mkelly@cs.uoregon.edu (Michael A. Kelly) (02/16/91)
How many MIPS is the Mac capable of? I see ads for workstations such as the SPARCs which say they run at 12 or 26 MIPS, and I think 'but how does that compare to my Mac IIcx?' Thanks in advance, Mike. -- _______________________________________________________________________________ Michael A. Kelly America Online: Michael792 mkelly@cs.uoregon.edu Compu$erve: 73567,1651 _______________________________________________________________________________
Your-MDF-name@mitre.org (Enter your name) (02/17/91)
In article <1991Feb16.025224.25758@cs.uoregon.edu> mkelly@cs.uoregon.edu (Michael A. Kelly) writes: > How many MIPS is the Mac capable of? I see ads for workstations such as the > SPARCs which say they run at 12 or 26 MIPS, and I think 'but how does that > compare to my Mac IIcx?' I am not an expert in this area but have looked in to it a little. You may be aware of the program Speedometer for the mac which is available on sumex. I have used this program to benchmark several of our machines here. The program provides , among other things, KWhetstones/sec and Dhrystones/sec. Using a spec I heard concerning a basic DEC machine that 1750 Dhrystones==1MIP, I have calculated the corresponding MIPs on the Macs. You may be a little dissapointed with your IIcx in comparison to 26 MIP workstation. MacIIcx MacIIci MacIIfx ---------------------------------------------------- 3816Dhry=2.18MIP 5791Dhry=3.31MIP 11029Dhry=6.3MIP I make no claim as to the validity of these numbers, this is just what I got. There is alot of debate concerning these benchmarks (see comp.sys.benchmark - mostly workstations though). From platform to platform there is compiler variability and instruction efficiency (i.e., one function may do the work of two) etc. As a note, I understand that the Dhrystone benchmark is register and memory intensive where as the Whetstone is math/FPU intensive. The math performance on the Mac seems fairly good to me when the FPU is utilized. As always, I would be interested in corrections , comparisons and comments. -GSM gmarzot@linus.mitre.org
dawg6844@uxa.cso.uiuc.edu (<blank>) (02/17/91)
mkelly@cs.uoregon.edu (Michael A. Kelly) writes: >How many MIPS is the Mac capable of? I see ads for workstations such as the >SPARCs which say they run at 12 or 26 MIPS, and I think 'but how does that >compare to my Mac IIcx?' >Thanks in advance, >Mike. >-- >_______________________________________________________________________________ >Michael A. Kelly America Online: Michael792 >mkelly@cs.uoregon.edu Compu$erve: 73567,1651 >_______________________________________________________________________________ Patterson & Resnick, the founders of RISC, spend a chapter of their latest book on hardware design explaining why MIPS and MFLOPS, the two most common 'measures' of computer performance are basically hogwash. Since Cycles per Instruction (CPI) and the design of the FPU (if it is there) are different from machine to machine, it is meaningless to compare machines of different architectures by using Mips or Mflops. (Similar to comparing cars by how many times the pistons go up and down). They argue that the only true measure of computer performance is the execution time of REAL programs (not 'toy' programs or benchmarks). example: a machine with optional floating point hardware. it generally takes more clock cycles per flop then per integer instruction, so programs that use the optional FPU take less time but have a LOWER MIPS than programs that use software flop emulation, which executes many more instructions(making it slower) but has higher MIPS because the instructions it executes are smaller. ie: MIPS can vary INVERSELY to performance in certain situations. So how does all this answer your question? It doesn't. The only true way to compare these two machines is to run the same software and compare execution times. (compile the same large program perhaps? but even that is bad, because are the compilers the same?) I am a smalltalk programmer, and smalltalk has the unique feature of being able to run across many platforms. A large package we developed here runs about one and a half to two times as fast on the Sparc I in the lab as it does on my IIci. That ought to give you some idea about relative performance. (Although even this isnt perfect, as the VM interface is different for the two machines) So anyway, have a nice day :) Dan
rose@beowulf.ucsd.edu (Dan Rose) (02/17/91)
dawg6844@uxa.cso.uiuc.edu (<blank>) writes: >mkelly@cs.uoregon.edu (Michael A. Kelly) writes: >>How many MIPS is the Mac capable of? I see ads for workstations such as the >>SPARCs which say they run at 12 or 26 MIPS, and I think 'but how does that >>compare to my Mac IIcx?' >Patterson & Resnick, the founders of RISC, spend a chapter of their >latest book on hardware design explaining why MIPS and MFLOPS, the two most >common 'measures' of computer performance are basically hogwash. . . . >... >The only true way to compare these two machines is to run the same software and >compare execution times. (compile the same large program perhaps? but even that is bad, because are the compilers the same?) Even this doesn't tell the whole story. It's nearly impossible to hold everything else equal. On a system with preemptive multitasking, are you measuring the speed with all the daemons disabled? Ones that you normally need to run the system? How full is the disk? What's the scheduling algorithm? What's the size of the "working set" of the program you're timing, versus the size of the memory? On the Mac, is the RAM cache turned on? How big is it? How much is the timing code itself affecting the speed of the program being timed? Etc. This is not to say that you can't ever measure relative differences. A IIfx really is faster than an SE. Just take any numbers with a large grain of salt. -- Dan Rose drose@ucsd.edu