Last week, when I published my preliminary test results documenting start-up times for Windows Vista, the most common comment, by far, was that my test was unfair because I used hardware that was too powerful. Personally, I think this particular Dell system (which costs around $1200 and is far from their top of the line) is pretty mainstream, but more data never hurts. So I went back to the lab and performed similar tests on two more systems.
First, a recap and an explanation. The original story in ComputerWorld was based on quotes from end users on the Windows Vista Performance and Maintenance Forum, complaining about dreadfully slow start-up times for machines with Windows Vista installed. Now, do I think those people were lying? Of course not. I’ve seen systems that take an exorbitantly long time to boot up, and I’ve usually been able to resolve the issue by finding the configuration problem or third-party application that was dragging down performance.
Do those randomly selected posts represent a trend? Not necessarily. Computerworld could just as easily have searched the Windows XP Support and Maintenance forum and found dozens of similar complaints about Windows XP. Or they could have gone to various Mac support resources, where they would have been overwhelmed with recent examples of people struggling with slow startups on their OS X machines.
Unfortunately, ComputerWorld didn’t include any testing to try to duplicate those results under controlled circumstances. Which is why I’m playing lab rat.
In my earlier controlled tests on a Dell XPS 410, I found that it took 61 seconds to load Windows XP and 72 seconds to load Windows Vista. Those results were consistent. But can they be extrapolated? What happens if I run a similar test on less robust hardware?
To answer that question, I added two more test systems to the one I tested last week. In addition, I performed startup tests on several other systems running Windows Vista without installing XP as a control. (For details of the methodology I used, see the end of this post.)
These are not sterilized, tweaked test systems. They’re machines I use all the time, including the ones on which I do daily work. Two of the three systems were using older, single-core CPUs and had a total of 1.5GB of RAM. (For detailed system configurations, see the end of this post.)
The results prove to my satisfaction that Windows Vista is not inherently slow. I was consistently able to achieve startup times on Windows Vista machines ranging from 44 seconds to 1:39 (the longest startup time I recorded was 2:04). In one of the three cases, Vista was much faster than Windows XP running on the same hardware. Here’s the data (all times in mm:ss, rounded to nearest second):
Vista: 1:12 (15% slower)
Vista: 1:20 (34% faster)
Windows XP Professional: 0:58
Windows Vista Home Premium: 1:14 (22% slower)
Ubuntu Linux 6.10: 1:49 (47% slower)
Additional systems (Vista only)
Dell XPS 210, Intel E6700, 2GB RAM, Vista Ultimate: 0:44
Dell Inspiron 6400, Intel T2050 (dual-core), 1GB RAM, Vista Business: 1:39
HP TX1000, AMD Turion TL-60, 2GB RAM, Vista Home Premium: 1:05
I’ll have a final post drawing some conclusions from this data and from my testing tomorrow.
Here’s a summary of how the three systems were configured:
System #1 is the Dell PC I used for last week’s tests. It has an Intel Core 2 Duo E6600, 4GB of RAM, and a 160GB 10,000 RPM SATA drive. I used Dell-supplied disks to install clean copies of Windows XP Professional and Windows Vista Business Edition. This system was built in 2006 and it’s one of two desktop systems I use for daily work.
System #2 is an Acer C310 series Tablet PC, designed and built in 2005. It includes a 2GHz Pentium M (a single-core CPU), 1.5GB of RAM, and a 100GB ATA hard drive. It’s exceptionally well built and is my primary traveling computer. For my tests I used the original OEM installation of Windows XP Tablet PC Edition as supplied by Acer, and upgraded that installation to Windows Vista Business using Acer’s Upgrade Assistant.
System #3 is a homemade PC using hardware that was common in 2003/2004. This is the same system I’m using for my experiments with Linux, with one crucial difference: I removed the add-in SATA controller and disabled the HPT370 RAID controller, reverting to the onboard IDE controllers. The system has a 2.8GHz Pentium 4 processor, 1.5GB of RAM, and two functionally identical 200GB ATA100 drives with very similar performance characteristics (one drive is for Linux, the other for Windows). I used retail copies of Windows XP Professional and Windows Vista Home Premium.
My methodology was consistent on all machines. I used a stopwatch, which I started as soon as the first BIOS text appeared on the display . If the system stopped at a logon dialog box, I paused the timing, entered the password or clicked the logon icon, and resumed timing as soon as I pressed Enter . After the desktop appeared, I immediately clicked the default browser icon (Internet Explorer 7 on all Windows machines, Firefox 18.104.22.168 on the one Linux machine) and stopped timing when the start page was fully loaded. I repeated this test for each system until I had three consecutive consistent results and then took the average of those results; I ran the test a minimum of 6 times per machine.
 Timing from the point of pressing the power button would have added no more than 2 seconds to each test.
 Results were similar on systems where the primary user account had no password or where auto-logon was enabled.