dhrystones/s on different machines. This is in a sense a measure of raw horsepower, integer performance. Along with integer performance, Dhrystone is also compiler and OS dependant, although it was constructed so as to minimize this variability. For a complete discussion of the Dhrystone see The Communications of the ACM Vol 27, No 10, 10/84 pg. 1013. With this in mind, allow me to post some Dhyrstone timings I have obtained on various machines. I used the Dhystone 1.1 written in C. On the Mac I built a single version with Think C. On Dos I used Turbo C. On the Unix/Vms systems I used the vendor's C. Computer O/S Dhrystones/sec Cray Y-MP UNICOS 16000 Zentith 386 33Mz MS-DOS 8333 Gateway 386 33Mz MS-DOS 8333 VAX 6420 VMS 7149 Mac IIci Macintosh 7145 Sequent Unix 3699 Mac IIcx Macintosh 3500 VAX 3200 VMS 2941 Mac II Macintosh 2777 AT&T 6386 MS-DOS 2631 AT&T 3B2 600 Unix System V 1785 MicroVAX II Ultrix 1077 Mac SE Macintosh 877 VAXstation II VMS 862 Mac Plus Macintosh 735 Zenith 8086 10Mz MS-DOS 476 Sorry for the poor formatting, I had this on an Excel spreadsheet. I have run this on a Sun SparcStation, but I think the number is way too low. I would also draw your attention to the Cray number. Remember that this is for INTEGER performance. If we would run the Whetstone or some other floating point benchmark, the Cray would show a BIG advantage. Actually, I only take this with a grain of salt. The only comparisons that I really believe are application oriented. See MacWorld March 1991 for a good comparison of Macs and PC's application speed. Dirk Herr-Hoyman UW-Madison, Dept. of Family Medicine and Practice dhoyman@fammed.wisc.edu 608-262-6368
dawg6844@uxa.cso.uiuc.edu (<blank>) (02/19/91)
dhoyman@vms.macc.wisc.edu writes: >An earlier posting asked for the MIPS rating of a Mac and later postings >have discussed the appropriateness of MIPS and FLOPS (unfortunately I have >lost those, so I cannot post a followup). >While I would agree that one cannot compare MIPS directly, the Dhystone >benchmark is often used as a measure of MIPS with the conversion factor >of 1750 Dhrystone/s = 1 MIPS. However, Dhrystone is actually measuring >the number of passes thru a loop, not actual MIPS. Thus, one can compare >Dhrystones/s on different machines. This is in a sense a measure of raw >horsepower, integer performance. Along with integer performance, Dhrystone >is also compiler and OS dependant, although it was constructed so as to >minimize this variability. For a complete discussion of the Dhrystone see >The Communications of the ACM Vol 27, No 10, 10/84 pg. 1013. >With this in mind, allow me to post some Dhyrstone timings I have obtained >on various machines. I used the Dhystone 1.1 written in C. On the Mac I >built a single version with Think C. On Dos I used Turbo C. On the Unix/Vms >systems I used the vendor's C. >Computer O/S Dhrystones/sec > >Cray Y-MP UNICOS 16000 >Zentith 386 33Mz MS-DOS 8333 >Gateway 386 33Mz MS-DOS 8333 >VAX 6420 VMS 7149 >Mac IIci Macintosh 7145 >Sequent Unix 3699 >Mac IIcx Macintosh 3500 >VAX 3200 VMS 2941 >Mac II Macintosh 2777 >AT&T 6386 MS-DOS 2631 >AT&T 3B2 600 Unix System V 1785 >MicroVAX II Ultrix 1077 >Mac SE Macintosh 877 >VAXstation II VMS 862 >Mac Plus Macintosh 735 >Zenith 8086 10Mz MS-DOS 476 >Sorry for the poor formatting, I had this on an Excel spreadsheet. I have >run this on a Sun SparcStation, but I think the number is way too low. I >would also draw your attention to the Cray number. Remember that this is >for INTEGER performance. If we would run the Whetstone or some other >floating point benchmark, the Cray would show a BIG advantage. >Actually, I only take this with a grain of salt. The only comparisons >that I really believe are application oriented. See MacWorld March 1991 for >a good comparison of Macs and PC's application speed. >Dirk Herr-Hoyman >UW-Madison, Dept. of Family Medicine and Practice >dhoyman@fammed.wisc.edu >608-262-6368 While these numbers are fun, they unfortunately are not any more informative than MIPS or MFLOPS. Toy programs, and 'benchmarks' like whetstone and drhystone, are really a test of the compiler-writers ingenuity. In fact since the code that whetstone and drhystone execute is so simple and publicly available (it is a simple loop) there have been cases of compilers written to detect whetstone and drhystone loops and perform special optimizations. Again, the only true measure of preformance is execution time of real pro grams. Note that the Cray number above is rather silly, and that the numbers would lead you to believe that the 386's substantially outperfom the IIci, despite studies showing quite the opposite. (using mac vs windows apps performance) I would also be interested in the Sparc number, as the sparc2 here in the lab vastly outperforms any macs and pcs we have. The bottom line is that you cannot possibly compare machines by using only one number, no matter what that number is. Dan Walkowski University of Illinois, Dept. of Computer Science walkowsk@cs.uiuc.edu
krk@cs.purdue.EDU (Kevin Kuehl) (02/19/91)
In article <1991Feb18.160733.20724@macc.wisc.edu> dhoyman@vms.macc.wisc.edu writes:
While I would agree that one cannot compare MIPS directly, the Dhystone
benchmark is often used as a measure of MIPS with the conversion factor
of 1750 Dhrystone/s = 1 MIPS.
