We're putting together an in-house development server at Low End Mac
headquarters. The original plan was to use our dual 500 MHz 'Mystic' Power Mac
G4, max out memory to 2 GB, get a copy of OS X Server, and maybe
find a low cost CPU upgrade. We mentioned this on the LEM Swap List, and received several donations -
including a "bare bones" Power Mac.
The 'Digital Audio'
Power Mac G4 was the first Power Mac with a 133 MHz system bus, a
step up from the 100 MHz bus in earlier AGP Power Macs. This machine
started out with a single 533 MHz G4 CPU, which we've already
benchmarked. We're comparing that to a second-hand Giga Designs
Dual 7447 G-celerator with two 1.8 GHz Freescale 7447 G4 CPUs. These
CPUs each have an onboard 512 KB level 2 cache that runs at full CPU
speed. This is the same CPU used in the 1.67 GHz G4 PowerBooks.
The upgrade is theoretically capable of over 3x the performance of
this Mac's original 533 MHz CPU under the Classic Mac OS (which can
only use one CPU) and over 6x under Mac OS X, which utilizes both
processors. We ran an extensive suite of benchmark tests under Mac OS X
10.3, 10.4, and 10.5 to measure performance. (This upgrade requires a
version of Mac OS 9 or Mac OS X 10.3.5 or later, although we were
unable to boot into OS 9.)
As I began putting everything together, I began to wonder if I might
not end up with a more powerful computer than the dual 1 GHz 'Mirror Drive Door' (MDD)
Power Mac G4 that's been my workhorse machine for several years.
More on that below.
Preparation
The first step is preparing your Power Mac G4 for the upgrade. This
discontinued upgrade is compatible with Sawtooth (AGP Graphics), Mystic
(Gigabit Ethernet), Digital Audio, Quicksilver, and Quicksilver 2002
models. You first have to make sure that your Power Mac has Boot ROM
4.1.9 or later installed - choose "About This Mac" under the Apple
menu, click on the "More Info..." button, and find the Boot ROM
Version.
If you don't have version 4.1.9 or later (version 4.2.8 is strongly
recommended), you can get it from Apple. Note that you must
boot into Mac OS 9.x to update the Boot ROM. This is just another good
reason to keep a copy of OS 9 handy - preferably on your Mac's hard
drive.
The next step should be installing the 7447 Firmware Update, which
is on the CD that comes with the upgrade. This is necessary because the
7447 CPU was designed after these Power Macs, so they can't take
advantage of all it offers without a firmware update. However, the
program never finished installing itself under any version of OS X, so
we switched to the PowerLogix software, which can be freely
downloaded from OWC. It worked!
Installation
The first thing I noticed about this upgrade is its weight. It's not
large, but it is heavy. I weighed it on the postage scale, and it came
in as just under 22 oz. with it's massive copper heatsink. The CPU can
be set to several different speeds up 1.8 GHz using DIP switches,
although I have not been able to achieve that speed.
Installation of the processor upgrade isn't difficult - all you need
is a #2 Phillips screwdriver, needle-nose pliers, and a compatible Mac
with OS X 10.3.5 or later installed. Before installing the upgrade, you
attach a 2-fan cooling unit. Then unclip the heat sink from the Mac's
CPU and remove 3 screws to take out the original CPU card. This upgrade
is integrated; the heat sink isn't separate from the CPU card. The
upgrade snaps into place, and putting the retaining 3 screws in is a
breeze, unlike the last CPU upgrade I reviewed.
Once the upgrade is installed, there are two more steps: You need to
set the DIP switches for your desired CPU speed and connect additional
power to the CPU with the supplied cable.
I have done a lot of fiddling with the DIP switches - the first time
I set them to 1.8 GHz, but the CPUs ended up running at 933 MHz. Other
settings that should have resulted in higher speeds ended up at 1.33
GHz. The only setting that gave me the expected speed was 1.53 GHz with
all 5 DIP switches set to the off position. I don't know if could be
due to using the PowerLogix software instead of the Giga Designs
software, which I was unable to install.
