Windows XP Inside & Out

Lab Tests: XP Performance--Satisfactory, Not Stellar

Microsoft touts XP as the fastest incarnation of Windows ever. But in our tests, we found its performance generally on a par with that of other recent Windows versions. It may not be a reason to upgrade, but neither is it a reason not to.

The PC World Test Center compared XP Home Edition and Professional with Windows Me and 2000 by putting all of the OSs through a battery of hand-timed application performance tests (see "How We Test," below). The XP-compatible version of PC WorldBench wasn't ready in time for use in this story.

We used two PCs representing the low and medium-high ends of the current market: an 800-MHz Celeron PC and a 1.4-GHz Athlon model. We tested them both with Microsoft's recommended minimum of 128MB of RAM and with 256MB. Since the difference in the two systems' results reflected only the Athlon's faster speed, we omitted the Celeron figures here. We also found no performance difference between the two versions of Windows XP.

Memory, generally considered a low-cost performance pick-me-up, had little impact except in the memory-intensive Photoshop 6 tests. If you spend a lot of time using Photoshop or other RAM-hungry applications--or if you typically run many active applications at once--upgrading to 256MB of RAM should help regardless of which Windows you use.

One place where our testing revealed a perceptible performance difference was in start-up and shutdown times. Windows 2000 took more than 20 seconds longer to boot, because it's much larger than Windows Me, and because XP uses new optimizations for reading and loading OS code into memory. Windows 98 users may see a marked improvement in boot-up times, too, since that OS doesn't support Fast Boot BIOSs as the three later Windows do.

At shutdown, Windows Me outperforms the others by unceremoniously dropping network connections.

To test how Windows XP's Fast User Switching--its ability to let one user's apps run in the background when one or more other users log in--affects performance, we ran the same Word 2000, Access 2000, Notes 5, and Photoshop 6 tests shown in the chart, but with a second user logged in and with Netscape Navigator, Lotus Notes, and Windows Movie Maker still running. Though we anticipated that this might degrade performance, and that additional RAM might restore the lost performance, we were wrong: We got virtually the same results with both 128MB and 256MB, whether a second user was logged in or not.

Still, your use of Fast User Switching may tax your system more severely. If you plan to keep more users logged in or to run more apps, extra RAM may keep your foreground user accounts running faster. And if you plan to use Fast User Switching on less-powerful systems with less RAM, be prepared for slowdowns.

How We Test

We tested each OS by hand-timing boot-up, shutdown, and common tasks in Microsoft Office 2000, Adobe Photoshop 6.0, Lotus Notes Release 5, and Netscape Navigator 4.08.

Microsoft Word: We timed application start-up, search and replace, auto-summarize, and mail merge.

Microsoft Access: We timed application start-up, importing a database, creating a form, and running a report.

Adobe Photoshop: We timed application start-up, opening a JPEG file, applying a filter, and exporting a file for the Web.

Lotus Notes: We timed application start-up.

Multitasking

We also conducted a multitasking test which consisted of timing the download of a large file with Netscape Navigator 4.08 in the background while separately timing three of the Access tasks (Import, Create Form, and Run Report) in the foreground.

To test Fast User Switching, we logged in as one user, started up Netscape Navigator, Microsoft Movie Maker, and Microsoft Word. We then used Fast User Switching to log in as another user, and ran our hand-timed application tests as previously described.

Equipment and Specs

We ran all tests on two PCs: a Micron Millennia Max XP2 with a 1.4-GHz Athlon, GeForce3 graphics card, and a 60GB hard disk, and an EMachines T1801 with a Celeron-800, integrated Intel i810 graphics, and a 20GB hard disk.

Testing on each system was done first with 128MB of RAM, then with 256MB.