Friday, August 3, 2007

Macintosh Performance Comparison


We published a detailed Mac performance comparison about a year ago, analyzing the strengths and weaknesses among eMac, iMac G5, PowerBook and Mac Mini systems, running Mac OS X 10.3 "Panther."

Apple is now changing the entire Macintosh computer architecture with its switch to Intel, and the current Mac OS X 10.4 "Tiger" operating system incorporates substantial changes from "Panther."
With the first Intel Macs available to customers, and the new Universal Binary (or Intel-specific) programs they require, the time seemed right for a new performance comparison with new benchmarks, as we evaluate PowerPC vs. Intel Macs and compare various Mac models, from laptops to power towers.

Evaluation and comparison is an on-going process, of course, and we'll be updating and refining our analyses over time, with a goal of using simple, standardized tests that can be reproduced easily by others without special applications or tricky procedures.

We also want to thank MacInTouch readers who have kindly contributed test results, as it's impossible for us to test every last Mac model and configuration with limited resources.
We hope the information we've generated and gathered will be useful.


The following list describes the details of each computer configuration tested, with a short name we use in the results tables below.

Mini G4

Mac Mini PowerPC G4/1.5GHz "stealth edition" [i.e. unannounced upgrade], purchased Dec. 2005, 512MB DDR SDRAM (1 x 512MB), 80GB ST9808211A hard drive, MATSHITACD-RW CW-8124 optical drive, ATI Radeon 9200 graphics w/ 64MB VRAM

Mini Solo

Standard Mac Mini Core Solo 1.5GHz, 512MB DDR2 SDRAM (2 x 256MB); 60GB, 5400RPM Seagate Momentus 5400.2 SATA drive (ST96812AS), MATSHITACD-RW CW-8124 "Combo" optical drive, Intel GMA 950 grapics, Mac OS X 10.4.5 (8H1619)


15" PowerBook G4/1.67GHz (pre-HD model), 1 GB memory (2x512MB PC2700U-25330), Fujitsu MHT2080AH 80GB, 5400RPM hard drive, ATI Mobility Radeon 9700 graphics with 64 MB video RAM.


MacBook Pro/2.16GHz, ATI Radeon X1600, Toshiba MK1234GSX 5200RPM hard drive, 1GB memory,
[Bajune Espiritu]

iMac Core Duo Maxtor

20" iMac Core Duo/2.0GHz, 1.5GB memory (1MB+512MB), 128 MB video RAM, Maxtor 6L250M0 hard drive.

iMac Core Duo 2GB

20" iMac Core Duo/2.0GHz, 2GB memory (1MB+1MB), WDC WD2500JS-40NGB2 hard drive, Matsushita UJ-846 optical drive, ATI Radeon X1600 w/ 256 MB VRAM,

iMac G5 Rev. A

20" iMac G5/1.8GHz Rev. A, 1GB of memory (2 x 512MB), Seagate ST3160023AS hard drive, Nvidia GeForce FX 5200 Ultra graphics hardware with 64 MBytes of video memory,

iMac G5 iSight

20" iMac (iSight)/2.1GHz, 1.5 GB memory (512MB+1GB), ATI Radeon X600 XT, 128 MB VRAM, Maxtor 6L250M0 hard drive.

G5 Quad

Power Mac G5 Quad/2.5GHz, NVIDIA GeForce 6600 w/ 256MB VRAM, WDC WD2500JS-41MVB1 drive, 2.5GB RAM (1GBx2 + 256MBx2)
[Rich Cruse]

Power Mac G5/RAID

Power Mac G5/1.8GHz single-processor, 1.25 GB memory (512x2 + 128x2), GeForce FX 5200, 64 MB VRAM, RAID-0 array with dual Seagate ST3160827AS drives. Mac OS X 10.3.9.


7200RPM Hitachi 7k100 in 1.5GHz 12" PowerBook G4


Maxon's Cinebench is a nice benchmark. We've found it consistent, straightforward and easy to use, measuring CPU and graphics performance. MacInTouch readers have already contributed a great collection of test results for various Mac models (and a few PCs, as well).

Note that changes in Apple's operating system graphics routines, or in the firmware of Mac graphics cards, can affect results.

Higher numbers, indicating faster processing rates, are better.

Cinebench 9.5 Processing Rates

MacBook PowerBook Mini
Core Duo
iMac G5
Power Mac
G5 Quad
Rendering 329 170 155 212 308 248 385
608 n/a n/a n/a 558 n/a 1180
1.85 n/a n/a n/a 1.81 n/a 3.07
Shading 371 180 163 256 357 260 386
831 498 427 869 981 730 1210
1427 889 530 438 1687 1075 2164

File Duplication

A simple Finder file copy has proven to be a very consistent and useful test of disk performance over the years. We create a folder containing ten copies of a 20MB file and duplicate that folder in the Finder, using a stopwatch to time how long it takes to complete the copy in seconds. (Changes in the Finder may have a small effect on the results; Tiger appears to be a little faster than previous Mac OS X versions.)