Anyhow, the system runs very reliably at 1.53 GHz and boots quickly
into Mac OS 9.2, 10.2, 10.3, 10.4, and 10.5. It's also very fast.
Performance
My primary computer is a 'Mirror Drive Door' Power Mac G4 with dual
1 GHz 7455 CPUs, 2 GB of RAM, a 400 GB 7200 rpm Deskstar hard drive,
and its stock ATI Radeon 9000 Pro video card. Subjectively, the dual
1.8 GHz upgrade in the Digital Audio (DA) feels every bit as fast for
everyday tasks, and our Xbench test results (below) verify that.
Our testbed computer has an 80 GB 7200 rpm Deskstar hard drive, 768
MB of RAM, and it's stock ATI Rage 128 Pro AGP 2x video card. The drive
has separate partitions for Mac OS X 10.3 "Jaguar", 10.4 "Tiger", and
10.5 "Leopard". Each has been updated to the latest version of that OS,
and the only additional software we installed prior to benchmarking are
System 9.2.2 for Classic Mode and the benchmark programs we ran.
The computer is booted, and the benchmark apps are then run in
order.
User Experience
With its stock 533 MHz CPU, the DA isn't a fast computer under
OS X, but it's usable. Drop in a fast dual processor upgrade,
though, and that changes. Safari and Camino load fairly quickly,
YouTube videos run smoothly, windows fly open, and it feels about as
fast as our 1 GHz dual G4. With more RAM, it might feel even
faster.
If you have an older AGP Power Mac that you're pretty happy with but
lacks horsepower, a "brain transplant" is worth considering. I only
wish this upgrade was compatible with my Mirror Drive Door Power Mac,
as I would appreciate the extra speed on a daily basis.
The big question is the value of the upgrade: Is it worth $250 or
$600 - or somewhere in between - to upgrade your older G4 Power Mac?
After all, a CPU upgrade could cost more than the whole computer is
work. It comes down a question of how much it's going to improve your
productivity.
Other World Computing has a wide-ranging line of G4 upgrades using
single and dual 7447A (512 MB onboard cache) and 7448 (1 MB cache)
CPUs:
- 1.6 GHz 7447A, $250
- 1.7 GHz 7448, $329
- 1.8 GHz 7448, $360
- 2.0 GHz 7448, $400
- 1.6 GHz dual 7447A, $450 (Xbench 1.1.3 = 175.0)
As fast as this upgrade is on a 133 MHz system bus, I have to wonder
how much it might benefit from the 167 MHz bus in a Mirror Drive Door
G4. Too bad there's only one CPU upgrade option for my Mirror Drive
Door, a dual 1.8
GHz 7447A from Sonnet that sells for $600 - the same price as a
brand new Mac mini with its 2.0 GHz dual-core CPU.
Mac OS 9.2.2
You can't boot into the Classic Mac OS with a dual-processor upgrade
unless you remove the Multiprocessing folder from the Extensions
folder. On startup, the computer will warn of a cache memory error -
just click OK and move forward.
Speedometer 4
The Classic Mac OS only runs the slightest bit faster than Classic
Mode on this dual-processor Power Mac.
10.3.9 10.4.11 native
CPU 93.40 93.47 93.67
disk 4.85 4.92 4.83
math 3954.3 3960.0 4009.5
MacBench 5
The system was tested under Mac OS 9.2.2. Display run at 1024 x 768
and millions of colors. The disk cache was set to its default of
8 MB. Results are relative to a Power Mac G3/300, which rates
1000.
CPU math disk graphics
3714 3773 3508 n/a
These figures show that our 1.53 GHz CPU is just 3.7x as fast as a
300 MHz G3. It is also 3x the performance of my old 400 MHz PowerBook
G4 and twice the power of the stock 533 MHz CPU. All of these are less
than predicted, but it's one wicked fast OS 9 machine.
