Q. When Apple couldn't get faster G4s from Motorola, it
released dual processor models with the slogan "Two Brains Are
Better Than One." Is that really true, or is it just hype?
A. Multiple processor computing on the Mac OS goes back to the
quad-processor DayStar Genesis
MP 528, which was released in October 1995. At the time,
Apple's fastest model was the Power
Mac 9500/132. With four 132 MHz 604 processors and the right
software, the Genesis could smoke it.
Without the right software, three of the four CPUs spent
their time idle. Although DayStar came up with a patch for the Mac
OS to support multiple processors - and even licensed it to Apple
for some dual processor models - neither the classic Mac OS nor
most applications could take advantage of multiprocessing.
Photoshop could, and Photoshop mavens made the quad-processor
Genesis line a big hit among the graphics crowd. With faster
processors, the Genesis PM
Plus eventually offered four 233 MHz 604e processors to this
demanding market.
That all ended in August 1997, the same month Apple discontinued
the Power Mac 9600/200MP. It
wasn't that multiple processors was a bad idea, but the fact that a
single 350 MHz 604e or 266 MHz G3 offered comparable performance to
two 200 MHz 604e processors in Photoshop - and about 75% more
everywhere else.
MP Redux
Introduced in October 2000, the Power Mac G4 promised 500
MHz performance - and finally delivered it in February 2001. Come
Macworld Expo in July 2001, Motorola still
hadn't delivered a faster G4, so Apple released the "Mystic" (a.k.a. Gigabit
Ethernet) Power Mac G4, which has dual G4
processors running at 450 or 500 MHz.
This is when the "Two Brains Are Better Than One" campaign was
launched. However, the Classic Mac OS had no more dual processor support
than DayStar had offered seven years earlier. A few applications in
the graphics and video fields supported both processors, but by and
large most Mac applications and the Mac OS itself were unable to
benefit from the second CPU.
I'll let history speak for itself: When Apple replaced the
Mystic G4s with the 5-slot
"Digital Audio" models in January 2002, only one of the five
configurations offered two CPUs - and it was a middle of the line
unit running at 533 MHz vs. a single 733 MHz G4 on the top-end
Power Mac.
At least under the Classic Mac OS, unless you ran programs
designed for multiple processors, two brains weren't any
better than one.
Mac OS X
That all changed with Mac OS X, which was released in March
2001. From the BSD core through the Aqua display engine, the new
operating system was designed to support multiple processors. And
although the Quicksilver
and Quicksilver 2002 lines had
only included one dual-processor model, the Power Macintosh G4 MDD (mirrored drive
doors) contained exclusively dual processor models.
Oddly enough, the current Power
Mac G4 models have only a single CPU in the entry-level model -
and offers it at the lowest cost yet for a Power Mac G4.
But it begs the question: Are two brains really better than one?
Put another way, is today's entry-level single CPU 1 GHz Power
Mac G4 a more powerful computer than last year's entry-level dual
867 MHz Power Mac G4?
As we're fond of saying at Low End Mac, there are no easy
answers. Under the classic Mac OS a 1 GHz single processor
machine would outperform a dual processor 867 MHz machine in almost
every task - but the new Power Macs don't even boot the classic Mac
OS, making that a moot point.
Finding meaningful Mac OS X benchmarks isn't easy. Here's what
I've been able to locate:
MacSpeedZone
In a
monumental comparison dated March 13, 2003, the new
1 GHz Power Mac is 16% slower than the January 2002 dual
1 GHz model, which is their benchmark. The dual 867 MHz model
is only 2% slower than it.
In a file search, the new 1 GHz machine is 13% slower than the
dual 1 GHz model, and the dual 867 falls midway between the
two at 93%.
In MacSpeedZone's Word and Excel macro tests using
Office X, the dual 867 lags at 84-88%, while the single
processor 1 GHz model has 90-94% the performance of the dual
processor 1 GHz benchmark machine. Office X apparently
doesn't take real advantage of the second CPU.
Neither does AppleWorks, which shows similar results.
For serious number crunching, the AltiVec Fractal benchmark show
the dual 867 at the expected 87% level compared with a dual
1 GHz machine. Today's single CPU 1 GHz model comes in at
50%, and the old single CPU 867 offers half the performance of the
dual G4/867.
MacSpeedZone also tests in a multitasking setup with iTunes
ripping, AppleWorks searching and replacing, and the Finder copying
- all at the same time. The single processor G4/1 GHz scored 75% on
this test, while the dual G4/867 offers 95% the performance of
their baseline dual G4/1 GHz machine.
MacSpeedZone also runs several other benchmarks, but some of
them seem more directly tied to hard drive and CD-ROM speed than
CPU performance. Based on the multitasking results, all else being
equal it appears that a dual processor system will offer about
one-third more performance when background tasks are taking
place.
Bare Feats
In an older
report, Bare Feats finds that Final Cut Pro 3.0 performs a
"render and blur" much faster on a dual G4/533 (132 sec.) than on a
single processor G4/933 (180 sec.). Dual 800 MHz and 1 GHz
machines are faster yet. Had a dual 933 been available, it would
have completed the test in roughly 116 seconds - 35% faster than a
single processor machine.
Another set of benchmarks pits the same G4/933 against the same
dual processor machines, and even the dual G4/533 outperforms the
933 in the Photoshop, iMovie, and Cinema 4D tests. With the right
software, dual processing rocks.
