PKG.SPARKMAN@MCC.COM (Aubrey Sparkman) (03/06/87)
Help anyone... I DON'T UNDERSTAND MIPS. CSA advertises an expansion box that includes a 14 MHz 68020/6881 combination and claims a 1 Mip rating. I read yesterday in ELECTRONIC ENGEERING TIMES that the new MAC is rated at 2 Mips with a 16 MHz 68020/6881 combination. 10 % increase in cpu speed can't give a doubling in MIPS can it? Where did I lose it? I am NOT interested in a Mac vs Amiga war, only an understanding on MIP rating. Thanks all, Aubrey Sparkman -------
mwm@eris.BERKELEY.EDU (Mike (No one lives forever.) Meyer) (03/08/87)
In article <12284247071.70.PKG.SPARKMAN@MCC.COM> PKG.SPARKMAN@MCC.COM (Aubrey Sparkman) writes: >Help anyone... I DON'T UNDERSTAND MIPS. Not uncommon. MIPS is best interpreted as "Mythical Instructions Per Second." Problems include: What's an instruction; how do you count them; cacheing of instructions; page faults; etc., etc., etc. The best way to treat MIPS claims in ads is as advertising hype. Wade through the net.arch archives for details. >CSA advertises an expansion box that includes a 14 MHz 68020/68881 combination >and claims a 1 Mip rating. The ad I'm looking at (page 48 of the March/April '87 AmigaWorld) claims that it meets the CMU MMM specs, making it at least a 1 MIPS box (of course, it don't got a million pixels). Given a rought rating of .6 MIPS for a 68000 Amiga (C dhrystones, based on Unix PCC), and the 120% increase claim, you get a 1.32 MIPS machine. >I read yesterday in ELECTRONIC ENGEERING TIMES >that the new MAC is rated at 2 Mips with a 16 MHz 68020/6881 combination. >10 % increase in cpu speed can't give a doubling in MIPS can it? Where did I >lose it? That 120% increase is for the Turbo-Amiga processor alone. It's still running on 16-bit wide memory. The Mac has only 32-bit wide memory. How much difference that makes will depend on the processor architecture. I think 30% is about right for 68K style architechtures, which would make that 1.32 MIPS box a 1.82+ box. <mike, master of the art of hand-waving mathematics But I'll survive, no you won't catch me, Mike Meyer I'll resist the urge that is tempting me, ucbvax!mwm I'll avert my eyes, keep you off my knee, mwm@berkeley.edu But it feels so good when you talk to me. mwm@ucbjade.BITNET
bj@well.UUCP (Jim Becker) (03/10/87)
If it's any help, MIPS is an arbitrary unit of intructions per seconds for CPUS. This is noramlly optimized for any given tested computer. For example, the test that scored the 80386 at three to four MIPs tested noops. --- Not Kidding ---- -Jim Becker Terrapin Software
billk@crash.UUCP (03/12/87)
Oh. I thought MIPS was Million Instructions Per Second. You know, 5 MIPS meant 5 million instructions per second. Is that wrong? By the way, I've got a Novix SC4000 which will do (at 8 MhZ) 5 MIPS. Pretty snazzy little dude. It's a Forth engine. You don't program in Assembler on this thing, or, rather, programming in Forth on this little bugger is like programming in assembler on a 'normal' processor. Interesting little thing... Bill Kelly
daveh@cbmvax.UUCP (03/12/87)
in article <897@crash.CTS.COM>, billk@pnet01.CTS.COM (Bill Kelly) says: > > Oh. I thought MIPS was Million Instructions Per Second. You know, 5 MIPS > meant 5 million instructions per second. Is that wrong? > Bill Kelly No that's correct. But its still an arbitrary measurement. We all know what a "second" is, but what's an "instruction". If you were benchmarking the 80386 chip, you might decide to calculate how many NOPs the chip can execute per second. A 68020 benchmarker might choose to do something a little more useful, like benchmark the number of add instructions per second the chip can perform. But do you want 8, 16, or 32 bit adds? Register to register or memory to register? So then we start talking about things like average and peak MIPS. Peak is probably the number of the fastest instruction of any kind (maybe a NOP) that can be executed in a second. Average? Well, it could be a simple average of all possible instructions in all possible addressing modes. But that drags the performance down with all kinds of powerful-yet-esoteric instructions that take a long time to run and are rarely used in real life. So you can see that in practical terms a MIPS figure without lots of further information can be pretty meaningless. By the way, did I mention that the C128 can run at over 1 MIPS? -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Dave Haynie Commodore Technology // /| ___ __ __ __ {ihnp4|caip|rutgers}!cbmvax!daveh |\ // /_| | / \ / \ / \ Commodore rarely admits to knowing me, \\// / | +--+ | | | | | | much less sharing my personal opinions. \/ / | |___ \__/ \__/ \__/ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
tim@ism780c.UUCP (03/14/87)
In article <2743@well.UUCP> bj@well.UUCP (Jim Becker) writes: > >For example, the test that scored the 80386 at three to four MIPs tested noops. > >--- Not Kidding ---- Oh yeah? Well I just did a benchmark that did more than noops ( I did a loop that did an addition of two memory variables, and stored the result in a memory variable. The loop index was in a register ), and got 3 MIPS. I was running on a production 80386 board from Intel, under Unix System V.3, with two other users on the system. They were both just editing, so they shouldn't have slowed me down much. On the other hand, what you say may not be wrong. Running my benchamark with the addition removed gave 3.1 MIPS. I guess the 386 either has a fast addition or a slow nop! :-) -- Religion: just say "no" Tim Smith USENET: sdcrdcf!ism780c!tim Compuserve: 72257,3706 Delphi or GEnie: mnementh
dvmark@cca.UUCP (03/14/87)