Classic Mode in OS X
With the upgrade installed, the Mac won't boot into the Classic Mac
OS - it reports a cache error and stops loading the OS. However, we can
benchmark Classic Mode in Panther and Tiger. (Update: Gene Osburn wrote
that the Multiprocessing folder must be removed from the Extensions
folder. We did that, and now it boots into Mac OS 9.2.2.)
Apple does something unexpected with Classic Mode: In many ways, Mac
OS 9.2.2 runs better inside Mac OS X than it does natively. Why?
Because OS X handles all of the disk access, networking, and
graphics, leaving OS 9 to handle the rest. Best of all, in a dual
processor system, Classic is free to take over one CPU while leaving
the other one to handle I/O in OS X.
Speedometer 4
With Classic Mode in Panther and Tiger, here are Speedometer 4
results:
10.3.9 10.4.11
CPU 93.40 93.47
disk 4.85 4.92
math 3954.3 3960.0
Speedometer scores in Classic mode are virtually identical whether
we're running Mac OS 10.3.9 or 10.4.11.
Mac OS X Benchmarks
Let 1000 Windows Bloom
The system was tested on 10 March 2009 at 533 MHz and also after
installing the dual processor upgrade.
533 MHz 1.53 GHz
OS X 10.2.8 49.7 sec.
OS X 10.3.9 40.1 sec. 39.4 sec.
OS X 10.4.11 22.4 sec. 16.1 sec.
OS X 10.5.6 44.4 sec. 32.0 sec.
As we've seen before, Mac OS X 10.4 seems to be the best platform in
terms of performance - approximately twice as fast as 10.3.9 and 10.5.6
with either CPU.
The system was retested on 6 July 2009 with the CPU upgrade running
at 1.6 GHz, 1.25 GB of RAM, a Radeon 9000 graphics card, and an booting
from an external FireWire hard drive. This benchmark ran in 14.0 sec.
under OS X 10.4.11 (a 13% improvement) and 24.3 under 10.5.7 (a 24%
improvement). The change in CPU speed can only account for 4.5% of that
difference; the rest comes from the better graphics card.
Power Fractal
Power Fractal doesn't run under Jaguar. Under Panther, this test
takes 7.0 sec. on the base computer and just 1.8 sec. with the CPU
upgrade. Under Tiger, it reports 7.9 sec. with dual 450 MHz CPUs and
2.1 sec. with the 1.8 GHz upgrade. I suspect Power Fractal is only
using one CPU.
533 MHz 1.53 GHz
OS X 10.3.9 2096 Mflops 13176 Mflops
OS X 10.4.11 2079 Mflops 13065 Mflops
OS X 10.5.6 2043 Mflops 12534 Mflops
Power Fractal seriously takes advantage of dual processors,
providing in excess of 6x the performance of the stock 533 MHz CPU. We
also see increasing system overhead as we move from Panther to Tiger to
Leopard - and corresponding small decreases in performance.
The system was tested on 6 July 2009 with a 1.6 GHz CPU, 1.25 GB of
RAM, Radeon 9000 graphics, and an external FireWire hard drive. Power
Fractal reported 13,290 MFlops under OS X 10.4.11 (1.7% better) and
under 10.5.7 (6% better). The change in CPU speed should result in a
difference of 4.5%, yet the score fall short of that under 10.4.11
while exceeding it with 10.5.7.
Xbench 1.3
Xbench 1.3 doesn't run under Jaguar, so we ran it under Panther,
Tiger, and Leopard. Xbench is calibrated to a 2.0 GHz dual Power Mac
G5, which would score 100.