Of course, we should stress here that Bare Feats tends to look
for the kind of applications that really demand G4 horsepower, most
of which also benefit greatly from dual processors.
Macworld
In the old days, Macbench 3, 4, and 5 were the standards. Not
only did Macworld publish the results, but they also made the
benchmarking software available to anyone.
No longer. For the past few years, Macworld has been using
Speedmark, their own in-house benchmark suite that seems to get
updated a few times a year, making direct comparisons between Macs
of different ages almost impossible.
Last December they gave Speedmark 3.2 scores to the original
Quicksilver 2002 dual 1 GHz Power Mac G4, the new DDR version,
and the then top-end 1.25 GHz Power Mac G4. These scores were 163,
161, and 184 respectively, and those ratings are relative to a 700
MHz eMac, which is assigned a score of 100.
At this point, Macworld has not yet published benchmark results
for the new Power Mac G4 models, nor can I find any Speedmark
results for the G4/867 dual.
A year earlier, the G4/800 dual processor had earned a Speedmark
2.1 score of 203, the single CPU G4/867 came in at 215, the G4/733
rated 183, and the older dual G4/533 managed 176. This time all
scores are relative to a 350 MHz 1999 iMac, which is assigned a
score of 100. And Speedmark 2.x was based on Mac OS 9.x, not
OS X.
A few months before that, Macworld compared the single processor
G4/450 and G4/500 with their new dual-processor replacements - also
using Speedmark 2.1. Here are the ratings: G4/450, 146; G4/500,
159; G4/450 dual, 158; G4/500 dual, 165.
Macworld has gone from having consistent, meaningful, widely
verifiable benchmarks to creating a proprietary suite of tests that
changes from time to time, making comparisons over time virtually
impossible.
Xbench
Where Macworld dropped the ball, Xbench has picked it up. Xbench
required Mac OS X 10.2 or later and reports simple scores for
memory, CPU, threading, Quartz, OpenGL, the user interface, and the
hard drive. Phil's Xbench
Benchmarks Site has a nice collection of user submitted
results.
There are some oddities. The same computer with different
amounts of memory will have different memory scores - the more RAM,
the lower the score. Quartz and OpenGL will vary with the OS,
drivers, and video card installed. And disk tests will also vary
depending on the brand and model of hard drive in the tested
computer.
Xbench's CPU score doesn't seem to benefit much from dual
processors. Comparing single- and dual-processor G4/533 machines,
the dual only scores 8% higher. But at least it gives us a standard
benchmark we can use to compare our OS X machines.
Conclusion
Are two brains better than one? Yes, but how much better is
the unanswered question. Or, more precisely, the question with no
easy answer. The brief form of this article:
Q. How much better is a dual-processor Power Mac
than a single processor one?
A. It depends.
For some tasks, dual processors can be nearly twice as fast, so
today's dual 1.42 GHz G4 Power Mac could potentially offer comparable
performance to a theoretical 2.5-2.8 GHz G4, and last year's dual 867 MHz model
could reach the levels of a 1.5-1.7 GHz G4 - if such a thing existed
today.
But for other tasks, you may see no difference at all.
One or two CPUs is only part of the equation. Other factors that
improve performance are a bigger and/or larger L2 and/or L3 cache, a faster
memory bus, whether your system has enough physical RAM or needs to
page out to virtual memory (VM) all the time, how fast your hard drive
is when VM is being used, what version of the Mac OS you're using,
whether Quartz Extreme is offloading display from the CPU to your
video card, your
video card and screen resolution, how many programs you have
active, which programs they are, and what haxies and other system
modifications may be installed on your Mac.
Mac OS X is very different from the Classic Mac OS - much more
different than you'd imagine by looking at it through its
interface. It uses memory differently. It manages resources
differently. It handles programs differently - even if one
application doesn't support multiple processors, the OS itself can
juggle other programs so they make more use of the other CPU to
balance things out.
There are no easy answers when it comes to a complex, modern
operating system like Mac OS X.
Like I said, it depends.
That said, things are likely to improve as more and more
programs are written specifically for Mac OS X instead of
simply ported over from the Classic Mac OS or other platforms - and
also tuned for multiple processors. On average, we could see dual
processor models offering 30% to 50% more performance
overall than a single processor machine. Over time, that should get
even better. It's mostly up to the developers to offer MP support
in their software.
Real World Comparison
If I were buying an entry-level Power Mac G4 today, I'd
definitely pick the recently discontinued dual G4/867 over the newer single processor G4/1 GHz machine.
Both have a 133 MHz system bus, 1 MB of L3 cache per CPU,
Nvidia GeForce 4 graphics, 256 MB of RAM, and a 60 GB hard drive, but the dual
867 also has the ability to boot into OS 9 if/when you need to.
Both are currently selling for about US$1,500, but when running OS X, overall
performance (thanks to the second CPU) should more than offset the 13%
slower clock speed, perhaps offering 15-20%
more power on average - and up to 70% more in some instances.
For the same price, I'd call it a no brainer and go for the
older dual 867.
Too bad Apple doesn't offer dual processor PowerBooks....
Update: One further benefit of dual-processor Macs comes in an
unexpected place - Classic Mac OS performance. With two CPUs, the
Classic Environment has full access to one CPU while OS X
uses the other CPU to handle keyboard and mouse input, graphics,
network access, disk reads and writes, etc. This means that the
CPU handling the Classic Mac OS has less to do than it does when
booted natively. An unexpected bonus for those of us who still
use Classic Mode!
