Mac IIci benchmarks

Mac IIci benchmarks

Post by Larry Hutchins » Fri, 02 Nov 1990 01:13:00



I finally had a chance to play with a Mac IIci last weekend and have
a few observations to report.

Incompatibilities:
        Stepping Out II -- cursor is invisible.  
        TMON  2.8.2 -- unreliable.  Bombs every time if use option-pwr-on
                or cmd-pwr-on when running MultiFinder (Programmer's FKey
                does not appear to be the problem -- mini-de* is ok)

Benchmarks:

You might wonder why I felt the need to run my own benchmarks rather than just
accept the results shown in the various magazines.  The answer is simple:
(a) their benchmarks frequently make no sense at all & (b) it is not
clear what their benchmarks are testing.  The following benchmark results
make (almost) perfect sense and I (pretty much) know what they are testing.

Noah Price of Apple confirmed my suspicions that the IIci is running two
wait states and that the SE/30 etc. runs one wait state.

Cynical bastards might note that the more wait states that a
machine has the more impressive a cache board becomes. This makes the
cache more desirable and perhaps a better money making opportunity for Apple.

Four of the benchmark routines consist of variations on a 4096 point
complex Fast Fourier Transform:
        FFT-SP-FPU      : single precision floating point (32 bits), uses FPU
        FFT-DP-FPU      : double precision (64 bits), uses FPU
        FFT-SP-SW       : single precision, uses software (not SANE)
        FFT-DP-SANE     : double precision, uses Apple's SANE package.

The first two test the machine's ability to perform numerically intensive
tasks. The 2nd slings around twice as much data as the first and will thus
be more affected by memory latency.  Both are highly optimized for the 882
and take maximum advantage of concurency.  These results should be highly
predictive of the performance of very well written floating point limited
applications.

The third benchmark tests the machine's integer performance (in spite of
the fact that it is performing floating point operations).  The SW is again
highly optimized and keeps memory references to an absolute minimum.  Thus
this benchmark should be predictive of the performance of well written
non-floating point limited applications.

The fourth benchmark tests the efficiency of Apple's SANE package.
*****NOTE***** It was discovered during the testing that TMON somehow
slows SANE down by a factor of two!!!  I have a dim memory of someone
mentioning this on the net a while ago.

        WHEN RUNNING BENCHMARKS INVOLVING SANE, MAKE SURE TMON IS NOT LOADED!!!!

A fifth benchmark was designed to be more sensitive to memory latency that the
others.  It is designated 'MEMORY' in the following results section.

These benchmarks do not test display oriented operations or disk access.

Results:

Time in seconds on the base IIci (internal video disabled)
        FFT-SP-FPU      : 0.88  +/- 0.02        
        FFT-DP-FPU      : 0.99  +/- 0.02        
        FFT-SP-SW       : 7.10  +/- 0.02        
        FFT-DP-SANE     : 39.47 +/- 0.02        
        MEMORY          : 2.77  +/- 0.02        

The numbers in the following table are 'xx' in the sentence:

        "The Mac IIci is xx% faster than the yy when running the zz benchmark."

Machine
                FFT-SP-FPU  FFT-DP-FPU  FFT-SP-SW  FFT-DP-SANE  MEMORY

IIci/bw1        2           2           4          4            6
IIci/color8     10          19          44         33           53
Mac II          178         167         75         82           81
Mac SE/30       56          55          50         57           48
Mac SE          5776        38901       631        863          628

Thus: The Mac IIci is 50% faster than the SE/30 when running the FFT-SP-SW
benchmark.

The numbers were calculated according to the following formula:
        (<time for yy>/<time for IIci>)*100 - 100

Thus 0% faster is the same speed and 100% faster is twice as fast.

Note: the numbers for the short tests are the average of 10 runs and
        the others utilized 5 runs except the SE at 1 run.  The IIci had
        about 3 times as much variability in run timing compared to the
        other machines.