- - - 533 MHz - - - - - dual 1.53 GHz - -
10.3.9 10.4.11 10.5.6 10.3.9 10.4.11 10.5.6
Overall 24.7 29.8 16.1 46.4 58.1 29.0
CPU 34.2 33.4 30.3 81.9 82.0 82.4
Threads 25.3 25.2 22.5 84.8 107.7 86.0
Memory 25.4 27.1 26.6 28.9 28.9 27.0
Quartz 29.7 29.9 20.0 27.8 57.6 32.6
OpenGL 53.0 58.4 13.7 103.7 108.1 31.3
User Int. 9.7 17.3 5.8 21.7 55.2 10.5
Drive 52.6 49.7 44.3 52.6 50.9 43.3
Under OS X 10.4.11, the overall score for the 1.53 GHz dual G4
upgrade is roughly double that of the single 533 MHz processor that
came in the computer (and over half the power of a dual 2.0 GHz G5),
but that's far from the whole story. The CPU score is 2.5 times as
high, the Threads rating roughly quadruple, Quartz and OpenGL are
almost twice as fast, and User Interface is over 3x as fast. Drive
scores are barely changed at all.
It's interesting that the Threads, Quartz, and User Interface scores
are highest under Tiger, yet Panther has a small edge in hard drive
performance. Threads and Memory are higher with dual processors under
Tiger vs. Panther but lower with the single CPU. Leopard doesn't do so
well with half the overall score of Tiger.
By way of comparison, my dual 1 GHz "Mirror Drive Door" Power Mac G4
benchmarks 42.4 overall under Panther and 50.4 under Tiger (54.3 for
the CPU). The CPU test score for the 1.53 GHz dual G4 upgrade is 51%
higher than the 1 GHz dual G4 - exactly what you'd expect. However,
Memory is over 30% faster on the Mirror Drive Door model (38.2) with
its 167 MHz system bus (vs. 133 MHz in the DA).
The system was tested on 6 July 2009 with a 1.6 GHz CPU, 1.25 GB of
RAM, Radeon 9000 graphics, and an external FireWire hard drive. While
the overall 10.4 score is a bit lower than before, the Leopard score
has improved significantly. Let's look more closely.
10.4.11 10.5.7
Overall 56.3 45.7
CPU 86.3 86.3
Threads 114.2 75.0
Memory 28.4 30.7
Quartz 62.8 65.7
OpenGL 74.1 64.5
User Int. 48.4 21.6
Drive 53.5 53.7
We would expect CPU and Threads results to be about 4.5% better, and
they're a bit better than that under 10.4. Memory results are almost
the same. Quartz scores 9% higher, OpenGL is 31% lower, and the User
Interface score is 12% lower. Drive scores with the FireWire 400 drive
are 5% better than with the internal Ultra ATA66 drive used in earlier
benchmarks. A real mixed bag of results under "Tiger".
Mac OS X 10.5.7 "Leopard", on the other hand, scores 57% better
overall. That's huge! The CPU score is exactly what we'd expect based
on the G4 upgrade running at 1.6 GHz this time vs. 1.53 GHz in earlier
tests. Threads takes a 13% hit, but Memory benchmarks 14% higher.
Quartz measures just over twice as fast, as do OpenGL and User
Interface - the Radeon 9000 graphics really shines here. And the Drive
score improves by an impressive 24%.
In contrast to Xbench scores using our 500 MHz dual G4 "Mystic"
Power Mac, where the internal Ultra ATA66 drive outperformed the
external FireWire 400 drive, this time around the FireWire drives are
superior. Between a slightly faster CPU, a much more powerful video
card, 512 MB of additional memory, and using a FireWire drive, we're
seeing significant improvements.
Geekbench
Geekbench only runs on Tiger and Leopard. Here are our results:
533 MHz 1.53 GHz dual 1 GHz dual
10.4.11 10.5.6 10.4.11 10.5.6 10.4.11 10.5.6
Overall 362 321 1140 1116 894 939
Integer 406 354 1627 1598 1159 1131
Floating Point 400 361 1288 1277 1015 1179
Memory 314 265 495 443 495 470
Streams 177 187 214 214 351 342
Overall results show the dual 1.53 GHz upgrade provides over 3x the
power of the original 533 MHz CPU, while the Integer score in Tiger
shows the upgrade has 4x the power.
