Can you change your motherboard without reinstalling your operating system? Sometimes you can, sometimes you can’t. Last week I decided it was time to upgrade the desktop PC on which I do much of my work. The old motherboard was nearly three years old – an Intel 915PBL, which I purchased in order to run 64-bit Windows XP. At the time I was indignant because the BIOS swallowed a full 1GB of my RAM.

Technology has moved on, so I purchased an Intel DP35DP board (I still have a high regard for Intel’s manufacturing quality) and a core 2 quad Q6600 processor. I appreciate that this is now not quite leading edge, but it has been out long enough to be a reasonable price. So I backed up my files, performed the motherboard surgery, and switched on.

For sure I was not depending on being able to continue without reinstalling the operating system. This is not a supported procedure. On the other hand, I figured it would be interesting to try; and so it proved. After all, I’ve had enough trouble installing things like Visual Studio 6 and Adobe Creative Suite 3, not to want to repeat the experience.

My desktop PC is a little complex. There are three SATA drives and three Windows installations. One is Window XP x64 (the original installation); one is Windows XP Pro 32-bit; and one is Windows Vista Pro 32-bit. I work mostly in Vista. Initially, none of these worked with my new motherboard. Attempting to boot Vista caused an immediate restart, and both XP installations blue screened.

The key problem is the storage controller. Typically, you will get a STOP 0x0000007B if your new motherboard has a different storage controller than the old one. The above link is to a Microsoft knowledgebase article for Windows XP, and you will see that it suggests two solutions. One is to perform an in-place upgrade, also known as a repair install. This is where you reinstall Windows to the same location, and it sorts out the device drivers and other critical problems without blowing away your documents or applications. The other is to do a new clean install – not really a fix, more an admission of failure.

Vista no longer supports repair installs. More on this later.

Note that the main issue is getting Windows to boot. Once it does boot, you will likely get all sorts of device errors, but these can usually be fixed. Fixing stuff if you cannot boot at all is more challenging.

Health warning

Since you are still reading, I assume you are in one of two categories. Either you are inquisitive about the innards of Windows; or you have a problem because Windows will not boot. Here’s my disclaimer. The steps I describe below are not supported, guaranteed ways to rescue Windows; and if you experiment with a working Windows installation, you could break it. If you try any of them, it’s at your own risk.

The AHCI problem

Intel’s new boards usually have three settings for the SATA controller. One is IDE, which is best for legacy compatibility. One is AHCI (Advanced Host Controller Interface), which is best for features. The third is RAID, which combines AHCI and a sort of hardware RAID (sometimes known as fakeRAID because it isn’t really hardware RAID, but never mind). Intel boards used to default to IDE, but this one defaulted to RAID which is now Intel’s recommendation. Even if you only have one drive, you can easily set up RAID later if you start with that setting. If you plan to install Linux, you might be better off with plain AHCI.

Repair Install on XP x64

For my first experiment, I tried a repair install on XP x64. It did not go well. There was some minor awkwardness. Intel supplies the crucial RAID drivers on antiquated media called floppy disk. You need this during Windows setup (press F6). However, the motherboard has no Floppy Disk connector; you have to use USB. Luckily I have a USB external floppy which came with a laptop some years ago, so that bit worked. Unfortunately, the repair install seemed to hang on a screen which said “Installing Devices” and promised to complete in 35 minutes. I left it overnight and it said the same thing in the morning, though it had not technically crashed and who knows, perhaps it would have completed one day.

If you ever hit this issue, read this thread. What you have to do is to restart the machine and press F8 to get the advanced startup options. Timing this may be tricky, since setup configures itself for immediate startup. If you get really stuck, find some way to edit boot.ini to insert a delay. Once you have pressed F8, choose the Debugging mode. The repair install will now complete within a sane time. There’s another solution discussed in the thread I’ve referenced above, which involves identifying which device is causing the hang and then preventing setup from attempting to install it; but the Debugging option worked fine for me.

As it happens, I didn’t discover this fix until later. I got diverted by 32-bit XP.

Fixing 32-bit Windows

After getting stuck on XP x64, I took a look at 32-bit XP. The X64 repair install had mucked up the boot menu; but that’s easy to fix from a recovery console. Being a little discouraged, I decided to try the legacy IDE settings for the SATA controller. Good old XP Pro then started, no hassle, no need for a repair install. It detected various new devices, and I ran the Intel driver utility from the CD that came with the motherboard. I also had to reinstall the NVidia graphics driver. My first success.