Conditions:
        All machines: no INITs, no TMON.
        Mac IIci: 4 meg memory, system 6.0.4, NuBus video except
                IIci/bw1        -- internal monitor driver at 1 bit/pixel
                IIci/color8     -- internal monitor driver at 8 bits/pixel
        Mac II,SE/30:           -- 5 meg memory, system 6.0.3
        Mac SE                  -- 1 meg memory, system 4.2, big screen + INIT

The benchmarks utilized the graphing and data analysis application Igor.
I can provide details of their operation if anyone is interested.

Larry Hutchinson, Tektronix, Inc. PO Box 500, MS 50-383, Beaverton, OR 97077
UUCP:   [uunet|ucbvax|decvax|hplabs]!tektronix!tekgvs!larryh


 
 
 

Mac IIci benchmarks

Post by Noah Pri » Fri, 02 Nov 1990 03:48:00



>Noah Price of Apple confirmed my suspicions that the IIci is running two
>wait states and that the SE/30 etc. runs one wait state.

Actually, the IIci runs three wait states for random reads, since they're now
synchronous which means a zero wait state random read would be 2 clocks.
It's not quite that easy though, since the IIci does burst reads into the
on-chip cache which are 5 clocks for the first longword access, followed three
2 clock accesses for the other three long words in the burst.

In summary, on the Mac IIci:

    For DRAM access:  5 clock random read, 5/2/2/2 burst read, 4 clock write.
        For ROM access:   5 clock read.

        RAM and ROM accesses are synchronous.

Please note that I've directed followups to comp.sys.mac.hardware...

noah

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

..!{sun,decwrl}!apple!noah                                 Apple Computer, Inc.

 
 
 

Mac IIci benchmarks

Post by Alexis Ros » Fri, 02 Nov 1990 05:00:00


This article was very interesting but it failed in one important way: it
was testing a IIci with 4MB RAM. Thus it was all in bank A, and all subject
to cycle-stealing for video. This leads to the numbers which show 1-bit
on-board video as being slower than the NuBus video.

I'd like to see the same tests done on a 5MB and 8MB Mac IIci. My IIci is
definitely faster now that it has 5MB.

Alexis Rosen


 
 
 

Mac IIci benchmarks

Post by Larry Hutchins » Fri, 02 Nov 1990 18:54:00



>I'd like to see the same tests done on a 5MB and 8MB Mac IIci. My IIci is
>definitely faster now that it has 5MB.

Sorry, I thouht this was obvious.

Applications that are lucky enough to be loaded into the 2nd bank will run
at the same speed as the numbers I gave for the base configuration (external
video).  Apple says that applications are preferentially loaded into the
2nd bank but obviously when that fills up they are put in the 1st bank.

Larry Hutchinson, Tektronix, Inc. PO Box 500, MS 50-383, Beaverton, OR 97077
UUCP:   [uunet|ucbvax|decvax|hplabs]!tektronix!tekgvs!larryh


 
 
 

1. Mac IIci Wait states (was Re: Mac IIci benchmarks)

Sorry about that.

I was trying to keep the comparison between the IIcx and the IIci
simple -- perhaps too simple.

The bottom line is that the IIci takes an extra cycle for reads relative
to the IIcx (same as SE/30, IIx).

Noah:  Why does the IIci use synchronous access?  To improve the cache
performance perhaps?  What is the situation with a chache board installed?

The '030 user's manual says "The data burst enable bit must be set to
enable burst filling of the data cache."

Anyone know if this bit is set when running user code?

Larry Hutchinson, Tektronix, Inc. PO Box 500, MS 50-383, Beaverton, OR 97077
UUCP:   [uunet|ucbvax|decvax|hplabs]!tektronix!tekgvs!larryh


2. DirectX 3D & Borland C++ Solution !!??

3. FoxBase+/Mac 2.0, some interesting benchmark results (!)

4. Drag Manager Questions

5. Mac OS X Server Benchmarks and Boundaries

6. Windows 2000 won't recognize floppy drive

7. Benchmarks for Macs

8. Solarian II recall (balls-up in posting)

9. MS FORTRAN benchmarks for Mac II

10. Benchmarks for Macs?

11. MAC IIsi benchmarks

12. Benchmarks on the Mac line coming soon

13. Hard disk benchmarks for Mac IIfx?