ali@navajo.UUCP (02/06/87)
---
I just downloaded the dhrystone posting from comp.sys.misc and tried it out
on the Amiga. The dhrystone info for the Amiga dated to about a year ago,
and was for the Manx 2.30a compiler... (Was there ever such a beast? Or does
that mean 3.20a?) Also the OS was probably 1.1, and I was curious what
values I would get with Manx 3.30e and 1.2 OS.
Here is the old info:
Dhrystones
System, OS, Compiler NoReg / WithReg
------------------------------------------------------------
Amiga 1.1, Manx 2.30a (?), 32-bit ints 643 / 684
Amiga 1.1, Manx 2.30a (?), 16-bit ints 880 / 915
Here's what I got:
------------------------------------------------------------
Amiga 1.2, Manx 3.30e, 16-bit ints 888 / 945 Workbench screen up
Amiga 1.2, Manx 3.30e, 16-bit ints 906 / 964 Workbench screen down
I did two tests above, and in one I pulled the WB screen way down (so
I could only see the drag bar). As you can see, that speeded up things
a tiny bit. Then I experimented with turning multitasking off:
------------------------------------------------------------
Amiga 1.2, Manx 3.30e, 16-bit ints 935 / 996 With Forbid/Permit
Amiga 1.2, Manx 3.30e, 16-bit ints 961 / 1026 With Disable/Enable
In both cases above the screen was pulled down (only 10% visible). Note
that Forbid/Permit simply assure that no other task gets to run, but
multitasking still continues (it is as if my priority is set to the max
and no other task's is). Disable/Enable, on the other hand, simply turn
of all interrupts associated with multitasking. I was first afraid to
try using Disable/Enable, but nothing tragic happened. (Actually, the
click-click from my empty second disk drive stopped for the duration
of the test!). But, some things still go on --- I was using DateStamp()
to get the time, and that still worked fine. (I guess because time is
not kept by interrupts but by the 8250s?)
Anyway, here are some other results from the dhrystone posting.
---------------------------------------------------------------
Mac 512, DeSmet C 625 / 625
Mac 512E, Lightspeed C 510 / 549
Mac Plus (7.8 MHz), Manx, 16-bit ints 714 / 769
Atari 520ST (8 MHz), Lattice 3.03 446 / 450
Atari 520ST,(8 MHz), Megamax C 1063 / 1136
One final note... I was not able to get results with "fast" memory,
as I am still in the 512K world. (Soon soon to be 2 megs I hope!).
Will the program run faster if in fast memory?
I sent the above results to the dhrystone people. It would be nice
if someone out there with a Turbo Amiga (like the people who did the
Savage floating point stuff?) tried the dhrystones out and also sent in
the results.
Ali Ozer, ali@navajo.stanford.eduaburto@marlin.UUCP (02/17/87)
In article <1365@navajo.STANFORD.EDU> ali@navajo.ARPA (Ali Ozer) writes: >--- >I sent the above results to the dhrystone people. It would be nice >if someone out there with a Turbo Amiga (like the people who did the >Savage floating point stuff?) tried the dhrystones out and also sent in >the results. > >Ali Ozer, ali@navajo.stanford.edu Ali, I have a Turbo-Amiga and a disk labled Drystones. I never ran the the Drystone Benchmark -- I suppose because it doesn't do floating-point tests. Anyway I'll run the Drystone program on the Turbo-Amiga and post the results. I seem to remember a result like 2200 drystones/sec which was posted by Dale Luck of CBM almost a year ago. Al Aburto
aburto@marlin.UUCP (02/20/87)
