September 21, 2001
Ok there is no Moe. But we did preliminary performance
testing on the new Quicksilver dual G4/800 and thought we
share the results with you. Like the 867 MHz Quicksilver,
the 800 MHz machine has 256K of on chip L2 cache running at
full processor speed, and 2 MB of L3 cache running at 1/4
the processor speed. In the case of the 800 MHz machine everything
is doubled. (L3 cache sits between the processor and main
You will find that in most of the tests below,
the 867 MHz Power Mac bests the dual 800 by a small margin.
That is because most of our test, like most software under
OS 9.2, does not take advantage of dual processors. The exception
to this is in our Photoshop tests and CineBench 2000 RayTracing
test, where the dual processor machine really comes into its
own. OS X is supposed to be built from the ground up to take
advantage of dual processors. Once you are running a Carbon
or Cocoa application,
you should see a significant performance improvement. We will
be testing that assumption in the next few weeks.
Anyway take a look at the results below. We
have included our best guess about what is going-on to generate
these differences. If you have a better explanation, or a
different opinion, please let us know. Either e-mail
us a message for us to post or use this link
to post your ideas on our bulletin board.
All machines were configured similarly and running
Note: This is our initial look at the performance
of the new Quicksilver Towers. We will be running more tests
and hope to have a full review of all the new Power Macs in
the next few weeks.
"Real World" Tests
The tests below are from our suite of real
world application tests. These tests feature a diverse
selection of applications commonly used by the Mac community.
The test suite was designed to render an accurate and well
rounded picture of a machine's performance. All of the tests
below (with the exception of the Quake III & CineBench
2000 tests) were timed with a stopwatch. The times are then
converted to percentages relative to the Quicksilver 733 MHz
machine which is set to 100%. For all scores, higher numbers
The test above copies a folder containing thousands
of files. The rotational speed of the 733 MHz's hard drive
is 5200 rpm. The 867 MHz and 800 MHz dual processor, 7200
rpm. These results show that not all drives are created equal.
Apple, in the past, has been known to change hard drive suppliers
mid-stream ... you go where you can get the best quality for
the lowest price.
In reading and writing a single, large data file, the 867
is the winner. Perhaps also the existence of the backside
cache is a factor in comparison with the 733.
AppleWorks 6 Tests
The additional speed of the 867's & 800's processors,
and and the backside cache account for the increased performance
here. The drive is not a factor in this test.
OK, here is where the machines with the backside cache come
into their own. The document we use is 1.5 MB ... theoretically
it could all fit completely into the 2 MB backside cache of
the 867 & 800. On the 733, without any kind of backside
cache, the processor must trudge all the way to main memory
to get much of the information it needs. So if a lot of your
work involves large databases, spreadsheets or documents,
you will want to be sure to consider one of the machines with
a backside cache. On small documents (under 256K),the performance
difference will not be that great. The 800 machine does a
good job of keeping up with the 867, even though dual processors
are not a factor in this test.
Quake III Tests
These scores are relative.
Here again the processor speed and improved memory subsystem
gives the 867 & 800 their disproportional advantage. No
dual processor work is being done here.
Why nearly identical results here? All three machines have
the same graphics card and at high-quality setting it appears
that Quake is running exclusively off the card ... using the
other subsystems little, if at all.
Photoshop 6 & Other Data Crunching Tests
We use an 20 MB image to launch Photoshop. Here the drive
plays a part as does the processor and memory subsystem. A
lot of rendering is going on here not just hard drive activity.
These two Photoshop test (above and below) are carried out
completely in RAM - there is no drive activity. The added
benefit of the 800 MHz machine's dual G4 processors makes
itself felt here.
Here, raw processing power is more important
than dual processors
The 733 did quite well in this test, though still 10% less
than it should when clock-speed is accounted for. No dual
processor or AltiVec code here
This shows you how much extra performance you can squeeze
out of a dual processor set up when the software is written
to take advantage of it. OS X is supposed to take advantage
of dual processors as a matter of course. We will be testing
this assumption in the next few weeks.
This test stresses both the processor and
its subsystems, and the graphics card
Again we see a similar pattern. The hard drive is accessed
here but is not much of a performance factor as it can deliver
data much faster than it can be processed. It appears that
under OS 9.2 QuickTime does not take advantage of dual processors
(at least for a Sorenson encode). However the G4 processor
stomps a similarly clocked G3, when doing this type of encoding.
The 733 has a 32X CD drive and the 867 & 800 have 24X
CD drives. The file being converted is coming off a CD.
Hard drive is a partial factor here.
The 733 writes CD-Rs at 12X and the 867 &
800 at 8X