mark@mips.COM (Mark G. Johnson) (07/12/90)
The June 1990 issue of _IEEE_Micro_ contains an article about the
Morotola 68040, written by some of its designers. The article agrees
with some of the advertising copy, saying "The sustained
performance level is 20 VAX-equivalent MIPS and 3 Mflops at a clock
speed of 25 MHz." (1st paragraph, 4th sentence).
Later in the article, Figure 2 is particularly interesting; its caption reads
"Processor performance relative to the 68020 versus cache size
(where the 68020 equals 1)."
For the cache sizes actually used in the 68040 (4Kbytes), the
performance plotted in Figure 2 [68040 normalized to 68020] is in
the range 3.6X to 4.3X, depending upon the workload. Most of the
benchmarks shown are at 4.1X.
So, the data and the claim that 68040==20VAXmips implies that the earlier
68020 has a "sustained performance level of 4.9 VAX-equivalent MIPS"
(4.9 = 20/4.1). Does anybody seriously believe this?
About the most impartial data I could find was for the Hewlett Packard
HP9000 model 370 machine. This uses a 68030 (not 68020) at 33 MHz (not
25 MHz) and achieves a geometric mean of 3.9 SPECmarks [ref. SPEC
newsletter v1#1]. It seems reasonable to suspect the 68020 is no
better than the 030 in performance {else who'd want the 030?}, so we
conclude that the 020's performance is, at most, 3.9 VAX-equivalent
MIPS. This makes the 68040 a 16 VAXmips machine (at most), not 20 VAXmips
as advertised.
Of course the best method would be to lay hands on an actual computer
system that uses the 68040 and benchmark it; presumably Motorola
and/or NeXT and/or HP will do this someday. Prediction: the SPECmark
will be significantly below 20.0.
Disclaimer: I'm biased. Check out the SPEC newsletter and the issue
of IEEE Micro to see if I've distorted the
facts. (I assert that I haven't)
--
-- Mark Johnson
MIPS Computer Systems, 930 E. Arques M/S 2-02, Sunnyvale, CA 94086
(408) 524-8308 mark@mips.com {or ...!decwrl!mips!mark}fox@VIXEN.NSCL.MSU.EDU (07/14/90)
In article <40088@mips.mips.COM>, mark@mips.COM (Mark G. Johnson) writes... >The June 1990 issue of _IEEE_Micro_ contains an article about the >Morotola 68040, written by some of its designers. The article agrees >with some of the advertising copy, saying "The sustained >performance level is 20 VAX-equivalent MIPS and 3 Mflops at a clock >speed of 25 MHz." (1st paragraph, 4th sentence). > >Later in the article, Figure 2 is particularly interesting; its caption reads > "Processor performance relative to the 68020 versus cache size > (where the 68020 equals 1)." > >For the cache sizes actually used in the 68040 (4Kbytes), the >performance plotted in Figure 2 [68040 normalized to 68020] is in >the range 3.6X to 4.3X, depending upon the workload. Most of the >benchmarks shown are at 4.1X. > >So, the data and the claim that 68040==20VAXmips implies that the earlier >68020 has a "sustained performance level of 4.9 VAX-equivalent MIPS" >(4.9 = 20/4.1). Does anybody seriously believe this? > >About the most impartial data I could find was for the Hewlett Packard >HP9000 model 370 machine. This uses a 68030 (not 68020) at 33 MHz (not >25 MHz) and achieves a geometric mean of 3.9 SPECmarks [ref. SPEC >newsletter v1#1]. It seems reasonable to suspect the 68020 is no >better than the 030 in performance {else who'd want the 030?