____________________________________________________________________________ Ali Ozer recently posted some Dhrystone Benchmark results for the Atari 520ST and the Amiga. He requested that someone with a Turbo-Amiga also run the Dhrystone and I said that I would, and I will --- I couldn't find my old copy of the Dhrystone program, but I'm getting another copy today. The Atari 520ST Dhrystone results were interesting to me because they show that the software can really make a huge difference in the performance of the system. There is nearly a factor of 2.5 in the performance of Lattice C V3.03 ( 450 Dhrystones/sec ) as compared to Megamax C V1.0 ( 1136 Dhrystones/sec ). Does anyone know why there is such a great variation in performance? Can we expect to see this type of performance variation on the Amiga and Turbo-Amiga? When I run the Dhrystone on the Turbo-Amiga with Lattice C and find (hopefully) that it clocks in at 2500 Dhrystones/sec am I, based on the Atari results, to expect that it really should be clocking in at 6250 Dhyrstones/sec with a 'good' compiler? I'd really like to get a handle on the real performance capabilities of the Amiga, so in addition to Ali's request would someone run the Dhrystone on the latest version of Lattice C ( and V3.03, V3.02 )? Does anyone know if there are versions of the Dhrystone program for Modula-2 and Fortran? Al Aburto _____________________________________________________________________________
ee173way@sdcc3.UUCP (02/20/87)
Regarding Dhrystones in Modula-2, I'd be glad to write the code and test it, given the specifications of the benchmark... John 7OHN
ali@navajo.UUCP (02/20/87)
In article <938@marlin.UUCP> Alfred A. Aburto writes: > The Atari 520ST Dhrystone results were interesting to me because >they show that the software can really make a huge difference in the >performance of the system. There is nearly a factor of 2.5 in the >performance of Lattice C V3.03 ( 450 Dhrystones/sec ) as compared to >Megamax C V1.0 ( 1136 Dhrystones/sec ). I think (from what I gather from the comments) one of the purposes of the Dhrystone program is to benchmark the code generated by the compiler as well as the machine itself. The Dhrystone results have to be posted with the compiler that the code was generated by --- So you can't just say "The Amiga benchmarks at 1000 dhrys/sec while this other machine benchmarks at foo," without mentioning what compiler you used, because, as you observed, the compiler plays a great role in the result. > Does anyone know if there are versions of the Dhrystone program >for Modula-2 and Fortran? No, Dhrystones come only for C. (Actually, the version I have was apparently translated from Ada... Hmmm.) > Al Aburto Thanks for the effort in getting the results on a Turbo. By the way, do you have a Manx compiler? If not, I can send you or post the executable (if I figure out how...) for the Dhrystone program I used. (It already includes the calls to DateStamp(), etc, for timing.) Ali
aburto@marlin.UUCP (02/21/87)
Keywords:Dhrystone,Turbo-Amiga,Amiga In article <936@marlin.UUCP> aburto@marlin.nosc.mil.UUCP (Alfred A. Aburto) writes: >In article <1365@navajo.STANFORD.EDU> ali@navajo.ARPA (Ali Ozer) writes: >>--- >>I sent the above results to the dhrystone people. It would be nice >>if someone out there with a Turbo Amiga (like the people who did the >>Savage floating point stuff?) tried the dhrystones out and also sent in >>the results. >> >>Ali Ozer, ali@navajo.stanford.edu ____________________________________________________________________________ I ran the Dhrystone Program last night on the Amiga and Turbo-Amiga using Lattice C V3.03. I just used the 'makesimple' file in the C-DEVEL disk so I only obtained results for the NOREG case. I got a compile time warning of 'No Reference to CharLoc at line 376'. Otherwise the program compiled and linked Ok. Anyway here are the results plus some others from the Dhrystone Listing: Atari 520ST, TOS , 8.00MHz, Lattice C V3.03 , 446 / 450 Amiga , V1.2, 7.16MHz, Lattice C V3.03 , 458 / --- Turbo-Amiga, V1.2, 7.16MHz, Lattice C V3.03 (020/16-bit