The Performance Edge: Eeny Meeny, Miney Moe - All The New G4 Towers Go Toe To Toe - A Performance Report

Thursday, March 21, 2002

We finally have benchmark results for all three of the new Tower Power Macs. In general, the results confirm our feelings about each one of these machines.

The 800 MHz is great for those on a budget, who will use the machine for more light duty activity in general, and only have need for more processing muscle occasionally. The 800 will be plenty fast for you, and you'll get to keep $700 in your pocket for other goodies.

The G4/933 is the best bet for those that need moderate amounts of processing bandwidth on a more regular basis, but don't use their machines in a production type manner. At the moment very few applications are written to take advantage of dual processors as a matter of course. So in the near term, unless you are going to be running multiple processor intensive applications at once, a high-powered single processor machine will suit you just fine.

If you are in a production type environment, or the applications you want to use, either are already tuned for dual processor machines or soon will be, the Dual Processor Tower is the machine for you. It will pay for itself in saved time, and more than justifies the $700 premium that you will pay for it. You will have to do some planning to see how best to harness all the power you will have underneath the hood. But this machine should be a real boon in a production type environment where you can be running several processor intensive applications at once - your productivity could soar.

Hardcore gamers will want to look at the two fastest machines. And in this arena, if Giants is a forbearer of things to come, those for whom gaming is a religion, may want to plunk down the extra brass for the dual processor machine. We achieved and average of 46 fps with Giants on the dual processor machine, and only 26 fps on the 933 - though they have the same graphics card.

Which Power Mac would you buy? Let us know why

Difference and similarities in processor and memory & graphics systems of each machine


  Power Mac G4/800 Power Mac G4/933 Power Mac Dual G4/1GHz

Additional Resources

Check prices at:

 

Processors G4/800 G4/933 2 x G4/1000
L2 Cache 256k @ 800 MHz 256k @ 933 MHz 256k @ 1 GHz
L3 Cache None 2 MB @ 233 MHz DDR 2 MB @ 250 MHz DDR
Memory 768 MB PC 133 768 MB PC 133 1 GB PC 133
Graphics Card ATI RADEON 7500 with 32MB DDR SDRAM NVIDIA GeForce4 MX with 64MB of DDR SDRAM NVIDIA GeForce4 MX with 64MB of DDR SDRAM
Drive 40 GB @ 7200 rpm 60 GB @ 7200 rpm 80 GB @ 7200 rpm
Operating System 10.1.2 10.1.2 10.1.2
Price $1,599 $2,299 $2,999


"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, except for the game tests, were timed with a stopwatch. The times are then converted to percentages relative to the Power Mac G4/800, which is set to 100%. For all scores, higher numbers are better.


Desktop Tests

 

Actual time: G4/800 : 44.02 sec .... G4/933: 36.75 sec .... Dual G4/1GHz: 34.38

 

Not sure why the Dual processor machine turned in poorer performance here. Perhaps more code needs to load to accommodate the extra processor. Time difference was only about 3 seconds.

Actual time: G4/800 : 32.14 sec Processor usage: 80% .... G4/933: 20 sec Processor usage: 100% .... Actual time: Dual G4/1GHz: 23.5 sec

Actual time: G4/800 : 12.15 sec Processor usage: 40% .... G4/933: 10.79 sec, Processor usage: 20% .... Actual time: Dual G4/1GHz: 10.54 sec, Processor usage: 10%

Actual time: G4/800 : 9.91 sec Processor usage: 20% .... G4/933: 8.62 sec, Processor usage: 10% .... Actual time: Dual G4/1GHz: 8.48 sec

The drives on all machines are very fast and comparable in performance

Actual time: G4/800 : 80.84 sec .... Processor usage: 15% G4/933: 122.73 sec, Processor usage: 10% .... Actual time: Dual G4/1GHz: 119.63 sec

The 933 MHz and Dual Processor machines have the same drive for burning CDs. They both burn CDs at 8X. The 800 MHz machine burns CDs at 24X. If you are mainly going to be burning CDs (and not DVDs), do yourself a favor and drop the SuperDrive from whatever machine you get.. You'll burn faster and save yourself $200 by dropping the DVD-R drive, in favor of a CD-RW one. The CD burning process is not a processor intensive task.

Actual time: G4/800 : 47.39 sec Processor usage: 100% .... G4/933: 34.45 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 34.14 sec, Processor usage: 60%

The test above creates and destroys 1,000 windows. See the Let1kWindowsBloom site for more info. The speed improvement of the two faster machines is probably due mostly to the better graphics card installed in each machine

Actual time: G4/800 : 21.68 sec Processor usage: 50% .... G4/933: 17.93 sec, Processor usage: 30% .... Actual time: Dual G4/1GHz: 17.92 sec, Processor usage: 10%

Large document is scrolled from one end to the other using Classic OS 9.2.2 when booted in OS 10. Both of the faster machines have the same graphics cards.