}, so we >conclude that the 020's performance is, at most, 3.9 VAX-equivalent >MIPS. This makes the 68040 a 16 VAXmips machine (at most), not 20 VAXmips >as advertised. For much of the stuff we run, 16Mhz 68020 + 68881 runs at about .8 VAX mips. (Scientific technical applications). So it seems to me that 25Mhz could be no faster than 25/16 * .8 = 1.25 VAX mips. Ron Fox | FOX@MSUNSCL.BITNET | Where the name NSCL | FOX@CYCVAX.NSCL.MSU.EDU | goes on before Michigan State University | MSUHEP::CYCVAX::FOX | the quality East Lansing, MI 48824-1321 | | goes in. USA
uppal@hpindda.HP.COM (Sanjay Uppal) (07/14/90)
Huh?? You are comparing apples to oranges: SPECmarks and VMIPS are not the same at all. HP 9000/350 (68020 @ 25 Mhz) is rated at 5.4 VMIPS and 1.7 SPECmarks. If (crudely) the 040 is 4.1X this machine it will rate at 22 VMIPS and 7 SPECmarks. Sanjay Uppal NN9T phone (408) 447-3864 Hewlett-Packard (IND) uucp: ...!hplabs!hpda!uppal NS arpa: uppal%hpda@hplabs.hp.com
rwallace@vax1.tcd.ie (07/16/90)
In article <1990Jul13.163849.4282@msuinfo.cl.msu.edu>, fox@VIXEN.NSCL.MSU.EDU writes: > > For much of the stuff we run, 16Mhz 68020 + 68881 runs at about .8 VAX > mips. (Scientific technical applications). So it seems to me that > 25Mhz could be no faster than 25/16 * .8 = 1.25 VAX mips. Sounds like either we don't agree on what VAX MIPS mean or there's something seriously wrong with your system or benchmarks. The 68000 at 8MHz is rated at about .8 MIPS. The 68020 at 16MHz should run about 3-4 MIPS (factor of 2 for the clock rate, factor of 2-3 for the better design). OK, processors run below rated performance with depressing frequency for reasons like not enough cache memory, badly written code etc. but by a factor of 4??? Are you sure you weren't running a benchmark that proves your compiler produces code that only runs 25% as fast as it should? Russell Wallace, Trinity College, Dublin rwallace@vax1.tcd.ie "To summarize the summary of the summary: people are a problem"
sritacco@hpdmd48boi.hp.com (Steve Ritacco) (07/16/90)
Some machines SPEC numbers and MIPS ratings are almost the same. Remember that the VAX 11/780 has a SPEC mark of 1. I'm not suggesting that some companies misrepresent their products. Nooooooooooo! "of course I don't speak for HP, I can't even speak for myself"
tif@doorstop.austin.ibm.com (Paul Chamberlain) (07/17/90)
In article <14900009@hpdmd48boi.hp.com> sritacco@hpdmd48boi.hp.com (Steve Ritacco) writes: >Some machines SPEC numbers and MIPS ratings are almost the same. >Remember that the VAX 11/780 has a SPEC mark of 1. I would be tempted to conclude that the relationship of MIPS to SPEC is not a linear one (for example MIPS ?= SPEC^2). Might this resolve the differences? Paul Chamberlain | I do NOT represent IBM tif@doorstop, sc30661@ausvm6 512/838-7008 | ...!cs.utexas.edu!ibmaus!auschs!doorstop.austin.ibm.com!tif
khb@chiba.Eng.Sun.COM (Keith Bierman - SPD Advanced Languages) (07/17/90)
In article <14900009@hpdmd48boi.hp.com> sritacco@hpdmd48boi.hp.com (Steve Ritacco) writes:
Some machines SPEC numbers and MIPS ratings are almost the same.
Remember that the VAX 11/780 has a SPEC mark of 1.
...
This is true by construction. The speed of a VAX 11/780 running the
SPECmark suite is defined to be 1 SPECmark.
It would be amazing if other machine SPECmarks were closely correlated
... it would mean that one would not need a SPEC suite ....
