G3 vs. Pentium III
As we suspected, the Apple Power Macintosh G3 beats a Dell
Pentium III on some tests and competes well on others.
By Laura Cox
March 26, 1999-- Once again, PC Magazine Labs tests
with FileMaker Pro 4.01 and Adobe Photoshop 5.02 reveal that the Apple Power
Macintosh G3/400 generally outperforms x86-based products--in this case,
a Dell Dimension PIII/500. Because the main architectural difference between
the PII/450 processor and the PIII/500 is the new Streaming SIMD Extensions
(SSE), we didn't expect to see much performance increase between the two
Intel processors--as well as compared with the G3--in most of our tests,
which are not SSE-enhanced (for more information, see "How
We Tested" in our First Looks review of the Apple Power Macintosh
G3/400).
The Dell PIII/500 marginally outperformed the Dell PII/450 (see our results
comparing the G3
vs. PII/450 for PII results) in every test. The only exception is the
Photoshop Pixelate Mosaic filter based on a 30MB file, which only takes
a few seconds to complete and is more likely the result of human stopwatch
measurement error than processor speed.
We ran two Katmai-enabled filters in our Photoshop suite. The Power Macintosh
G3/400 still outperformed both Intel platforms on the Polar coordinates
Katmai-enabled filter. However, the PIII/500 outperformed the G3/400 in
the Photoshop Twirl filter (12 to 15.1 seconds), which likely takes better
advantage of Intel's SSE instructions (for more information, see "How
We Tested" in our review of PIII/500 machines). The PII/450 finished
Twirl in 19.6 seconds.
We also tested the Power Macintosh G3/400 with an IBM DTTA 351290 7200-rpm
EIDE hard disk to see if the G3's impressive disk scores were due to its
SCSI hard disk. We chose a subset of the most disk-intensive tests, including
FileMaker Pro and certain filters in Photoshop using a 100MB file. The IDE
disk performed similarly in the FileMaker Pro tests and a little slower
in our Photoshop tests. Apple has told us that the SCSI drive performs better
for larger streams of sequential transfers, but that the IDE disk performs
similarly to the SCSI drive for mixed disk operations, and in some cases
even beats the SCSI drive.
FileMaker Pro & Word Results
Results in seconds; Note: LOWER numbers are
better |
Power Mac G3/400 (SCSI HD) |
Power Mac G3/400 (IDE HD) |
Dimension PIII/500 (Katmai) |
FileMaker Pro 4.01 Import |
23.7 |
23.8 |
26.4 |
FileMaker Pro 4.01 Sort |
41.4 |
41.0 |
52.6 |
FileMaker Pro 4.01 Export |
16.3 |
16.4 |
19.0 |
|
|
|
|
Microsoft Word Search/Replace |
16.9 |
|
9.1 |
Microsoft Word Spell Check |
18.3 |
|
13.8 |
FileMaker Pro 4.01 Import.
We imported a 52MB CSV data set and converted it to FileMaker Pro format.
This test is disk intensive.
FileMaker Pro 4.01 Sort. We sorted the imported and converted file
according to different sort keys. This test is both integer-calculation
and disk intensive.
FileMaker Pro 4.01 Export. We exported the sorted database to WKS
format. This test is integer-calculation and disk intensive.
Microsoft Word
Search and Replace. This is a simple search and
replace test. We replaced the word "the" with the word "whatever"
in a 968K Word file. 15,250 changes were made.
Microsoft Word Spell-Check. This is a simple spell-check of a 968K
Word file. We used Microsoft Office 98 on the G3 and Microsoft Office 97
on the PC.