Large Document & Database Type Tests

Actual time: G4/800 : 29.14 sec , Processor usage: 100%.... G4/933: 21.56sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 19.79sec, Processor usage: 50%

This test takes place in a large AppleWorks document. Here you can see the 800 MHz suffering a disproportionate performance hit because of its lack of a L3 backside cache

Actual time: G4/800 : 64.43 sec Processor usage: 95%.... G4/933: 45 6sec, Processor usage: 95% .... Actual time: Dual G4/1GHz: 43 sec, Processor usage: 45% (Non-multitasking)

Only one processor at a time was used on the dual processor machine. It does not appear that the indexing function of Sherlock is multi-threaded. However the 800 MHz is hit again in the performance solar plexus by that lack of a L3 cache


Number Crunching & Rendering Tests

Actual time: G4/800 : 28.74 sec Processor usage: 100%.... G4/933: 27.62 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 28.38 sec, Processor usage: 50%.

Pretty even performance here, but the dual processor machine was cooling its heals waiting for data

Actual time: G4/800 : 206.03 sec Processor usage: 100% .... G4/933: 171.40 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 149.38 sec, Processor usage: 50%

The faster machines are helped not only by the increase in speed, but also, moderately by their L3 cache

Actual time: G4/800 : 66.55 sec Processor usage: 100% .... G4/933: 56.69 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 26.54 sec, Processor usage: 100%

The Fractal program has been highly tuned to take advantage of the G4 and is precisely the type of work that the G4 was made for. It will also gobble up whatever processing capability is present. Now why can't all processor intensive tasks take advantage of the dual processor machine like this one does?


Encoding/Decoding Tests

Actual time: G4/800 : 248.64 sec Processor usage: 100% .... G4/933: 195.33 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 175.05 sec, Processor usage: 50%

This version of Sorenson does not take advantage of dual processors ... you have to pay big bucks to get that capability. L3 cache helps here too

Actual time: G4/800 : 172.27 sec Processor usage: 100% .... G4/933: 116.71 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 76.88 sec, Processor usage: 80%

Converting QuickTime movies to DV allows you to import them into iMovie. Now that is better, 80% usage on the dual processor machine.

Actual time: G4/800 : 132.06 sec Processor usage: 100% .... G4/933: 98.36 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 94.39 sec, Processor usage: 75%

Both of the faster machines have the SuperDrive and thus read data from and to the CD at the same speed. The 800 MHz machine reads data faster than the other two, but in this case processing speed trumps optical drive speed

Actual time: G4/800 : 12.42 sec Processor usage: 100% .... G4/933: 9.67 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 8.92 sec


Multitasking

Actual time: G4/800 : 516.37 sec Processor usage: 100% .... G4/933: 424.84 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 280.81 sec

For the above results we fired up two copies of QuickTime and ran two encodes, one on each processor at the same time (something that cannot be done under OS 9). Need to do this type of heavy duty rendering on a regular basis? Run, don't walk, to the nearest dual processor machine you can find!

Actual time: G4/800 : 311.38 sec Processor usage: 100% .... G4/933: 424.84 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 280.81 sec

Same as previous only a different type of encode ... one that is multitasking even on one processor. I think we are beginning to see a pattern here as far as the multiprocessor machine is concerned

Actual time: G4/800 : 522.21 sec Processor usage: 100% .... G4/933: 424.84 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 280.81 sec

A Sorenson encode and the AltiVec Fractal are performed at the same time

Actual time: G4/800 : 175.99 sec Processor usage: 100% .... G4/933: 115.66 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 73.53 sec

An MP3 encode, Search & Replace in AppleWorks and Folder Copy are all carried out at the same time

Actual time: G4/800 : 311.55 sec Processor usage: 100% .... G4/933: 65.24 sec, Processor usage: 100% .... Actual time: Dual G4/1GHz: 45.77 sec, Processor usage: 100%

Again the above encodes are performed at the same time


Gaming

Actual fps: G4/800 : 18 fps .... G4/933: 26 fps .... Dual G4/1GHz: 46 fps

Again here the Dual Processor machine really shines, even though the 933 has the same graphics card. Giants is a game that has been developed to take advantage of dual processors .... let's hope that all developers begin to take this kind of care in developing their products.

Actual fps: G4/800 : 19 fps .... G4/933: 29.5 fps .... Dual G4/1GHz: 50 fps

On the lowest quality setting of Giants you only pick up a couple of frames.