--
Keith H. Bierman |*My thoughts are my own. !! kbierman@Eng.Sun.COM
It's Not My Fault | MTS --Only my work belongs to Sun* khb@chiba.Eng.Sun.COM
I Voted for Bill & | Advanced Languages/Floating Point Group (415 336 2648)
Opus<khb@eng.sun.com> "When the going gets Weird .. the Weird turn PRO"bartho@obs.unige.ch (PAUL BARTHOLDI, OBSERVATOIRE DE GENEVE) (07/17/90)
In article <6535.26a0e67f@vax1.tcd.ie>, rwallace@vax1.tcd.ie writes: > In article <1990Jul13.163849.4282@msuinfo.cl.msu.edu>, fox@VIXEN.NSCL.MSU.EDU writes: >> >> For much of the stuff we run, 16Mhz 68020 + 68881 runs at about .8 VAX >> mips. (Scientific technical applications). So it seems to me that >> 25Mhz could be no faster than 25/16 * .8 = 1.25 VAX mips. > > Sounds like either we don't agree on what VAX MIPS mean or there's something > seriously wrong with your system or benchmarks. The 68000 at 8MHz is rated at > about .8 MIPS. The 68020 at 16MHz should run about 3-4 MIPS (factor of 2 for > the clock rate, factor of 2-3 for the better design). OK, processors run below > rated performance with depressing frequency for reasons like not enough cache > memory, badly written code etc. but by a factor of 4??? Are you sure you > weren't running a benchmark that proves your compiler produces code that only > runs 25% as fast as it should? > > Russell Wallace, Trinity College, Dublin > rwallace@vax1.tcd.ie > "To summarize the summary of the summary: people are a problem" We have a vax 780 with FPA (== 1 MIPS) and sun 3/60 at 16 MHz with 68881, that is the same configuration as above. All my comparisions show the sun to be between .5 and .8 of the vax for pure computational tasks (no io). One exception, TeX runs about twice as fast on the sun ... why ? In all cases, I used optimization, with the fortran compiler and standard libraries (TeX is written in pascal or C), on real problems we have run on many other machines too. So Russells data are correct for me. Who has a 68020/68881 that runs 3-4 times faster than a 780 ? Paul Bartholdi, Geneva Observatory
mslater@cup.portal.com (Michael Z Slater) (07/17/90)
Speaking of 68040 performance, has anyone on the net actually seen one running? I'm trying to collect some data on the current status of the chip. Michael Slater, Microprocessor Report mslater@cup.portal.com 707/823-4004 fax: 707/823-0504
daveh@cbmvax.commodore.com (Dave Haynie) (07/18/90)
In article <14900009@hpdmd48boi.hp.com> sritacco@hpdmd48boi.hp.com (Steve Ritacco) writes: >Some machines SPEC numbers and MIPS ratings are almost the same. >Remember that the VAX 11/780 has a SPEC mark of 1. However, the classic "MIPS" rating, which gradually became standardized as VAX 11/780 MIPS, doesn't include floating point performance. SPECmarks contain quite a bit of floating point information. Even if the SPECmarks, when designed, were scaled to make a VAX 11/780 equal to 1 at some point (eg, an impossible task given compiler/OS variations), that is still not the same as permanently equating VAX MIPS and SPECmark ratings. Certainly the SPECmark is a better number for overall machine performance, though as I understand it, the reason that all the SPEC benchmarks are quoted in a report, as well as the composite that gives you a SPEC number, is that no one considers the single SPECmark number to be all-telling. It is also meaningless to quote a "SPECmark for the 68040", since that's a system benchmark. Certainly "SPECmark for the VAX 11/780" and "SPECmark for the HP 9000/360" are meaningful numbers, at least as far as these things go. -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "I have been given the freedom to do as I see fit" -REM
gerry@zds-ux.UUCP (Gerry Gleason) (07/18/90)
In article <713@obs.unige.ch> bartho@obs.unige.ch (PAUL BARTHOLDI, OBSERVATOIRE DE GENEVE) writes: >In article <6535.26a0e67f@vax1.tcd.ie>, rwallace@vax1.tcd.ie writes: >> In article <1990Jul13.163849.4282@msuinfo.cl.msu.edu>, fox@VIXEN.NSCL.MSU.EDU writes: >>> >>> For much of the stuff we run, 16Mhz 68020 + 68881 runs at about .8 VAX >>> mips. (Scientific technical applications). So it seems to me that >>> 25Mhz could be no faster than 25/16 * .8 = 1.25 VAX mips. >> Sounds like either we don't agree on what VAX MIPS mean or there's something >> seriously wrong with your system or benchmarks. The 68000 at 8MHz is rated at I'd go with confusion over what "MIPS" measures. >We have a vax 780 with FPA (== 1 MIPS) and sun 3/60 at 16 MHz with 68881, that >is the same configuration as above. All my comparisions show the sun to be >between .5 and .8 of the vax for pure computational tasks (no io). One MIPS is a measure of integer CPU performance, so I wonder why people are being so careful to tell us about the floating point hardware. If you want to talk floating point, the metric is MFLOPS, not MIPS. Gerry Gleason
paulr@mips.COM (Paul Richardson) (07/18/90)
And do not forget that the Vax 11/780 is technically not a 1 mips machine.
