You'll get your Mac news here from now on...

Help TMO Grow

Subscriber Login

Advertising Info


The Mac Observer Express Daily Newsletter


More Info

Site Navigation

Home
News
Tips
Columns & Editorials
Reviews
Reports
Archives
Search
Forums
Links
Mac Links
Software
Reports
Contact

The Devil's Advocate
by John Kheit

The Real Megahertz Myth: Is Apple The Real Slim Shady?
May 9th, 2003

By now most have heard Apple's mantra. Megahertz are a myth. Or something like that. If you believe Apple's corporate propaganda, you might believe that Intel's megahertz are second-class citizens when compared with Motorola's. Depending on which flavor of Kool-Aid you're drinking, various Apple PR related materials suggest that it takes two-or-three Intel Pentium cycles under Windows to equate to a single Motorola G4 cycle under OS X. Don't believe the hype. In practice, if anything, the converse seems more likely to be true. If you want Apple's OS X to perform common functions at around the same speed as an Intel Pentium III (P3) system running Windows XP, you will need roughly two G4 cycles for every one P3 cycle.

The Tests
What do I base this on? Some tremendously unscientific tests, and some empirical and anecdotal evidence as well. Now, no one is arguing that a G4 is more efficient per clock cycle than a P3/P4 in a vacuum; however, we don't live and work in a vacuum. The real world demands functionality throughput, not processor throughput. How fast can you perform certain functionality, like scrolling, or loading a Web page? And in the real world, OS X and its applications exact a performance tax causing Macs to run more slowly. As such, Macintosh G4s cannot deliver two-or-three times greater throughput for average functionality. Perhaps the G4 is 30-40% more efficient in a vacuum, which is great, but the disparity in overall available hertz is well over 100% in Intel's favor (i.e., 1.42 Ghz G4 vs. 3.06 Ghz P4), which compounds the performance gap between Macs and PC systems. With those provisos out of the way, here are my unscientific test reference systems:

General System Description

System 1 (PC)

System 2 (Mac)

Test System:

Wintel Clone

Powerbook G4 DVI

Processor

500 Mhz Pentium III

667 Mhz G4

RAM

768 Mb

1 Gb

Hard Drive(s)

ATA 100, two 80 Gb 5400 RPM

ATA 66, 60 Gb 5400 RPM

Video

(old) ATI Rage 128 All-In-Wonder 32 Mb

ATI Radeon 7500 32 Mb

Operating System

Windows XP (latest patches applied)

OS X 10.2.5 (latest patches applied)

Test Software:

 

 

Suite

Microsoft Office XP (latest patches applied)

Microsoft Office X (latest patches applied)

Internet

Explorer 6

Explorer 5.2.2

Internet

Explorer 6

Safari Beta 2

MP3 Player

MusicMatch 7.5

iTunes 4.0

Media Player

QuickTime 6.1 (latest)

QuickTime Pro 6.2

Java App

Limewire Pro 2.7.3

Limewire Pro 2.7.3

Acrobat Reader

Adobe Acrobat Reader 5.05

Adobe Acrobat Reader 5.05

PIM

Outlook

Mail

 

 

iCal

 

 

AddressBook

IM Client

AOL

iChat

 

 

 

 

This system is running the following programs at any given time: MusicMatch, Explorer, AOL IM, Word, Acrobat, Outlook, ATI TV program (tuned to CNN at about 400x300 pixels), Norton System Works (AntiVirus, etc.).

This system is running the following programs at any given time: Finder, LaunchBar, Mail.app, Safari, iChat, Watson, iCal, Address Book, iTunes, Acrobat, Word, and Stickies.

I chose a bunch of programs in an attempt to mimic common operations for users, and ran a battery of tests, which were timed by stopwatch. I ran the following tests (three times, taking the average) on both systems using the exact same documents. I didn’t include the standard Photoshop tests because (1) the average person isn’t using Photoshop in any heavy way, and (2) everyone seems to concoct Photoshop tests that rely so heavily on processor media acceleration functions, whether AltiVec, or MMX/SIMD/iSSE, that the results only suggest that optimizing application code for media acceleration is more important than any design differences between processors.

Instead, I did scrolling tests because that is one of the most common user/computer interactions—a user might regularly scroll hundreds of times a day. Another frequent user activity that served as a test was loading and rendering a Web page. All documents and Web pages were opened to roughly the same visible window dimensions on each system. I used this Slashdot.com thread for a test Web page. I used a 194 page patent application with moderate formatting and no graphics for a test Word document. I also ran a Word macro to renumber the patent claims in the Word document; the macro comprises relatively intense find/replace, copy/paste and recalculation functionality. A 72 page "AGFA SnapScan[ner] 600 User's Manual," filled with art, was used for the PDF test document. I then re-ran these tests under a heavy load, which included running LimeWire, QuickTime looping through the new Matrix trailer, and iTunes looping "Breathe" by télépopmusik, all in the background. I also tested the boot-up time for each system from power-on until my login applications were completely loaded. Finally, I measured delays in switching between applications' windows and drawing selected menus.