First attempt with Vista

Vista on the other hand still would not work. It is less amenable than XP to controller changes; I gather this may be something to do with speeding up the boot process by disabling “unnecessary” drivers. Anyway, it blue-screened irrespective of the setting for the SATA controller. It was then I discovered that Vista has removed the option for a repair install. What you do instead is to boot from the setup CD or DVD and choose the option to “Repair your computer”. Then it runs a Startup Repair wizard which tries to fix your Vista. I tried this on all the SATA settings, and it failed. On the first try it thought it might have succeeded and invited me to restart Windows, but it still blue-screened. On subsequent attempts the wizard made some lengthy disk-checking efforts only to inform me that I had a corrupt volume and should give up.

I’m not impressed with Startup Repair. I think XP’s Repair Install is brilliant, and Vista’s Startup Repair is brain dead. I’ll explain shortly.

You can do a sort-of repair install with Vista by “upgrading” an existing install – but you can only start an upgrade from within the instance of Windows that you want to upgrade. In other words, if you can’t boot then this is not an option. Do NOT do an over-the-top clean install. You will lose all your settings.

At this point I nearly gave up and did a clean install of Vista. I still had the nagging thought that if only I could get Vista to load the right storage controller driver, it would probably start up. So I left it for the moment and went back to 32-bit XP, my success story. I wanted to discover if I could switch to the SATA RAID option without breaking it.

Switching an installed Windows OS to AHCI or RAID

This thought led me to an illuminating discussion. This explains that if you want to switch Windows to use a different storage controller, for which the driver is not installed, you do so by manually copying the driver to system32\drivers; and editing the registry so that Windows can use the new driver when booting. For this to work, you need to know the hardware ID of your controller. This is something like:

PCI\VEN_8086&DEV_2822&CC_0104

One way to find this ID is to clean install a second copy of Windows to a temporary location, and then look at the storage controller properties in device manager. Alternatively you can experiment with some likely values, or try several at once. The discussion I’ve linked provides some reg files which show the critical entries.

This looked interesting, so I followed the steps in the first post of the thread, shut down, then restarted with the BIOS setting changed to AHCI. Success – XP Pro started, and I was able to run Intel’s driver setup to clean up the installation.

Getting more confident now, I found the hardware ID for the RAID controller (the hardware ID changes when you change the BIOS setting) and added that to the registry. The actual driver is the same. Success again – XP now started using the Intel Matrix Storage RAID.

XP was done now, especially as I also discovered the fix for x64 described above. But Vista, the one that mattered to me most, was still broken. Nevertheless, I now had a much better idea of how to fix it.

Offline registry editing

Now is the moment to mention a little-known feature of the Windows registry editor. 99% of the time, you edit the Windows registry from within a running instance of Windows. However, the registry editor has an entry in the File menu called Load Hive. The way this works is not particularly intuitive. You select the root of HKEY_LOCAL_MACHINE, then choose Load Hive and navigate to a registry database file that is not part of the active registry. In my case, the hive I needed to load was SYSTEM, located in System32\Config. You may possibly know this if you have ever had to recover Windows when it will not boot because of registry corruption. When you click OK to load the hive, the registry editor prompts you for a key name. Enter something like “Vista Repair”. The loaded hive then appears as a sub-folder under HKEY_LOCAL_MACHINE.

There is no Save; simply make the edits you want, then select the top level of the loaded hive – eg. “Vista Repair” – and choose File – Unload Hive.

This had to be worth a try. I booted my poorly machine into Windows XP, then went to another machine on the network running Vista. I copied the SYSTEM hive from the broken Vista across the network, made a backup, then loaded it into registry editor. Finally, I manually made the changes pretty much as in the reg file used to migrate XP to a new storage controller, though what I actually did was to pattern the changes on an existing storage controller entry in the working Vista registry. Note that an offline registry has no “Current Control Set” hive. This is actually a runtime link to ControlSet01. So, for offline editing, you edit ControlSet01.

I also made the normal registry fix for enabling AHCI on Vista. Then I unloaded the hive and copied it back to my broken Vista. I also added the actual driver to system32\drivers, of course. In my case the driver is called iastor.sys, from Intel’s floppy disk.

Vista boots

The result was amazing. I restarted the PC with the new motherboard, and instead of a blue screen I got:

Press CTRL+ALT+DELETE to log on

Nothing much to add now. I logged  on, installed various drivers, and everything works. I have a quad core CPU:

and an Intel RAID storage controller:

Vista itself seems fine. It is solid and significantly faster than before, as you would expect.

Thoughts on Vista Startup Repair

Despite my success, I’m not impressed with the Vista Startup Repair. First, why couldn’t Microsoft’s wizard do what I did manually? Second, why has the excellent Repair Install option disappeared from Vista? Vista’s repair and restore options seem to presume that the hardware has not changed; yet I’d have thought a replaced motherboard would be one of the more common scenarios where you would want to repair a Vista installation.

