In early April, we spent a day driving around San Francisco and measuring the performance of the major carriers around the city on their 3G networks and the newer 4G networks. We were also able to include Verizon's brand new 4G network as they have just recently released their first 4G handset, the HTC Thunderbolt. Over the course of the day, we conducted more than 550 speed tests measuring connection quality at 9 different locations around the city.
From our results its clear that 4G has well and truly arrived. Data speeds on 4G were far higher than the 3G networks across the board in every aspect of the testing; download speeds, upload speeds & ping tests. However, the carriers confusingly place the 4G label on a lot of different things. Verizon and Sprint have both rolled out new technologies (LTE & WiMax) whilst T-Mobile has simply upgraded their 3G networks to a slightly improved version (HSPA+) and performance varies between these technologies.
Verizon has been slow to market their 4G network, but our reports show that they've built out a next-generation network that is head and shoulders above the competition. Their LTE service is significantly more responsive than any other cellular network (including Verizon's own 3G services), and offers download and upload speeds that mark a significant improvement over older technologies. Its the closest the mobile experience has ever come to matching what you can get at home with a broadband/WiFi set-up and is likely to herald a new class of mobile applications and devices that take advantage of this. Most significantly, Verizon's 4G network is built on the 700MHz band that was auctioned in 2007; this lower frequency compared with the 1900/2100MHz PCS and AWS 4G services offered by T-Mobile and Sprint means that coverage was generally excellent everywhere we tested. If Verizon can maintain the same quality as more users adopt the new network and the airwaves become more congested, customers will be very pleased with their new offering.
3G: 414 kbps
4G: 4290 kbpsUpload
3G: 538 kbps
4G: 4640 kbps
Verizon's 4G service consistently had the fastest service at every location we tested, averaging 4.6Mbps on downlink and 4.3Mbps on uplink. It should be noted that Verizon's 4G service benefits from having very little network congestion since the service was only recently launched; it'll be interesting to see how Verizon's performance scales as more users hop onto the upgraded network. Notably the 4G service also had considerably lower latency than any of the other 3G or 4G networks, which means subscribers can expect a considerably more responsive browsing experience than with competitors.
However, Verizon's 3G service also performed very solidly in our tests, performing faster on average than Verizon, Sprint, AT&T or T-Mobile's 3G services. Based on these results, downloading or uploading a 4MB MP3 file using Verizon's 4G network would take just 8 seconds, compared with 42 seconds to download and 190 seconds to upload on AT&T, and 58 seconds to download and 79 seconds to upload on Sprint's 4G network.
3G: 764 kbpsUpload
3G: 169 kbps
AT&T have made it clear that they're making large investments to improve their coverage in the San Francisco metro area, and it's evident that their efforts are beginning to have some effect. In our tests, AT&T averaged 764kbps download and 169kbps upload speeds. However, their upload speeds slowed to a crawl in a few of our tests, in particular at Civic Center (53kbps upload) and SoMa (58kbps upload), two areas where we've anecdotally heard a number of complaints about signal problems.
Unfortunately, AT&T's latencies were the highest out of the networks that we tested, averaging 1114ms and peaking at 1931ms in the financial district. The higher latencies mean that despite raw download speeds being decent, browsing websites on an AT&T smartphone can be frustratingly slow.
3G: 334 kbps
4G: 556 kbpsUpload
3G: 247 kbps
4G: 405 kbps
Sprint managed to fare quite solidly in our tests, though neither their 3G nor their 4G service was anything to write home about. The 3G services seemed to fare particularly poorly in the Mission, managing a weak 127kbps download speed and 44kbps upload speed at Delores Park. Their 4G WiMax services seemed quite consistent, but speeds were 10 times slower than Verizon's network in most areas where we tested.
Latencies on Sprint's 3G network were comparatively fast at 618ms, but their newer 4G WiMax network clocked in at 502ms - a far cry from Verizon 4G's 208ms times. This means that the browsing experience on Sprint 4G smartphones will often be considerably less responsive than on Verizon's 4G phones.
3G: 40 kbps
4G: 665 kbpsUpload
3G: 46 kbps
4G: 571 kbps
T-Mobile's results were really a mixed bag. Their 3G performance was shockingly terrible, with averages of just 40kbps for downloads and 46kbps for upload. At those speeds, it would take more than 11 minutes to upload or download a 4MB mp3 file, compared with around 70 seconds on any of the other carrier's 3G networks. Their 4G network, however, fared much, much better, with average download speeds of 665kbps and uploads of 571kbps.
Unfortunately, latencies on both the 3G and 4G networks were very high: on average, connections took around 1 second to be established. For end users, this means a laggy web browsing experience, despite the high 4G download speeds. In conclusion, if you're on a T-Mobile contract in San Francisco, make sure to upgrade to a 4G phone, and if you can, switch networks.
For experimental fairness, we wanted to make sure that we used the same phone to test all the different networks. The only Android phone currently available on all the major US carriers is the Samsung Galaxy S. For the experiment we purchased six Galaxy S phones, one from each carrier and two each from Sprint and T-Mobile (one each for 3G and 4G). However, since the HTC Thunderbolt is the only Verizon 4G LTE phone currently available, we were forced to use that model for testing Verizon's 4G network. We modified our existing OpenSignalMaps app specially for these tests, allowing the app to choose the file sizes for the download and upload speed tests based on the connection speed.
Driving around, we tried to pick out locations that we thought would be fairly representative of the whole city. For each location, we performed five consecutive speed tests on each phone - these results were averaged for the map view shown above. There was certainly a bias towards locations with available parking, since not all tests were conducted at the same time of day, network congestion isn't factored in. However, we did perform tests across all the networks at each location simultaneously, so the overall results are still valid.