Before it can tackle Windows, Chrome must leave Safari in the dust

Just a few months ago, Apple Safari 4 could stake a claim to being the fastest Web browser available for Windows. But although its speed has improved even since then, especially in the second update since its official launch released late Wednesday, Safari is now as much as 30% slower than the latest beta of Google Chrome 3, released the following morning. This according to Betanews tests completed late Thursday.

Since the report of last week's score following the release of the first stable Mozilla Firefox 3.5 browser, we've seen a security update for Safari 4 Beta, three updates to the Opera 10 beta, and two major improvements to Chrome 3. The first of those two releases saw major improvements to window handling routines -- the first clear indication that Google is working to build a windowing environment of its own. The second, released this morning, brings a revised layout to the default "New Tab" window (an alternative for the user's home page), allowing more thumbnails in the frequented sites list.

Also we've seen some updates for Windows Vista. Installing those actually improved our Internet Explorer 7 scores in Vista SP2 by a slight amount, forcing our Betanews index scores for the stable Opera 9.64 and Safari 3.2.3 browsers down slightly (our index compares all Web browsers' performance to that of IE7 in Vista SP2).

After these latest updates, both Safari 4 and Chrome 3 posted slower index scores in Windows 7 RC (slower than the Vista speed adjustment would mandate), but faster index scores in Windows XP Professional SP3. Chrome 3 posted a record index score of 18.01 in XP, running real-world benchmark tests at 2123% (no, readers, there's no missing decimal point in that figure) the speed of IE7 in Vista. That's 46% faster than Chrome 3 in Vista SP2, whose score is 12.97. As of now, the average Web browser runs 29% faster in XP SP3 than Vista SP2.

The latest Safari is also showing impressive speed gains even over its immediate predecessor, scoring 16.16 in XP and 11.77 in Vista. The gap between those two platforms is about the same as for Chrome 3, at 45%.
Meanwhile, the development tracks for the first bug fix of Firefox 3.5 and the experimental build of 3.6 appear headed in opposite directions. While the bug fix could conceivably make 3.5 almost 2% faster, the indications from the 3.6 Alpha 1 nightly build are that developers there are trying stability improvements as opposed to speed. Thursday's nightly build of 3.6 continued heading south in the speed department, now at 7.01 on the index in Windows 7 RC versus a 9.21 score from the stable 3.5 on the same platform.

We'd like to be able to accurately test nightly improvements to the WebKit rendering engine being produced for Safari 4. WebKit's nightly builds are designed to install on top of an existing Safari installation, replacing just the rendering and processing components while keeping Apple's front end. In the tests we have conducted, we've seen clear evidence that Safari's math and string processing speeds could catch up with and even exceed those of Chrome, as some of the WebKit nightly build's benchmark scores are notably faster.

Some. The problem is, the renderer in the WebKit nightly builds appears to be a temporary solution, so its DOM and AJAX scores are too slow to be counted as legitimate -- far slower even than IE7 in Vista. The results would render the WebKit build's final index scores irrelevant, though there does appear to be fragmentary evidence that Safari could eclipse Chrome's speed at some point, perhaps soon.

We've also taken a suggestion from a few readers and started investigating the V8 benchmark suite, which Google assembled for testing the relative performance of Chrome. (Some of our readers, in response to our calling the SunSpider benchmark suite "independent," note that it was developed for the WebKit renderer used most prominently in Safari.)

As expected, Chrome performs better in the V8 suite than every other browser. But the speed gap it records of almost exactly two-thirds between Chrome 3 Beta and Safari 4, and the 951.5% performance gap between Chrome 3 Beta and Firefox 3.5 (again, there's no loose decimal point in that last score), seem to us uncorroborated by real-world experience. Chrome just isn't ten times faster than Firefox. We suspect that the speed gaps registered by the V8 suite, rather than linear in nature, are actually exponential, such that higher scores are more...shall we say, pronounced. For now, we're comfortable with the performance suite we're currently using, though we continue to investigate possible alternatives and improvements.