mem) , 574 / --- Amiga , V1.2, 7.16MHz, Manx Aztec C V3.30E , 888 / 945 Atari 520ST, TOS , 8.00MHz, Megamax C V1.0 , 1063 / 1136 Turbo-Amiga, V1.2, 7.16MHz, Manx C V3.30E (020/16-bit mem) , ???? / ???? Turbo-Amiga, V1.2, 14.32MHz, Lattice C V3.03 (020/32-bit mem) , 1585 / --- Turbo-Amiga, V1.2, 14.32MHz, Manx C V3.30E (020/32-bit mem) , ???? / ???? Well the results seem to be telling me that I ought to be using Manx Aztec C with the Amiga instead of Lattice C. Manx Aztec C V3.30E appears to be somewhat more than a factor of 2 more efficient in running the Dhrystone than Lattice C V3.03! I didn't run Manx Aztec C V3.30E with the Turbo-Amiga but it looks like its going to do a pretty darn good job! It looks like the Dhrystone is more useful as an indicator of software performance than system performance or hardware performance. Apparently software performance is too great of an unknown variable and the Dhrystone should not be used (except with great caution) to compare computer system or hardware performance. I hope someone with Manx Aztec C and a Turbo-Amiga can run the Dhrystone too ( Jere Marrs ? ). Al Aburto
ee173way@sdcc3.UUCP (02/22/87)
I seem to remember someone posting Turbo Amiga dhrystone benchmarks at something like 2745. I also remember the article saying that was faster than a vax11/750. (The article may have been in the Best of Bix, Byte Magazine). (Manx was the compiler). Doesn't anyone care to see what Modula-2 does? Latest version is faster than Aztec (without register vars) doing the sieve... John 7OHN
aburto@marlin.UUCP (02/23/87)
In article <3761@sdcc3.ucsd.EDU> ee173way@sdcc3.ucsd.edu.UUCP (John Schultz) writes: > > I seem to remember someone posting Turbo Amiga dhrystone >benchmarks at something like 2745. I also remember the article >saying that was faster than a vax11/750. (The article may have been >in the Best of Bix, Byte Magazine). (Manx was the compiler). >Doesn't anyone care to see what Modula-2 does? Latest version is >faster than Aztec (without register vars) doing the sieve... Go for it John. Manx Aztec C V3.30E on the Turbo-Amiga ran at 2815 Dhrystones/sec (NOREG) and 2978 Dhrystones/sec (REG) --- I just ran these last night. AL Aburto
aburto@marlin.UUCP (02/24/87)
In article <1395@navajo.STANFORD.EDU> ali@navajo.UUCP (Ali Ozer) writes: >In article <938@marlin.UUCP> Alfred A. Aburto writes: >> The Atari 520ST Dhrystone results were interesting to me because >>they show that the software can really make a huge difference in the >>performance of the system. There is nearly a factor of 2.5 in the >>performance of Lattice C V3.03 ( 450 Dhrystones/sec ) as compared to >>Megamax C V1.0 ( 1136 Dhrystones/sec ). >Thanks for the effort in getting the results on a Turbo. By the way, do you >have a Manx compiler? If not, I can send you or post the executable (if >I figure out how...) for the Dhrystone program I used. (It already includes >the calls to DateStamp(), etc, for timing.) > >Ali I did get Manx Aztec C V3.30E to run the Dhrystone on the Turbo-Amiga. Here are the results(16-bit int): Amiga ,68000 , 7.16MHz, 16-bit memory , 1006 / 1082 Turbo-Amiga ,68020 , 7.16MHz, 16-bit memory , 1064 / 1151 Turbo-Amiga ,68020 , 14.32MHz, 32-bit memory , 2815 / 2978 You know when the Dhrystone Program (I'm not calling it a Benchmark anymore) finishes with a run it prints out something like the following: "This Machine Benchmarks at 6000 Dhrystones/second". Well this is not accurate. A more accurate output would be: "This System Executes 6000 Dhrystones/second" "CAUTION: This System with a different compiler" "may produce an entirely different result." Well we just shot ourselves in the foot, because the next question that can be asked is: 'Well then of what use is the Dhrystone?'. It is useful for the comparison of software(compiler) performance on the same or very similar machines as we saw for the Amiga/Lattice C/Manx Aztec C results. To go further than this and compare one machine or system against another without accounting in some manner for the variance in software performance is pretty shaky if not meaningless as some 'soldiers of the benchmark wars' have pointed out to me. Al Aburto