Photoshop Results
Results in minutes:seconds; Note: LOWER numbers
are better |
Power Mac G3/400 (SCSI HD) |
Power Mac G3/400 (IDE HD) |
Dimension PIII/500 (Katmai) |
Photoshop Gaussian Blur (30MB file) MMX
|
0:09.3 |
|
0:11.4 |
Photoshop Gaussian Blur (100MB file) MMX |
1:34.0 |
1:59.0 |
2:21.0 |
Photoshop Lighting Effects (30MB file) Floating
Point |
1:17.0 |
|
0:58.0 |
Photoshop Lighting Effects (100MB file) Floating
Point |
4:31.0 |
4:40.0 |
4:03.0 |
Photoshop Noise Median, 8 pixels (30MB)
|
0:28.3 |
|
0:30.2 |
Photoshop Noise Median, 8 pixels (100MB)
|
2:39.0 |
2:57.0 |
3:12.0 |
Photoshop Pixelate Mosaic, 64 pixels (30MB)
|
0:01.7 |
|
0:04.0 |
Photoshop Pixelate Mosaic, 64 pixels (100MB)
|
0:52.0 |
0:58.0 |
1:42.0 |
Photoshop Polar Coordinates (30MB) Katmai enabled
filter |
0:16.8 |
|
0:32.3 |
Photoshop Stylize Trace Contour, level 128,
edge lower (30MB) |
0:03.1 |
|
0:03.9 |
Photoshop Stylize Trace Contour, level 128,
edge lower (100MB) |
0:45.0 |
0:51.0 |
1:01.0 |
Photoshop Unsharp Mask (30MB file) |
0:04.5 |
|
0:06.1 |
Photoshop Unsharp Mask (100MB file) |
1:15.0 |
1:36.0 |
2:00.0 |
Adobe Photoshop test files: We used two source TIFF images in
our Photoshop tests. We scaled these images to create two different files
sizes per image: 30MB and 100MB. Filter operations on the 30MB file resulted
in heavy processor and memory utilization and virtually no disk activity;
the 100MB file tests the entire system, including processor, memory, and
disk. These tests are not graphics intensive, as they simply display a single
image. We used the latest versions of Photoshop (v5.02) for both platforms.
Photoshop 5.02 Gaussian Blur.
This is an integer-intensive image-blurring filter that uses MMX functionality.
We set the radius to 2 pixels.
Photoshop 5.02 Lighting Effects. This test changes the image lighting;
it is extremely floating-point-intensive.
Photoshop 5.02 Noise Median. This filter can smooth edges, decrease
pixelization, and add shadow effects, among other things. We set the radius
to 8 pixels.
Photoshop 5.02 Pixelate Mosaic. This filter clumps pixels of similar
color values into square blocks. We set the cell size to 64 pixels.
Photoshop 5.02
Polar Coordinates. This filter geometrically reshapes
a selection to its polar coordinates and is very memory intensive. We used
the default settings.
Photoshop 5.02 Stylize Trace Contour. This filter displaces pixels
based on major brightness transitions in the test image. We used the default
settings.
Photoshop 5.02 Unsharp Mask. This filter adjusts the contrast of
edges by producing a lighter and darker line on each side of an edge to
emphasize it and create the illusion of a sharper image. We used the default
settings.
Quake Results
Results in frames per second; Note: HIGHER
numbers are better |
Power Mac G3/400
(Rage 128 card) |
Dimension PII/450
(??? card) |
Dimension PIII/500
(nVidia TNT card) |
Quake TimeDemo 1 (s/w xlr8d) |
31.4 |
38.8 |
41.5 |
Quake TimeDemo 2 (s/w xlr8d) |
30.9 |
44.6 |
47.6 |
Quake TimeDemo 3 (s/w xlr8d) |
30.4 |
37.4 |
39.9 |
|
|
|
|
Quake TimeDemo 1 (3D h/w xlr8d) |
92.3 |
112.7 |
114.5 |
Quake TimeDemo 2 (3D h/w xlr8d) |
97.1 |
111.2 |
112.1 |
Quake TimeDemo 3 (3D h/w xlr8d) |
91.3 |
107.4 |
109.3 |
Quake Timedemo 1, Timedemo 2, and Timedemo 3. These are built-in Quake demos that deliver a result in frames
per second. All tests were run at 640x480 game resolution.
[Page Disclaimer: This page appears
w/o ZDnet's permission, and appears primarily as their report on one page
over their
report spread over four pages. This page was also proofed in detail
for print consistency.]
[Report Summary: Out of 24 tests, Apple's G3/400 bested Dell's PIII/500
in 14 of them, two in half of the PIII's time (Pixelate Mosaic). Dell's
PIII beat Apple's G3 in all Word and Quake tests, but in only two Photoshop
tests (Lighting Effects).]
|