The whole business of changed hardware is something of a dark secret in Windows backup and restore. In an enterprise environment it is not so bad, because you have redundancy, and a single failed machine (server or workstation) should not be a problem. Something like Small Business Server is different, because typically this is a single, business-critical machine. If your 2-year old Small Business Server box disappears in the night, and you purchase a new box and stuff in your backup tapes, it will not necessarily be easy to restore.

Still, I must also credit Windows with considerable resilience once it can find a working storage controller driver. It pretty much sorts itself out after that.

Coming next: a clean install of 64-bit Vista. It makes sense.

PostScript:

Anyone interested in this esoteric topic is also advised to read this article on ArsTechnica and this related thread. Note the advice on removing non-present devices from device manager. There’s also this Microsoft article (non-Vista) which is a variant on the Repair Install. Let me also note the main area of uncertainty regarding repair installs: it may not be safe with regard to recovering a fully patched system. In principle you can re-apply service packs or hotfixes, but what if Windows Update thinks a hotfix has already been applied when in fact it has been undone? It’s a good question and I’m not sure at the moment what the answer is.

The Microsoft article above also makes an interesting point about OEM licenses. If you replace the motherboard with something other than the same model, it is considered a new computer and OEM licenses cannot be transferred.

I’ve been covering Microsoft for enough years to have seen and heard Bill Gates on numerous occasions. But I’ve not done so for enough years to have seen the best of him. I gather from other journalist friends that until maybe the early nineties, Gates was excellent value for the IT press, showing his technical side and chatting in-depth about some of the details of his products. Note this comment from Joel Spolsky:

Bill Gates was amazingly technical. He understood Variants, and COM objects, and IDispatch and why Automation is different than vtables and why this might lead to dual interfaces. He worried about date functions. He didn’t meddle in software if he trusted the people who were working on it, but you couldn’t bullshit him for a minute because he was a programmer. A real, actual, programmer.

Sadly I was a little too late to see this side of Gates. Microsoft grew too big; Microsoft execs grew too distant. In the keynotes I’ve heard, he talks about the company vision and the state of computing and leaves the technical details to others to explain. He occasionally takes questions, to which he typically gives long, circuitous answers, a favourite technique used by senior execs with, I suspect, the goal of reducing the number of questions that can be asked and answered in the time available. Nonetheless I respect him for steering the company through its path from the early days of DOS through to having its products installed on nearly every desktop and in nearly every home.

What prompts this post? billg is retiring in July and confirmed this at CES:

It’s the middle of this year, in July, that I’ll move from being a full-time employee at Microsoft to working full-time at the foundation.

This isn’t news; it’s in line with a previous announcement in June 2006; even the date, July 2008, was announced then.

Is Vista really half the speed of XP? That is what CNET reports here, in an article which does not bother to link to the source of the tests, as far as I can see.

The tests seem to be based mainly on scripting Office through OLE automation. This is certainly interesting, but it would be more helpful to have a performance breakdown, looking at various aspects of OS performance. Otherwise, how do we know whether this is more to do with, say, OLE performance than Vista performance?

Second, Vista has richer graphics than XP. Think of it like a game which offers difference graphics options to suit your hardware. Typically, you can set varying levels of background detail, shadow effects, etc, in order to find the right compromise between appearance and performance. Vista sets this higher than XP by default, so you would expect screen operations to be slower.

Third, Vista’s UAC security imposes a significant performance overhead. Again you would expect that, since it applies additional checks to numerous operations.

You could conduct a more useful test by configuring Vista to work as much as possible like XP. Turn off UAC, turn off Aero, turn off the indexer, turn off visual effects. I would also suggest checking what is running in the background and matching it as closely as possible on the two machines. I would be surprised if the performance difference is so striking after doing these things.

Perhaps you will argue that this is not the real-world experience. In practice, users take the machine as supplied and start using it; they do not open Control Panel and look at Performance Information and Tools, or get rid of all that third-party foistware. It’s a fair point, though on a corporate network admins can set these things on the user’s behalf.

When it comes to real-world performance, I also have gripes about Vista. However, I have little problem with Vista’s overall performance, say when working in Office, aside from Outlook 2007 which is a disgrace. What bugs me is unexpected delays. For example, I click the Start button, then All Programs, then scroll down to a program group and click. Vista just stops. I tried it just now, and it took 12 seconds to open the group I chose. I’ve had to train myself not to keep clicking. What’s going on there?

There are real issues here, but not helped by misleading reporting of inadequate tests.