File Duplication Time (seconds)

MacBook Mini G4 Mini Solo iMac
Core Duo
iMac G5
Power Mac
Internal Drive 12 19 20 14 12.5 9 12 6.5
USB 2.0 27 22 35
FireWire 400 11 12 16
FireWire 800 8 n/a n/a n/a n/a 12

Disk performance comparison chart - FW400, FW800, USB2, eSATA, internal

The eSATA RAID-0 system was lightning-fast, equivalent to an internal RAID-0. A single-drive eSATA system was as fast as an internal drive, too.
(We used a 70GB Raptor boot drive inside the dual-core Power Mac, a Hitachi 250GB drive in the quad-port enclosure, and two 250GB Hitachis in the external eSATA RAID enclosure.)
We have no idea why USB 2 performance was even worse on the dual-core G5, although it's awful across the board - worse than the speed of a slow laptop drive. (We double-checked the USB2 folder copy against other disk benchmarks, which confirmed the bad USB 2 results.)

MPEG-4 Conversion

Video compression is a good test of a computer's real throughput, and we use Apple's standard QuickTime features (in the $29.95 "Pro" version) to convert a high-quality DV file to an MPEG-4 file, using default settings.

(The 50-second, 182MB video is in DV format, NTSC at 720 x 480, 29.97 fps, 48KHz stereo.)

MPEG Encoding Time

PowerBook iMac
Core Duo
iMac G5
Mini G4 Mini Solo
Time (sec.) 58 66 37 63 96


Geekbench is a new benchmarking program that's still in development, as we're compiling this comparison. We're using it primarily because we really like the approach taken by its developers; it's exactly the sort of sensible, standard test suite we'd write ourselves if we had any spare time to do that. It currently tests memory and CPU performance. (Perhaps a future version will tackle other subsystems.)

Results shown below are performance rates calibrated against a Power Mac G5/1.6GHz uniprocessor system, which scores 100 for each test, so higher numbers are better.

Geekbench Preview 2 Processing Rates

MacBook PowerBook Mini G4 Mini Solo iMac G5
Core Duo
G5 Quad
Integer Emulate6502 192 82 80 132 151 183 202
Emulate6502 4 threads 376 87 81 127 151 355 745
Blowfish 314 258 214 259 132 351 179
Blowfish 4 threads 589 216 197 226 127 608 600
bzip2Compress 138 83 86 99 126 140 168
bzip2Compress 4 threads 275 92 87 99 119 262 615
bzip2Decompress 141 82 76 92 141 139 181
bzip2Decompress 4 threads 279 80 74 85 138 267 678
Floating Point Mandelbrot 1 thread 153 79 72 108 131 148 165
Mandelbrot 4 thread 302 78 71 102 124 269 625
Memory Latency 537 85 83 547 164 495 304
ReadSequential 313 48 42 256 189 363 282
WriteSequential 143 91 90 131 141 145 251
StdlibAllocate 148 205 184 106 126 142 139
StdlibAllocate 4 threads 168 205 186 102 122 157 152
StdlibWrite 146 44 43 125 63 156 170
StdlibCopy 174 46 43 149 80 180 171
Stream StreamCopy 108 36 32 107 115 107 181
StreamScale 109 35 32 106 113 107 180
StreamAdd 162 23 20 159 117 157 189
StreamTriad 158 22 20 157 112 154 187

Java SciMark 2

Another new benchmark is Java SciMark 2 from NIST. There are a number of issues involved in benchmarking Java, and it's a little trickier to run this than some other tests, but we like its tests.

We find the simplest way to run SciMark 2.0 is by typing:


into Apple's Terminal application, then expand the window it creates before clicking the Execute Benchmark button, so you'll be able to see all six result numbers.

After it runs, select Show Table from the pop-up menu at the top, and use Command-Shift-4 to capture the six values shown for ">>Your Computer<<."

Results are shown in Mflops (Millions of floating point operations per second), so higher numbers are better than lower ones.

(The iMac G5/2.1 GHz iSight results for this test were provided by John Gerwing.)

SciMark Processing Rates (Mflops)

iMac G5
Rev. A
15" PowerBook MacBook iMac
Core Duo
iMac G5
G5 Quad
SciMark Composite 201 127 234 228 233 281
FFT 142 78 124 121 167 204
Jacobi 283 259 - 545 328 402
Monte Carlo 19 25 - 25 22 11
Sparse Matrix Multiply 283 125 - 194 325 396
LU Factorization 277 149 - 259 321 394

Xbench 1.2

Xbench has been the standard for Mac benchmarking in recent years. This shareware application is nicely packaged and easy to run, producing detailed results for many different performance parameters, as well as some helpful information about the test configuration. The new 1.2 version is a Universal Binary.

Unfortunately, Xbench results can vary widely from run to run and can be overly sensitive to small configuration details. The Overall Result and the User Interface Test have been particularly inconsistent. You also have to be careful not to compare results for different versions, which are calibrated differently (to a "standard" Mac system that changes from time to time).

We find MacBench helpful as a quick sanity check and overall performance test, as well as an indicator of potential problem areas. We use a special procedure, trying to tame its variability, rebooting immediately before running the program and averaging results for multiple runs. It's also important to run other benchmarks for perspective on Xbench numbers.

Results shown are processing rates, calibrated against a 2GHz Power Mac G5 running Tiger that scores 100 on each test, so higher numbers are better.

(The Power Mac G5 Quad results for this test were provided by Rich Cruse.)

Xbench 1.2 Processing Rates

MacBook PowerBook Mini G4 Mini Solo iMac Duo
iMac G5
Power Mac
G5 Quad
CPU 71 67 63 52 77 103 125
Thread 211 74 67 83 198 59 247
Memory 102 29 29 88 103 88 135
Quartz 54 73 59 53 68 94 138
OpenGL 140 85 77 160 138 132 131
Disk 30 35 30 37 91 93 64

No comments: