View the test results.

Our test bed for this story went far beyond the walls of our labs in New York City. We traveled to the metro areas of Raleigh, North Carolina; San Francisco; Dallas; and Las Vegas to give four services—Cingular EDGE, Cingular UMTS, Nextel Flash-OFDM, and Verizon Wireless EV-DO—a trial run.

We were able to do some of the work, however, in our own backyard; we tested EV-DO in New York City, as well as in Hoboken, New Jersey, in Bucks County, Pennsylvania, in Loudoun County, Virginia, and in Las Vegas. We tested Cingular EDGE in New York City and in Loudoun County, as well as in Durham and Chapel Hill, North Carolina. This North Carolina region, known as the Triangle, is the only area we could test Flash-OFDM. Finally, we headed to San Francisco and Dallas to test Cingular UMTS.

Testing Methodologies

At each of these locations, we tested the high-speed networks at various times of the day between 7:00 A.M. and 8:00 P.M.

Most of our tests were performed standing (or sitting) still, but we also tried EV-DO on an Amtrak train and OFDM on a fast-moving bus. Being in motion didn't seem to cut data rates by much, as long as we were able to keep a consistent signal level.

To determine the average download speed, we downloaded 1MB and 5MB text and ZIP files. The text files contained compressible data, while the ZIP file contained noncompressible data. The files resided on our PC Magazine servers in New York, which are connected to the Web via a 1.5-Mbps SDSL connection. We also downloaded a roughly 5MB file from Microsoft and ran Web-based "bandwidth testers" from two sites— www.toast.net and www.bandwidthplace.com .

To judge upload speeds, we uploaded a 1MB ZIP file to our servers. All the networks were slower uploading than downloading. With a 54-Kbps average upload speed, UMTS couldn't beat even the slower EDGE network.

Latency, or the time it takes to begin loading a Web site once you've entered the URL, is a major factor in Web browsing, as Web pages are made up of many small, separate data requests. We tested each network by repeatedly making ten "ping" requests to four Web sites: www.pcmag.com , www.disney.com , www.bmg.com , and www.mta.info . OFDM had excellent latency results, similar to our office's 802.11b/DSL network. EDGE's results were awful; the latency felt slower than the 160-Kbps data rates we got.

We tested using PC Cards in a laptop running Microsoft Windows XP Professional with Service Pack 2. For EDGE, we also hooked up several phones to our laptop via Bluetooth and USB 2.0. We factored only the 6230's results into our final data because, of the phones, only the 6230 uses the fast Class 10 EDGE.

We also tested EV-DO speeds using the Audiovox XV6600 PDA/phone. Hooking it to a laptop via Bluetooth cut its download speeds by half, to an average of 300 Kbps. (Bluetooth has a real-world throughput of 700 Kbps.) The bottleneck seems to be in the Bluetooth PDA-to-laptop connection, not in the PDA itself.

Network Surprises

On most of our tests, OFDM was a star. We should note, though, that our OFDM download speeds could be slightly misleading. We got spectacular results with at least three bars of signal, but in only two of our five test sites. Signal strength had less of an impact on the other networks.

The biggest surprise was with EDGE and UMTS: The real gap between the two right now is in latency rather than speed. While Web surfing on UMTS will feel snappier, we don't think that outweighs UMTS's availability problems and limited list of devices.