The Results
Here are the unexpected tests results, which are all measured in seconds or percentage use of the CPU:

Test

System 1 (PC)

System 2 (Mac)

Performance Advantage in %

Load Web page Internet Explorer

3 seconds

35 seconds

PC by 1066.67%

Load Web page Safari

NA (3)

5

PC by 66.67%

Scroll down Web page in Internet Explorer

 

Note: To accurately measure scrolling on such different systems (which make different assumptions on how much data they skip when scrolling long documents) I used the page-down key on the keyboard or clicked in the page-down area of scroller-wells.

8

15

PC by 87.50%

Scroll down Web page in Safari

NA (8)

7

Mac by 14.29%

Scroll down PDF document

23

29

PC by 26.09%

Scroll down Word document, 194 pages

18

40

PC by 122.22%

Renumber 54 patent claims

7

20

PC by 185.71%

% CPU playing MP3

3% (2-4%)

10% (8-24%)

PC by 233.33%

% CPU playing QuickTime movie

45%(30-60%)

40% (30-55%)

Mac by 12.50%

% CPU running LimeWire

 

Note: LimeWire is a Java application and a notorious performance hog.

20% (15-25%)

8% (6-13%)

Mac by 150.00%

% CPU running Word with document open

10% (7-20%)

20% (14-25%)

PC by 100.00%

Load Web page Internet Explorer (under load)

11 seconds

115 seconds

PC by 945.45%

Load Web page Safari (under load)

NA (11)

15

PC by 36.36%

Scroll down Web page in Internet Explorer (under load)

11

56

PC by 409.09%

Scroll down Web page in Safari (under load)

NA (11)

13

PC by 18.18%

Scroll down PDF document (under load)

59

52

Mac by 13.46%

Scroll down Word document, 194 pages (under load)

52

254

PC by 388.46%

Renumber 54 patent claims (under load)

12

82

PC by 583.33%

Time to boot system

232

207

Mac by 12.08%

Bring Window to Focus of Another Program

0.1-0.3

0.1-0.3

0%

Bring Window to Focus of Another Program (under load)

0.2-1

0.2-1

0%

Delay in Menu Drawing from First Click.

 

Note: I have noticed that under OS X, sometimes when you slam the cursor all the way up the screen and your cursor tip goes beyond the menu bar, thereafter, when you click on a menu choice for the first time (i.e., after loading or activating a program) that there is an appreciable delay before the menu draws, and in some instances, the drop down menu fails to draw until you move the cursor.

0.0-0.2 seconds

 

This is likely not an accurate measurement because it is too short to accurately measure with the stopwatch or my limited perceptions.

0.1-0.5 seconds

 

This is likely not an accurate measurement because it is too short to accurately measure with the stopwatch or my limited perceptions.

PC by 200.00%

Delay in Menu Drawing from First Click. (under load)

0.0-0.2      seconds

 

1-5 seconds

 

PC by 1900.00%

 

 

 

 

 

 

Average

101.94%*

*[Author’s Addendum: Erick Wong correctly pointed out that the Average "PC Performance Advantage in %" was incorrectly calculated; as such the article has been updated to correct the error. The original 272% advantage I presented was erroneously calculated by taking a straight average, which doesn’t work out because of the negative values in the column. The new value was calculated by summing all the times (the % CPU values were thrown out because they total to the same amount), and then computing the PC's Advantage in %. Thanks to Erick Wong for pointing out the obvious error.

Of course, averaging this in anyway is nonsensical to start with (and why I noted early on in the article that this is "tremendously unscientific") because each item does not have an equal weight. In other words, if I surf 98% of the time, and I use Word only 2% of the time, it is nonsensical making a straight-line performance average.

The main point is this. Lets assume the above tests show that the PC had a 0% advantage over the Mac. When Apple suggests that the megahertz myth results in Macs smoking PCs of greater megahertz ratings, yet the reality is that a 667Mhz G4 Mac has no speed advantage over a 500Mhz P3, then the megahertz myth itself may be a myth.]

Basically, the P3 system spanked the G4. On the PC, scrolling and navigating through Word is instantaneous; scrolling redraw on the PC appears to move text lightly into view as if it were etched on a slab of granite. Conversely, the Mac's display/redraw feels, relatively, like you're manipulating it through molasses when you try live scrolling, particularly under heavy loads. Aqua is an appropriate moniker in this regard; when I grab the scroller and drag it, the screen updates in waves and splashes. I observed a similar performance differential when resizing windows; however, Aqua uses a far better redraw technique, which does not cause redraw "blitter" evident in Windows.

