Ubuntu 11.04: i686 vs. i686 PAE vs. x86_64

John The Ripper with its DES test also favored the x86_64 kernel on both systems.

The Blowfish performance is also up on 64-bit.

With the TTSIOD 3D Renderer on the Core 2 Duo / 4GB notebook there was a small drop in performance with the PAE kernel but it was much faster on the x86_64 kernel. The Core i7 notebook had been boosted by the PAE kernel with its 8GB of system memory while the x86_64 environment was certainly the best.

In nearly all tests, the Linux 2.6.38 kernel with PAE enabled had not caused any drastic change in performance (Linus Torvalds and others in years past have cited performance issues when enabling PAE in some configurations) on Ubuntu 11.04. In tests that could exploit a great deal of system memory, the i686 PAE kernel on the System76 notebook with 8GB of system memory had resulted in a respectable performance boost, but still it wasn't as fast as a native 64-bit environment. In a majority of the tests, the x86_64 kernel was decisively better on both Intel notebooks than using the 32-bit Ubuntu software.

It is not new or surprising that the x86_64 Ubuntu is much faster than i686 Ubuntu on supported hardware, but it is somewhat surprising that Canonical continues to push the 32-bit version as the "recommended" version of Ubuntu from their web-site, etc. Most hardware shipping in recent times is 64-bit capable and there really is no reason not to use the version. Adobe offers a 64-bit Flash Player plug-in that is not updated as regularly as the 32-bit build, but it is available and there are also open-source solutions such as Gnash and Lightspark that operate under 64-bit just fine. Canonical does not even ship the Adobe Flash Player by default -- or any other non-free software that prefers a 32-bit environment -- and they have voted down changing this policy anyways (by checking "install restricted software" as the default). For those with new hardware, they are just impairing the initial Ubuntu experience with reduced performance for those who follow the recommendations/defaults.

There are more test results from the different configurations in this OpenBenchmarking.org result file.