The clock cycle is 200ns.I guess assuming a performance rating of unity
makes life easier for the marketing types to synthesize other more 'useful'
performance numbers :)
--
Peace,
/pgr
"Most of my heroes don't appear on no stamps" - Public Enemy (Chuck D.)
{ames,prls,pyramid,decwrl}!mips!paulr or paulr@mips.comdaveh@cbmvax.commodore.com (Dave Haynie) (07/19/90)
In article <40231@mips.mips.COM> paulr@mips.COM (Paul Richardson) writes: >And do not forget that the Vax 11/780 is technically not a 1 mips machine. >The clock cycle is 200ns. Well, if the critter did one instruction per clock cycle, like a RISC machine, it would only have to have a 1000ns clock cycle to be a 1 MIPS machine. Obviously it takes multiple clocks for the average VAX instruction anyway, but the bigger issue is what, exactly, constitutes an instruction in the first place. I mean, the Transputer people have been claiming 17MIPs for years, yet running real code a single T800 does integer stuff about as fast as a 68020. So for real comparisons, the further away from the concept of MIPS you get, the better. Strangely enough, the marketing folks at most companies still quote MIPS and Dhrystone 1.1 figures, since they produce larger and more amazing numbers than SPECmarks. "These go to 11", etc. >Peace, >/pgr And I'm sitting here typing this on a 25MHz 68030 based machine name "kahuna", telnetted to a Mips-based computer sold by DEC, named "cbmvax". All these MIPS flying around, and a Commodore C64 could probably get this job done... -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "I have been given the freedom to do as I see fit" -REM
johnl@esegue.segue.boston.ma.us (John R. Levine) (07/19/90)
In article <40231@mips.mips.COM> paulr@mips.COM (Paul Richardson) writes: >And do not forget that the Vax 11/780 is technically not a 1 mips machine. >The clock cycle is 200ns. Depends what you mean by MIPS, other than the accurate Meaningless Index of Processor Speed. It's true, a 780 typically executes about 600K instructions per second. The 1 MIPS figure came about when someone observed that the 780 is about as fast as an IBM mainframe (370/158?) that was rated 1 MIPS. Evidently it accomplished somewhat more work per instruction than a 370 did. IBM mainframes have been rated in MIPS for 25 years. IBM 360/370 MIPS are somewhat more meaningful than others since the comparison is among machines that implement the same architecture and at least at first was calibrated to reality, i.e. a 1 MIPS 360 was one that actually executed a million instructions per second in some instruction mix. It was also easier to do such benchmarks on 360s since pesky things like caches and restartable string moves didn't mess up the measurements. -- John R. Levine, Segue Software, POB 349, Cambridge MA 02238, +1 617 864 9650 johnl@esegue.segue.boston.ma.us, {ima|lotus|spdcc}!esegue!johnl Marlon Brando and Doris Day were born on the same day.