One thing no one has mentioned when comparing MIPs is that not all MIP
measurements are the same. Some vendors use ``true'' MIPs which is
very misleading, especially for RISC architectures. Others use Vax
MIPs where they run a piece of benchmark code on a machine and compare
it to the speed of a Vax 11/780. If it runs the code 15 times faster
than the Vax, then the machine is declared to be 15 Vax MIPs.
But MIPs are pretty irrelevant. Bus speed, memory speed, disk speed,
etc. all go into deciding how fast a machine is.
--
Kevin Kuehl
krk@cs.purdue.edu
kuehlkr@mentor.cc.purude.edu
emg@Metaphor.COM (Mike Greenawalt) (02/22/91)
In article <1991Feb18.160733.20724@macc.wisc.edu>, dhoyman@vms.macc.wisc.edu writes: ... some material deleted ... |> With this in mind, allow me to post some Dhyrstone timings I have obtained |> on various machines. I used the Dhystone 1.1 written in C. On the Mac I |> built a single version with Think C. On Dos I used Turbo C. On the Unix/Vms |> systems I used the vendor's C. |> |> Computer O/S Dhrystones/sec |> |> |> Cray Y-MP UNICOS 16000 |> Zentith 386 33Mz MS-DOS 8333 |> Gateway 386 33Mz MS-DOS 8333 |> VAX 6420 VMS 7149 |> Mac IIci Macintosh 7145 |> Sequent Unix 3699 |> Mac IIcx Macintosh 3500 |> VAX 3200 VMS 2941 |> Mac II Macintosh 2777 |> AT&T 6386 MS-DOS 2631 |> AT&T 3B2 600 Unix System V 1785 |> MicroVAX II Ultrix 1077 |> Mac SE Macintosh 877 |> VAXstation II VMS 862 |> Mac Plus Macintosh 735 |> Zenith 8086 10Mz MS-DOS 476 |> ... some material deleted ... |> Dirk Herr-Hoyman |> UW-Madison, Dept. of Family Medicine and Practice |> dhoyman@fammed.wisc.edu |> 608-262-6368 There is something wrong with these numbers. Recent experience with some RISC processors give Dhrystone numbers in the 30000+ range. These seem too small to me. But, I don't want to fuel a long controversy over Dhrystone numbers. A few years ago I was involved in a project to get the best Dhrystone numbers out of a newly-designed architecture. Deep study of the Dhrystone code and the factors which influence its performance revealed that nothing counted more than the performance of the strcpy routine in the C library (unless the compiler could recognize and in-line the string moves). Dhrystone really measures moving characters around. Maybe that is why the Cray number seems relatively low. -- +---------------------------+--------------------------------------------+ | Mike Greenawalt | emg@metaphor.com | | Manager, Software Quality | ...!{apple|decwrl}!metaphor!emg | | Assurance | | | Metaphor Computer Systems | Clever quote under construction. | | Mountain View, CA | | +---------------------------+--------------------------------------------+
russotto@eng.umd.edu (Matthew T. Russotto) (03/06/91)
In article <1182@exua.exeter.ac.uk> kt@msor.exeter.ac.uk (Keith Tizzard) writes: >In article <13486@medusa.cs.purdue.edu> krk@cs.purdue.EDU (Kevin Kuehl) writes: >> >>One thing no one has mentioned when comparing MIPs is that not all MIP >>measurements are the same. Some vendors use ``true'' MIPs which is > > >What is a MIP? It's as valid an indicator of speed as a MIPS. (well, actually, vax MIPS have meaning on the Vax 11/780.... says a vax runs as fast as it runs :-) ) -- Matthew T. Russotto russotto@eng.umd.edu russotto@wam.umd.edu .sig under construction, like the rest of this campus. I mine 600 wells, and whaddo I get? Another day older and deeper in debt! --- Saddam Hussein.