Power PC performance comparisons
- To: mathgroup at yoda.physics.unc.edu
- Subject: Power PC performance comparisons
- From: purswelj at vt.edu (Jerry Purswell)
- Date: Tue, 21 Jun 1994 16:54:49 -0800
Dear mathgroupers:
Several mathgroupers running Mma on the Mac platform have posted
some performance comparisons of Mma on a PowerPC and Mma on quaddras,
centris, and Mac II machines. I pulled down some of the test results and
ran the same tests using an IBM PS/Vaule point computer with an
Intel486DX-2 66 Mhz machine. The computer has 16 megs of RAM and I believe
the video is (VL) local bus.
A note about the results - Mma ver 2.2.1 under Windows requires you to set
how much memory the kernal will take before you start computations. (This
may be true for the MAC front-end too - I don't know) Whatever is left gets
used for Displaying results and such. If there is not enough actual RAM
left after the Kernal is started, Mma can use Disk Space as Virtual RAM)
Using virtual RAM tends to make stuff run more slowly. One can eliminate
the need to use virtual RAM if the Kernal is alloted smaller amounts of RAM
so that some actual RAM is left for the front-end. Consequently, I ran the
tests using kernal memory (k-m) setting of 8 megs and 16 megs. The 16 meg
case should force Windows to use some virtual RAM, but Mma did not seem to
run much more slowly this way.
The timings are for what my Windows front-end reported (not verified by
stop watch)
The IBM did better than the all the older MACs and ran surprisingly close
to the PowerPC. In one case (I don't know whether to believe this) the IBM
actually did better (See comparison 3).
The original post of the first three tests actually included 2 more tests
(loading a package and another plot of some sort). I could not get these
to execute on my PC, but if the original poster will send me suggestions as
to what might be wrong, I'll have another go at it. The execution times
shown for the MAC and PowerPC machines are simply reposts of the earlier
ones - I did not replicate them.
I'm somewhat dissappointed that the performance improvements indicated with
the Power PC. I was hoping that this Apple-IBM-Motorola combo would come
up with something that would really knock my socks off. The performance
improvements seem dissappointingly incremental. (Remember, the IBM machine
is just a mid-range PC, not a 100 Mhz pentium) Even so, the PPC did do
substantially better on the serious number crunching.....
Results follow......
Computation #1: Arithmetic - polynomial
f[x_]:= 4x-4x^2;
Nest[f,0.6,5000]
Results #1:
PowerMac 7100 601/66mhz: 5.87 seconds 1.00
NeXT 68040/25mhz: 14.84 seconds 2.53
Mac Centris 650 68040/25mhz: 18.02 seconds 3.07
Mac IIcx 68030/50mhz: 38.33 seconds 6.53
IBM PS/ValuePoint (k-m =16) 7.36 1.25
IBM PS/ValuePoint (k-m =8) 7.25 1.24
Computation #2: Arithmetic - transcendental
f[x_]:= BesselJ[0,x];
Nest[f,0.6,2500]
Results #2:
PowerMac 7100 601/66mhz: 11.28 seconds 1.00
NeXT 68040/25mhz: 25.79 seconds 2.29
Mac Centris 650 68040/25mhz: 46.88 seconds 4.16
Mac IIcx 68030/50mhz: 81.12 seconds 7.19
IBM PS/ValuePoint (k-m =16) 15.05 1.33
IBM PS/ValuePoint (k-m =8) 14.56 1.29
Computation #3: Graphics - 3D
Plot3D[Sin[x y],{x,-2,2},{y,-2,2},PlotPoints->30];
Results #3:
PowerMac 7100 601/66mhz: 6.70 seconds 1.00
NeXT 68040/25mhz: 14.84 seconds 2.21
Mac Centris 650 68040/25mhz: 11.12 seconds 1.66
Mac IIcx 50mhz: 15.32 seconds 2.29
IBM PS/ValuePoint (k-m =16) 5.22 seconds 0.78
IBM PS/ValuePoint (k-m =8) 4.78 seconds 0.71
--------------------------------------------------------------------------------
start of second Mac-PowerPC comparison post
Timing[3^10000;]
0.066667 Sec - PPC 1.00
0.4 sec on Quadra 700 6.00
0.16 sec IBM (k-M = 16) 2.40
0.16 sec IBM (k-m =8) 2.40
Timing[N[Pi,3500];]
1.3 sec on PPC 1.00
9.78 sec on Quadra 700 7.5
IBM (k-m =16) 3.35 2.58
IBM (K-m = 8) 3.29 2.53
First[Timing[Eigenvalues[Table[Random[],{200},{200}]]]]
13.216 sec on PPC 1.00
114.5 sec on Quadra 700 8.66
IBM (k-m =16) 35.76 2.71
IBM (k-m =8) 35.98 2.72
----------------------------------------------------------------------------
---------
Last two test from first post for which I could not get results from my PC
Computation #4: Package Loading
<<calcE.m (a package slightly over 200K in size)
Results #4:
PowerMac 7100 601/66mhz: 14.42 seconds 1.00
NeXT 68040/25mhz: 66.85 seconds 4.64
Mac Centris 650 68040/25mhz: 49.13 seconds 3.41
Mac IIcx 50mhz: 78.85 seconds 5.47
Computation #5: Complex 2-D Graphics and Logic
Plot[4x^2+9y^2=36,{-3,3},{-2,2},Frame,
Grid[{0.25,0.25},Fill[4x^2+9y^2=36]]
];
Results #5:
PowerMac 7100 601/66mhz: 19.88 seconds 1.00
NeXT 68040/25mhz: 36.12 seconds 1.82
Mac Centris 650 68040/25mhz: 58.18 seconds 2.93
Mac IIcx 50mhz: 103.47 seconds 5.20