I have to admit to feeling a bit cheated after running MacBench 5 on
my wife's new indigo iBook. After all,
I've been a cheerleader for the new G3e (PowerPC 750CX) processor since
IBM 
announced it in June. (See Should Apple Use the New G3? for all the
details.)
But MacBench 5 showed the new iBook with the new processor was
slower than the old processor in the old iBook - and not just a
bit slower. All things being equal, you'd expect a 366 MHz G3e to
outperform a 300 MHz G3. Instead, MacBench showed CPU scores 28%
lower!
Yeah, I felt a bit cheated. Maybe we should have picked up the older
blueberry iBook, saved some money, and had a faster computer.
I ran the tests four times: with and without virtual memory, with
different VM settings, and once with all extensions off. Every time the
CPU score was terrible.
Oddly enough, the 750CX way outperformed the old G3 on the FPU
benchmark, scoring 28% better even though the clock speed is only 22%
higher. That was encouraging.
Still, something seemed very, very wrong. After all, IBM claimed the
750CX with its on-chip 256 KB cache would pretty much match the
performance of the old 750 (G3) with a 512 KB backside cache running at
half CPU speed. Since the original iBook ran the cache at 40% of CPU
speed, I certainly didn't expect the new iBook with a 366 MHz 750CX
brain to run slower than the old 300 MHz iBook.
I spent most of a day reflecting on this. Was the cache too small to
work efficiently with the Mac OS? Was something in OS 9 dragging down
performance? What could explain this abysmal performance when the new
processor at a higher clock speed should have blown past the old
iBook's scores?
Then came the thought, "What if MacBench 5 is flawed?"
One of my favorite benchmarks is
Speedometer 4.02. It's old enough that a score of 1.0 is Quadra 605
performance, but it's new enough to run in PowerPC native mode on newer
Macs. Best of all, it runs on any Mac with a 68020 processor or later,
so you can use it to compare the Mac II with the Power Mac G4.
When I ran Speedo 4 on my wife's new iBook, I got a very
different set of results from those I saw under MacBench 5. The
iBook and its 366 MHz 750CX processor roared past the G3/333 in my
SuperMac S900.
Clockometer shows the G3 in my SuperMac actually runs at 330 MHz,
not 333. All else being equal, the iBook should score about 14-15%
better than the SuperMac. Of course, the card in my G3 has a 1 MB
backside cache running as 220 MHz, but it's also stuck on a 44 MHz
system bus. In contrast, the iBook has a small 256 KB cache running at
366 MHz and a 66 MHz system bus.
Final results: the iBook outperformed the SuperMac on all three
benchmarks.
test CPU disk math
iBook 28.2 3.24 1016
S900 25.2 2.95 891
What's really impressive is that it scored 12.2% better on the CPU
benchmark and 14.1% better on the math benchmark. Considering the
larger 1 MB cache on my G3-upgraded SuperMac, that's impressive
indeed.
Of course, all things are never equal: every benchmark has biases
and only tests part of the whole computer - something Macworld is
addressing with their new Speedmark 2.1 benchmark suite that uses real
applications.
In the end, Speedometer helps me not feel cheated because of poor
MacBench 5 scores - MacBench is only one benchmark. Speedometer and
IBM's benchmarks show a different story, leaving me grateful Macworld
won't be benchmarking the new iBook with MacBench 5. I'm also looking
forward to their real-world findings using Speedmark.
I still wonder if a 256 KB cache is well matched with the Mac OS,
especially since all the G3 Macs have used 512 KB or 1 MB caches. And I
wonder how much better a next-generation 750CX with a 512 KB or 1 MB
cache might perform.
Most of all, I now feel that my
cheerleading for the 750CX was not in vain and that Apple has made
an excellent processor choice for the new iBook.
Further Reading