Worse yet, OS X multitasking is more painful. If I play iTunes, run LimeWire, run QuickTime, and run Word on the Mac, the system becomes unresponsive. It's like I'm trying to perform brain surgery with mops tied to my hands. To Apple's credit, even under heavy load, OS X never dropped QuickTime audio, while QuickTime audio dropout was severe under Windows XP. Under heavy load, video frame dropout of the QuickTime movie was severe for both systems. Otherwise, Windows XP was much more responsive under load in almost every other way. In particular, user interactions (i.e., moving windows, scrolling text, engaging menu selections, etc.) were much more responsive on the XP system when multitasking. More impressive still, XP did so running a slew of additional applications (including the ATI TV, which never dropped a frame of video or audio, and Norton System Works, which is a performance hog). Impressively, both systems experienced no MP3 audio dropout.

Of course, the above tests are not scientific, but they are symptomatic -- indicative rather than definitive. I can't imagine what a 3.06 Ghz P4 would do to a 1.42 Ghz G4. As it stands, I was surprised that my lowly 500 Mhz P3 system so soundly trounced all over my 667 Mhz G4 -- leaving my PowerBook's keyboard-marred screen and chipped-titanium carcass in its wake.

Reality Disparity
So why the apparent disparity between the PR claims and reality? Well, the truth is that overall system throughput is more complicated to measure and varies based on things other than just processor pipeline length or raw megahertz. It also depends greatly on operating system efficiency, efficiency in handling spin locks and context switches, shared runtime library efficiency, program code efficiency, compiler optimizations, memory bandwidth, instruction set capabilities and efficiency, etc. Sure the AltiVec processing units on the G4 do a great job of accelerating vector processing functions, but how about the more mundane activities that are executed far more often?

Lets not forget, OS X is an operating system that requires more horsepower to provide much of its advanced functionality; it's simply doing more stuff. For example, a pervasive object layer, which provides huge libraries of functionality (e.g., where any text box can be spell checked), takes its toll. Also, providing real-time display-PDF, which is animated and composited in 3D alpha-transparency layers, to your screen taxes the system. Such robust, extreme and beautiful technology and OS X's greater capabilities compound the need for speed on the platform. Apple's system is often, and I believe rightfully, considered to be the most advanced and vital operating system available for general computing applications. Nevertheless, from the end user's perspective, the result is painful. You need two hertz from a Motorola G4 under OS X to get similar performance to a single hertz from an Intel Pentium III under Windows XP. In other words, a 1 Ghz G4 is almost a requirement for a decent experience with OS X.

Reality Distortion Expected
I cannot blame Apple for trying to spin its lemons into lemonade. I expect it. They have been dealt a bad hand from Motorola as of late. So it stands to reason that its megahertz myth video talks of pipe-line efficiency, yet Apple subtly switches to discussions of AltiVec acceleration on its Web page to tout a two-to-three time speed boost, which has a lot less to do with the over-explained pipeline differential, and more to do with far less used vector functionality. No matter how snazzy John Rubenstein's Keynote presentations are, megahertz does matter. Sure, not all megahertz are created equal, but overall system throughput and disparities can at least, roughly, be measured and compared.

However, many seem to be in some kind of denial about the woeful state of performance of OS X on G4s. In that regard, the press has been slacking. I'm not sure if they've all gone to the Apple Kool-Aid trough one too many times, or if there are other externalities at play. It seems strange to me that such performance disparity and the high prices of Apple's offerings are not more of an issue to the press. Supposedly it is their job to question and bring such issues to everyone's attention, but instead, everything is sugarcoated. It's as if Mac pundits have their heads so far up Apple's ass that they can't see things for what they are. Current Macintosh systems are slow compared to current Microsoft XP systems. Apple needs to provide significantly faster machines, and they cannot come soon enough.

Then again, maybe it's just me. As always, your mileage (and perception of speed) may vary.

is an attorney. Please don't hold that against him. This work does not necessarily reflect the views and/or opinions of The Mac Observer, any third parties, or even John for that matter. No assertions of fact are being made, but rather the reader is simply asked to consider the possibilities.

You can send your comments directly to me, or you can also post your comments below.

Most Recent Columns From The Devil's Advocate

The Devil's Advocate Archives



Today's Mac Headlines

[Podcast]Podcast - Apple Weekly Report #135: Apple Lawsuits, Banned iPhone Ad, Green MacBook Ad

We also offer Today's News On One Page!

Yesterday's News

 

[Podcast]Podcast - Mac Geek Gab #178: Batch Permission Changes, Encrypting Follow-up, Re-Enabling AirPort, and GigE speeds

We also offer Yesterday's News On One Page!

Mac Products Guide
New Arrivals
New and updated products added to the Guide.

Hot Deals
Great prices on hot selling Mac products from your favorite Macintosh resellers.

Special Offers
Promotions and offers direct from Macintosh developers and magazines.

Software
Browse the software section for over 17,000 Macintosh applications and software titles.

Hardware
Over 4,000 peripherals and accessories such as cameras, printers, scanners, keyboards, mice and more.

© All information presented on this site is copyrighted by The Mac Observer except where otherwise noted. No portion of this site may be copied without express written consent. Other sites are invited to link to any aspect of this site provided that all content is presented in its original form and is not placed within another .