Thank you everyone who has contributed towards analyzing this bug for Jaunty.  As many of you know, the Intel performance issues were not completely fixed for the Jaunty release, and I understand you are probably as frustrated as I am that it remains an issue for so many people.  To those who have remained patient through this process, it is especially appreciated.

For the background behind why all this regression happened to begin with, I'd best refer you to Keith Packard himself (you may want to skip down to the "Pick One From Each Column" section, if you're not so interested in the technical stuff):

    http://keithp.com/blogs/Sharpening_the_Intel_Driver_Focus/

It is unfortunate that Jaunty (and Intrepid, to a lesser degree) hit in the midst of this major rearchitecturing work, and suffered the consequences as a result.

One thing that is important to note is that not everyone sees the performance regression.  Some have no problem whatsoever.  Some see it only with compiz, and are fine sticking with 2D.  Some see a slight regression but not so bad that it affects their workflow.  This is not to trivialize the importance of having good performance with compiz (since we ship it on by default).  As well, there are a sizable number of people who have significant performance problems both with 2D and 3D.

Given all the construction work being done upstream, the obvious question is why not to keep Ubuntu to an older, more stable version of the -intel driver such as 2.4 or 2.2.  There are four reasons.  First, it would undo fixes that were gained in 2.6 to bugs that were even worse than this performance problem.  Second, some newer Intel hardware was enabled in 2.6, which would be lost if we moved to an older driver.  Third, it would inhibit our ability to work with upstream to gain real fixes to the problem.  And fourth, it is the wrong thing to do.

The right thing to do is to figure out what causes the problem, and fix it, and that is what many of us have been quietly working on the past couple months.

Let me explain what we have done to date, the current status, and the plan going forward.

In our testing, we've uncovered a lot of different ways to work around the problem.  Sadly, there is no single workaround which solves the issue for everyone across the board.  In some cases there has been sufficient consistency (such as a workaround that worked for particular families of chipsets) that we were able to roll out the change.  In most cases, however, our testing found the workaround only seemed to solve the problem for some cases, and caused new problems (corruption, freezes, crashes...) for other cases randomly.

I generally consider stability to be higher up on the priority list than performance, so where we have an option that would give performance at the expense of stability I've opted to pass it by.  Because the driver seems to be so sensitive to changes, we're being very deliberate in doing thorough testing; last thing we want is to rush some change out that causes more damage than it solves.

However, since these workarounds do in fact help people in some cases, I've capturing them in the Ubuntu-X wiki.  My feeling is that even if we can't use them for Ubuntu as a whole, by getting the information out there others can benefit from it to mitigate the problems locally while we keep searching for a true fix.  This documentation is located here:

  https://wiki.ubuntu.com/X/Troubleshooting/IntelPerformance

I hope for anyone still having performance problems, that one of the ideas contained there will help you out of the hole.

So what's the plan going forward?

We've known that there are upstream kernel patches that were introduced as _real_ fixes to the problem.  Since they're kernel patches, they're a bit harder to test, however we've found they do indeed address the problem.  There were concerns that they cause secondary issues (freezes, etc.) so the kernel team didn't include them for jaunty, and those issues will obviously require further study so we don't end up causing worse problems.  However, they're still on the kernel team's radar to give ample review and attention, and if we are lucky and they don't cause other regressions then we may see them available as SRUs.  I will keep you updated as I know more.

So, meanwhile you can help by being patient, helping keep discussions about this civil, and disseminating useful information on workarounds to affected users.