The dual 1 GHz G4 has an overall score of 894 (Tiger) and 939
(Leopard) - the upgraded DA Power Mac is 27.5% faster overall in Tiger,
18.9% in Leopard. It's interesting that the Memory results under Tiger
are identical for the dual 1 GHz on its 167 MHz system bus and the dual
1.53 GHz upgrade on a 133 MHz system bus. This indicates the DA with
the 1.53 GHz upgrade may be a suitable replacement for the dual 1 GHz
MDD.
The system was tested on 6 July 2009 with a 1.6 GHz CPU, 1.25 GB of
RAM, Radeon 9000 graphics, and an external FireWire hard drive. While
the overall 10.4 score is a bit lower than before, the Leopard score
has improved significantly. Here are the results:
10.4.11 10.5.7
Overall 1218 1311
Integer 1673 1638
Floating Point 1471 1784
Memory 486 454
Streams 210 235
The improved graphics card and use of a FireWire boot drive should
have negligible impact, and the 4.5% faster CPU should result in better
Integer and Floating Point scores.
Under 10.4, the Integer score is just 2.8% better, but Floating
Point results are 14% better. Memory and Streams are just a bit slower,
and overall the system benchmarks about 6.8% better than before.
Under 10.5, the overall score is 17.5% higher. The Integer score is
only 2.5% better, but Floating Point is nearly 40% higher, which points
to some optimization in 10.5.7 vs. 10.5.6. This could be due to the
operating system moving more graphics work to the graphics processor.
Memory scores 2.5% higher, while Streams takes a 31% performance
hit.
Where Next?
At present, the DA Power Mac suffers from too little memory (768 MB)
to unleash the power of OS X and an antiquated video card. We hope to
boost this to 1.5 GB of RAM using donated memory, and I suspect this
will make it feel quite a bit snappier.
UPDATE: We have since upgraded to 1.25 GB of RAM.
There are two drawbacks to the older Power Macs:
- They don't support hard drives over
128 GB on the internal bus without third-party drivers, and I have
two 400 GB drives in the MDD Power Mac. [UPDATE: Since moved to
external NewerTech miniStack hard drive enclosures.]
- The ATI Rage 128 Pro video card is hopelessly outdated. (UPDATE:
The 2002 Radeon 9000 makes a significant difference, doubling several
graphics benchmarks. More modern graphics cards would do even
better.)
We'll solve the first problem by moving both drives to external
NewerTech miniStack v2
enclosures, which use FireWire 400 and will make it possible to move to
a different computer easily. FireWire doesn't have the "big drive"
problem, and being less resource intensive than IDE, FireWire drives
could result in better performance than using the drives inside the
computer.
We'll solve the second problem by swapping the ATI Radeon 9000 Pro
from the MDD into the DA Power Mac. Since our server won't be doing any
graphic-intensive work, it shouldn't matter that it will have an older,
slower video card.
The end result may well be moving to the "Digital Audio" Power Mac
for production and setting up the MDD as our in-house server. It's fast
system bus, 2 GB of RAM, and twin 1 GHz CPUs should make it a great
server.
Time will tell.
Conclusion
The Power Mac G4 line was the last family of PowerPC Macs that can
readily take a CPU upgrade, and while the 350 MHz and faster G4
processors that shipped with them offered a lot of power in their day
(the era of the Classic Mac OS), Apple never offered G4 Power Macs with
1.6 GHz to 2.0 GHz CPUs.
If you have a Power Mac G4 (Sawtooth through Quicksilver 2002) that
you're happy with but need more horsepower, CPU upgrades are certainly
worth consideration.
Only you can put a dollar value on increased performance and
productivity. One huge plus to upgrading your current setup is that you
don't have to worry about memory, drives, add-in cards, etc. not being
compatible, because you're not migrating to a newer Mac.
Manufacturers and distributors: Interested in having
your product reviewed? Please read our review
policy.