Don't feel bad, there is much confusion about MIPS. I have had to deal with this for 15 years (notice my gray beard showing :^)= ). IBM use to quote Millions of Instructions Per Second ratings on new machines years ago. They *ACTUALLY* sampled some real user's old 360 mainframes and got a standard mix of instructions to use as a standard benchmark for IBM machines. The instruction set chosen contained a mix of binary arithmetic, address manipulation, character handling (strings to you C kind of guys) and packed decimal arithmetic. These were really good benchmarks, and gave useful MIPS ratings within the IBM mainframe line. However, competition killed the good benchmark. Sadly, IBM refuses to run their standard benchmark on newer machines. What happened is that some mainframe competitors started running benchmarks with NOPs in a loop and reported twice the MIPS of IBM's machines (though the IBM machines were really faster). This is like deja-vu with the INTEL 80386 claim of 3-4 MIPS (really NOPs) better than 68020 (running a good instruction set mix). IBM's official line is: MIPS stands for Meaningless Indicator of Performance. IBM refuses to quote MIPS anymore. Anyone can create a benchmark that favors their machine and beat someone else's honest MIPS rating taken to try to explain to customers how much faster the new machine is than the old one. To help answer the question, assuming the same bus width (16 or 32 bits), and that all parts of the system are upgraded in step (if your CPU is running faster, you better have memory that doesn't hold up the CPU), a 14 mhz machine will be twice the speed of a 7 mhz machine. This only applies to an identical hardware architecture. The Amiga has hardware assists (in it's custom chips) for some heavily CPU intensive stuff like graphics and sound that the Mac has to do all in software. That is why equivalent benchmarks on the 7.14 mhz Amiga are better than the 8 mhz Mac. The Mac has a different architecture than the Amiga, so comparing mhz rating is not useful. About all you can say is that X architecture running at double the mhz of the old X architecture should run twice as fast. I am not going to go into cacheing, pipelining and fancy stuff being done in the CPU to make them faster (like the 68000 vs the 68020), read the 68k news group for that stuff. _____ I wrote this with 3D glasses on, could you tell?
wagner@gpu.utcs.toronto.edu (03/16/87)
In article <2704@jade.BERKELEY.EDU> mwm@eris.BERKELEY.EDU (Mike (No one lives forever.) Meyer) writes: >In article <12284247071.70.PKG.SPARKMAN@MCC.COM> PKG.SPARKMAN@MCC.COM (Aubrey Sparkman) writes: >>Help anyone... I DON'T UNDERSTAND MIPS. > >Not uncommon. MIPS is best interpreted as "Mythical Instructions Per >Second." I've heard it as "Mythical Indication of Processor Speed" But the idea is the same. With varying workloads, the same hardware can exhibit vastly different MIPS ratings. This is most prevalent in microcoded machines like small 370s. Michael (another michael in this discussion group?) Wagner
vic@bobkat.UUCP (Vic Sohal) (03/17/87)
>>Not uncommon. MIPS is best interpreted as "Mythical Instructions Per >>Second." > >I've heard it as "Mythical Indication of Processor Speed" >But the idea is the same. With varying workloads, the same hardware can >exhibit vastly different MIPS ratings. This is most prevalent in microcoded >machines like small 370s. > I've NEVER EVER heard of MIPS referring to "Mythical Indication of Processor Speed", where did you hear it meant that? As far as I know, MIPS means "MIllion Instructions Per Second" Just Clearing The Air....Vic Sohal
tony@artecon.artecon.UUCP (Anthony D. Parkhurst) (03/18/87)
In article <763@bobkat.UUCP> vic@bobkat.UUCP (Vic Sohal) writes: >>>Not uncommon. MIPS is best interpreted as "Mythical Instructions Per >>>Second." >>I've heard it as "Mythical Indication of Processor Speed" >I've NEVER EVER heard of MIPS referring to "Mythical Indication of Processor >Speed", where did you hear it meant that? >As far as I know, MIPS means "MIllion Instructions Per Second" That is just the point, the acronym has been modified to reflect reality. Just what exactly does "Million Instructions Per Second" mean?!? Define "Instruction" (extra credit if you can indicate the different definitions from the different manufacturers). > Just Clearing The Air....Vic Sohal Just Clearing The Air....Tony Parkhurst -- **************** Insert 'Standard' Disclaimer here: OOP ACK! ***************** * Tony Parkhurst -- {hplabs|sdcsvax|ncr-sd|hpfcla|ihnp4}!hp-sdd!artecon!adp * * -OR- hp-sdd!artecon!adp@nosc.ARPA * *******************************************************************************
wagner@utgpu.UUCP (03/19/87)
In article <763@bobkat.UUCP> vic@bobkat.UUCP (Vic Sohal) writes: >>>Not uncommon. MIPS is best interpreted as "Mythical Instructions Per >>>Second." >> >>I've heard it as "Mythical Indication of Processor Speed" >>But the idea is the same. With varying workloads, the same hardware can >>exhibit vastly different MIPS ratings. This is most prevalent in microcoded >>machines like small 370s. >> > >I've NEVER EVER heard of MIPS referring to "Mythical Indication of Processor >Speed", where did you hear it meant that? > >As far as I know, MIPS means "MIllion Instructions Per Second" > > Just Clearing The Air....Vic Sohal From a talk given by an IBM performance expert. You can also substitute "Misleading" for the first word. IBM nowadays uses throughput numbers. And MIPS did originally mean Million Instructions Per Second. But, as the conversation on this topic has already pointed out, it's a pretty useless speed indicator. Michael