alex@xios.UUCP (Alex B Laney) (02/24/87)
In article <939@marlin.UUCP> aburto@marlin.nosc.mil.UUCP (Alfred A. Aburto) writes: >Keywords:Dhrystone,Turbo-Amiga,Amiga > >In article <936@marlin.UUCP> aburto@marlin.nosc.mil.UUCP (Alfred A. Aburto) writes: > >Atari 520ST, TOS , 8.00MHz, Lattice C V3.03 , 446 / 450 >Amiga , V1.2, 7.16MHz, Lattice C V3.03 , 458 / --- >Turbo-Amiga, V1.2, 7.16MHz, Lattice C V3.03 (020/16-bit mem) , 574 / --- >Amiga , V1.2, 7.16MHz, Manx Aztec C V3.30E , 888 / 945 >Atari 520ST, TOS , 8.00MHz, Megamax C V1.0 , 1063 / 1136 >Turbo-Amiga, V1.2, 7.16MHz, Manx C V3.30E (020/16-bit mem) , ???? / ???? >Turbo-Amiga, V1.2, 14.32MHz, Lattice C V3.03 (020/32-bit mem) , 1585 / --- >Turbo-Amiga, V1.2, 14.32MHz, Manx C V3.30E (020/32-bit mem) , ???? / ???? > This value shows that the Lattice V3.03 isn't a speed demon! (Get to work on that optimizer!) As a comparison, the results on a CT MightyFrame (68020 @ 12 MHz) are about 2200 to 2500 (don't have the exact number in front of me) I do think the CT though has fast memory/data paths and such. Also, I ran the benchmark without 68881. Still, those results make it tempting to get a Turbo Amiga! Considering that a Mighty Frame costs about 10 times as much as a regular Amiga or 5 times as much with Turbo Amiga. ( I can't say exactly - considering configurables). Or a Sun! And it runs Unix SVR2 really well, and SVR3 soon. Although it does come with hard disk and cartridge tape drive, etc. But no graphics built in. -- : : ::: ::: ::: : : : : :.. Alex Laney, Xios Systems Corp, 150-1600 Carling Av, Ottawa. : : : : : {utcsri,utzoo}!dciem!nrcaer!xios!alex alex@xios.UUCP : : ::: ::: ::: (... whatever gets you thru the net)
eric@amc.UUCP (02/25/87)
In article <941@marlin.UUCP> aburto@marlin.nosc.mil.UUCP (Alfred A. Aburto) writes: > >Amiga ,68000 , 7.16MHz, 16-bit memory , 1006 / 1082 Amiga ,68010 , 7.16MHz, 16-bit memory , ???? / ???? >Turbo-Amiga ,68020 , 7.16MHz, 16-bit memory , 1064 / 1151 >Turbo-Amiga ,68020 , 14.32MHz, 32-bit memory , 2815 / 2978 How about it folks? The '010 should help a little, but does it? If someone will send me the Dhrystone sources, I'll return a report. >"CAUTION: This System with a different compiler" >"may produce an entirely different result." Amen! Eric McRae Engineering Fellow, Applied Microsystems Corporation UUCP: ..uw-beaver!tikal!amc!eric ATT: (206) 882-2000 USNAIL: PO Box 97002 Redmond, WA 98073-9702 MISSILES: 122 8 27 W / 47 39 14 N
johnl@tw-rnd.SanDiego.NCR.COM (John Lindwall) (11/15/89)
Has anyone performed the Dhrystone benchmark on the various models of Amigae? I am interested in receiving results for the following systems: Amiga 1000 Amiga 500 Amiga 2000 - (standard 7.14 MHz 68000) Amiga 2500 - (14.32 MHz 68020, 68881, 2 meg 32 bit RAM,right?) Amiga 2000 w/GVP 25Mhz 68030/68882 + 32 bit RAM. I'm sure the three lowend systems 500, 1000, 2000 will have the same rating, but it would be nice to run the benchmarks anyway. What is the meaningful result of the Dhrystone benchmark? Is a rating of X Dhrystones mean that the system under test averaged X instruction executions per second? Thank you! -- John Lindwall | "Not my employer opinions; mine" johnl@tw-rnd.SanDiego.NCR.COM | a man, a plan, a beer, reeban alpa nama ----------------------------------+--------------------------------------------