aburto@marlin.NOSC.MIL (Alfred A. Aburto) (07/20/90)
In article <713@obs.unige.ch> bartho@obs.unige.ch (PAUL BARTHOLDI, OBSERVATOIRE DE GENEVE) writes: >In article <6535.26a0e67f@vax1.tcd.ie>, rwallace@vax1.tcd.ie writes: >> In article <1990Jul13.163849.4282@msuinfo.cl.msu.edu>, fox@VIXEN.NSCL.MSU.EDU writes: >>> >>> For much of the stuff we run, 16Mhz 68020 + 68881 runs at about .8 VAX >>> mips. (Scientific technical applications). So it seems to me that >>> 25Mhz could be no faster than 25/16 * .8 = 1.25 VAX mips. >> >> Sounds like either we don't agree on what VAX MIPS mean or there's something >> seriously wrong with your system or benchmarks. The 68000 at 8MHz is rated at >> about .8 MIPS. The 68020 at 16MHz should run about 3-4 MIPS (factor of 2 for >> the clock rate, factor of 2-3 for the better design). OK, processors run below >> rated performance with depressing frequency for reasons like not enough cache >> memory, badly written code etc. but by a factor of 4??? Are you sure you >> weren't running a benchmark that proves your compiler produces code that only >> runs 25% as fast as it should? >> >> Russell Wallace, Trinity College, Dublin >> rwallace@vax1.tcd.ie >> "To summarize the summary of the summary: people are a problem" > >We have a vax 780 with FPA (== 1 MIPS) and sun 3/60 at 16 MHz with 68881, that >is the same configuration as above. All my comparisions show the sun to be >between .5 and .8 of the vax for pure computational tasks (no io). One >exception, TeX runs about twice as fast on the sun ... why ? In all cases, >I used optimization, with the fortran compiler and standard libraries (TeX >is written in pascal or C), on real problems we have run on many other machines >too. So Russells data are correct for me. Who has a 68020/68881 that runs >3-4 times faster than a 780 ? > > Paul Bartholdi, Geneva Observatory ----------- I noticed in IBM advertisements that 1 MIPS relative to the VAX-11/780 was based upon the VAX-11/780 doing 1757 (?) Dhrystones/sec. This was Version 1.1 of the Dhrystone. I don't know what particular compiler or what degree of optimizations were used (ahem, just minor points of course :-)). One VAX-11/780 may be different from another too ... In any event IF we ASSUME the 1757 Dhrys/sec is accurate for a 1 MIPS VAX-11/780 reference then we can go on and derive other possibly equally meaningless numbers (based upon other compilers with various sorts of optimizations applied). Well, I know from personal measurements that the Amiga with 25 MHz 68030 and fast nibble mode dram (with burst mode on) can do about 8200 Dhrys/sec (this is Dhrystone V2.1 which is said to be less optimizable than version V1.1 :-), but otherwise it is the same benchmark). These results indicate that the 68030 at 25 MHz is a 8200/1757 = 4.6 MIPS machine. ------------ Actually 5 MIPS for the 25 MHz 68030 is not all that unreasonable. Afterall most 68030 instructions execute in 3 Clock Cycles (implies 8.3 MIPS peak operation). Also if one operates in "synchronous" mode then many instructions execute in 2 CC's (implies 12.5 MIPS peak performance). However if we could assume (there I go again) typically 5 CC's per 68030 instruction then this implies a 5 MIPS machine. This seems very reasonable to me, but I have no hard data to examine ..... Al Aburto aburto@marlin.nosc.mil
jiml@uwslh.slh.wisc.edu (James E. Leinweber) (07/24/90)
paulr@mips.COM (Paul Richardson) writes: >And do not forget that the Vax 11/780 is technically not a 1 mips machine. Sigh. I don't have hard data on the relative merits of Vax, IBM and other MIPS, but I first started seeing this canard when some trade press idiot divided the clock rate by the average cycle count of the instruction set. This is, of course, totally bogus. You have to pay attention to the pipeline depth and weight the cycle counts according to relative frequency of use, interlock delays, etc. In short, on something the complexity of a Vax, you have to run a performance benchmark. I haven't kept up with the SPEC ratings of IBM versus DEC boxes; would someone care to follow-up with real numbers? Jim Leinweber (608)262-0736 State Lab. of Hygiene/U. of Wisconsin - Madison jiml@sente.slh.wisc.edu uunet!uwvax!uwslh!jiml fax:(608)262-3257 -- Jim Leinweber (608)262-0736 State Lab. of Hygiene/U. of Wisconsin - Madison jiml@sente.slh.wisc.edu uunet!uwvax!uwslh!jiml fax:(608)262-3257