|
Dual G4 Shootout: DP 533 vs. DP 800
by Michael Flaminio
8.27.01
Apple's dual processor 800 MHz G4 is the new king of Macs. The new G4 replaces
the 733 G4 as the fastest and most expensive Mac, surpassing even the 867 MHz G4
in Apple's price structure. Apple has placed a premium on its dual processor configuration,
making it only available in top configurations. Multiprocessor G4s  have become more and more in demand
as Apple's Mac OS X heads for mainstream usage. But not to be forgotten is the round
of multimedia applications, such as Apple's suite of video editing apps, which also
benefit greatly from multiple processors.
When announced earlier this year, the DP 533 was considered a diamond in the rough.
By utilizing two CPUs based on older, more efficient PowerPC 7410 processors, the
DP 533's proved to pack a punch. Moderately priced, multimedia creators and OS X
adopters found the machines to hold great value as it chewed through multi-threaded
apps.
Apple's newest DP G4 is the G4 800, using the newer PowerPC 7450 processors. Since
it was announced, the question on our minds was how would this Mac perform against
the popular DP 533?
PowerPC 7450
In 2000, Apple's PowerMac G4 line sat at 500 MHz. While dual processor machines were
made available, nothing could be found beyond the 500 MHz rate. Presumably to better
compete in the MHz wars, the next revision of the G4 processor included measures
to allow it to achieve higher clock speeds more rapidly. The ultimate trade off was
the chip would not be as efficient as previous chips. For example, a 533 G4 based
on the PPC 7410 was not 27% faster than a 733 G4 based on the PPC 7450.
One reason for this is the difference is cache handling. The new chips include a
256K L2 cache and a 1 MB L3 cache, compared to a 1 MB L2 cache. This design difference
limited performance compared to the more traditional chip designs. Ideally, once
software is recompiled to accommodate the chip design, performance will improve.
According to Metrowerks, however, software changes will only offer around a 5% boost
in performance for PPC 7450 chips. While this will help, this boost will not overcome
the complete performance loss of the new designs.
With the latest round of 800 MHz and 867 MHz G4 systems, the G4 CPUs feature some
different cache configurations. Mainly, the 256K now operates at the same speed as
the CPU, and the L3 cache now sports 2 MB, rather than 1 MB sizes. These changes
appear to improve performance further, helping to reduce the impact of the new designs
even more.
Tests
So with that preamble, lets get to the tests. The results were a bit of a mixed bag.
For the most part, the DP 800 faired as expected or better than than expected compared
to the DP 533. There were, however, a few results that left us scratching our heads.
We chose a variety of tests in hopes of gaining a good spectrum on performance. All
tests had a dual CPU portion, except for Quake 3. Test were performed on a DP 533
(7410 v1.3) and DP 800 (7450 v2.1) systems with as many shared components as possible.
The machines were quipped with 768 MB of RAM (same DIMMs, 2-2-2 speed) 60 GB hard
drive (IBM, 7200 RPM) GeForce3, Mac OS 9.2.1 and Mac OS X 10.0.4 (4S10). Test details
are as follows:
QuickTime 5.02 under Mac OS 9.2.1. Sorenson Video 3 Pro was tested as well as Apple's
MPEG-2 codecs. The test file was a 770 MB DV file. SV3 compressed the file to 360x240
at high quality, 12 fps. Both videos were compressed without sound.
Throughput was using Throughput 1.5. Gauge Pro was version 1.1, Cinebench was Cinebench
2000.
Quake was under 9.2.1 and was version 1.29f. Locki script was version 1.3. Each test
was done from a fresh config file. Fastest, Normal and High Quality are the built
in config settings.
SoundJam tests used a 55 MB AIFF file and converted to a standard MP3, and back to
AIFF.
SMP Bench was Ars Technica SMP Benchmark. This was performed under Mac OS X.
| Test |
DP 533 |
DP 800 |
Diff. |
| QT - SV3 |
224.2 |
163.9 |
26.9% |
| QT - MPEG-2 |
233.5 |
184.9 |
20.8% |
| |
|
|
|
| Throughput - CPU |
50.1 |
214.0 |
76.6% |
| Throughput - FPU |
171.8 |
214.3 |
19.8% |
| Throughput - AltiVec |
180.7 |
221.3 |
14.9% |
| Throughput - CopyBits |
335.1 |
251.6 |
(-24.9)% |
| |
|
|
|
| Q3 - Fastest |
46.7 |
78.1 |
40.2% |
| Q3 - Normal |
42.3 |
69.5 |
39.1% |
| Q3 - Highest |
41.1 |
69.8 |
41.1% |
| Q3 - Locki |
88.0 |
146.1 |
39.8% |
| |
|
|
|
| Guage Pro |
309.2 |
151.0 |
-(54.4)% |
| |
|
|
|
| Cinebench - Cinema 4D |
6.05 |
6.71 |
10.0% |
| Cinebench - OpenGL |
7.72 |
9.04 |
14.6% |
| Cinebench - Raytrace SP |
5.66 |
9.71 |
41.7% |
| Cinebench - Raytrace MP |
11.89 |
17.12 |
30.5% |
| |
|
|
|
| SJ - MP3:AIFF |
18.75 |
14.84 |
20.9% |
| SJ - AIFF:MP3 |
23.72 |
24.16 |
(-0.2)% |
| |
|
|
|
| SMP Bench - 1 Thread |
186.28 |
216.72 |
14.0% |
| SMP Bench - 2 Thread |
328.23 |
376.28 |
12.8% |
| SMP Bench - 4 Thread |
342.44 |
388.51 |
11.9% |
| SMP Bench - 6 Thread |
357.42 |
405.69 |
11.9% |
| SMP Bench - 8 Thread |
350.31 |
398.34 |
12.1% |
Conclusion
Performance was interesting. Ideally we would see a 33% increase in performance
between 533 and 800 MHz systems. Video marks were both respectable, as were the Cinebench
scores. Quake 3 scores were in my opinion outstanding at a consistant 40% increase.
SoundJam had a curious flat score for AIFF to MP3, but the other way around looked
good. SMP Bench wer a little low, but consistent overall. The memory bandwidth tests
have always been low for PPC 7450 CPUs. Such a drastic difference suggests the application
is not working correctly with the new CPUs, but these results seem to be accepted
by some.
Hopefully these test provide a little perspective on these systems. To some, it may
justify a DP 800, to others, it may make a discontinued or used 533 more attractive.
|