My guess is that the OS is easily the biggest factor in system reliability. Memory may be another; with almost no price premium for Kingston and Crucial using a good brand seems to be a no-brainer. I also suspect that power supplies are underrated as a source of problems, and may be (along with a good UPS) the real key to system longevity. My last system to bite the dust died not long after I had to replace the power supply, and I suspect there was damage done by that power supply as it died.

Good peripherals may be underrated also. After trying various setups and experimenting with varous software on my Red Hat 7.2 box I decided that the easiest way to get back to a clean install was to scrub down and do just that. Four or five straight attempts failed at various points, abruptly aborting out of the install program. I have 2 cd drives in the system, a fairly fast el cheapo several years old and a name brand but very slow, somewhat old cdrw. The cdrw has usually worked when I’ve occasionally had problems installing something with the faster el cheapo.

I tried both drives, in both graphical and text mode, with no success. Finally, having a second unused older hard drive in the system, I decided that I’d better do an install from the second hard drive. I’d done my original install with downloaded ISOs, but now had no working system on the PC to use for downloading. I’m guessing that there’s probably some decent dos ftp programs still around, but I decided the easiest thing was to download to my other (Windows 2000 pro) PC, then use Ghost.

I ghosted the partition that I downloaded to a partition on the second Red Hat hard drive and tried to install. The install program didn’t recognize the partition. At first I thought that the install program didn’t recognize it because it was an extended partition on the drive, but eventually I realized that when I ghosted I inadvertently converted the drive to ntfs. Trying again, with a fat partition worked like a charm, and the install went without a glitch.

Long story, but I suppose that the lesson is that older/cheaper peripherals cost me a lot of hours.

Cheap motherboard designs transcend chipsets. The very worst motherboards I’ve ever seen were based on Intel 430VX chipsets. The VX was a decent low-end chipset for its day, but these particular boards were of such shoddy quality that the virtues of the VX chipset as opposed to any other chipset were completely covered over.

I have run various motherboards with VIA chipsets in them since 1995, starting with 486s and moving on up. I’ve run systems with VIA and Intel chipsets side by side since 1999, and I can’t say I’ve noticed any demonstrable difference in stability. All of the systems have crashed mysteriously on a couple of occasions. And there have been times that I’ve done something stupid that I figured probably would crash the system and it didn’t. Was it the fault of Windows or the hardware? More likely it was Windows.

At work, we mostly have Intel CPUs on Intel motherboards. We bought a batch of Intel CPUs on Tyan boards with VIA chipsets last spring. The default BIOS settings needed a little tuning to run NT4 reliably, but I have yet to see any indication that those systems are any less reliable than the strictly-Intel based systems.

This Web server is running on a Soyo motherboard with a SiS chipset in it. It’s been running 277 days straight. And SiS is supposed to be the worst of the chipset makers.

I am aware of bugs in VIA chipsets. I’m also aware of bugs in Intel chipsets. That’s why it’s important to get current drivers when you install the OS.

In my experience, when you buy a motherboard from a reputable maker (I prefer Asus; I’ve also been successful with Abit, FIC, and AOpen boards, but I’ve yet to see an Asus fail) and pair it with good-quality power supplies, memory, and IDE cables, and if you run Windows, stay reasonably current on your drivers, and you won’t have much trouble regardless of the chipset.

Frequently when people have problems they attribute to a particular component (VIA chipsets are in vogue now; Cyrix CPUs are another scapegoat of ages past that come to mind) there are other possibilities involved. Are they using brand-name memory? Are they using knockoff video and/or sound cards? Are they using no-name power supplies? Frequently the answer to those questions is yes.

Take multiple systems with identical power supplies, memory, video cards, CPUs and cases, with the only variance being the motherboard and the chipset on it. Ideally, testing should be done on a mature Linux or FreeBSD kernel to eliminate as much possibility of OS stability issues as possible.

Now, in that environment, show me an intense process that will run reliably on the systems with Intel chipsets but not on the systems with VIA chipsets. Then I’ll be convinced. So will any reasonable individual. I haven’t seen that test yet.

Pournelle’s just repeating what people have told him. I suspect he’s repeating what one person has told him. That person has been talking poorly about VIA chipsets for a couple of years, but he’s never cited any test (certainly not like the one whose parameters I just outlined) to corroborate his statements.

Without seeing the methodology, we know nothing useful. And stating something over and over doesn’t make it fact.

Pournelle knows the scientific method and I’m disappointed that he didn’t ask for backing before publishing that statement about VIA chipsets. He states that if the problems are numerous enough to mention, they’re too high. But you have to have something worth measuring. Since VIA chipsets cost less, they are more likely to be paired with questionable components than their Intel counterparts. That’s why a valid stability test would have to eliminate those other possible variables.

CPU temperatures are another topic. Fortunately it’s also a simpler topic. Cooler is generally better. AMD’s Athlon series tends to run hotter than Intel’s P2/3/Celeron series, but have you seen the monster heat sinks and fans for the P4? And both of them run alarmingly hot compared to Motorola’s PowerPC. Apple enthusiasts tout this as a huge advantage, but while there’s no doubt that Apples are more reliable than Packard Bells or eMachines, I’ve never seen anything that demonstrates Apples have more longevity than the business-class PCs from Tier-1 PC makers. I used to worry about CPU temperature but now the only thing I care about is whether it’s running within the design specs.

As for AMD vs. Intel, I’ve followed Jerry Pournelle’s recent discussion. It sounds like the informed opinion is that there are perhaps small differences in reliability, likely of more concern in an server environment where minimizing downtime is a big deal. The impression that I get from Pournelle is that any measurable differences have been more likely due to AMD chips being matched with poorer quality motherboards (via chipsets). I don’t keep up with hardware in that much detail, but I understand that there are now more quality motherboards for AMD, so that this shouldn’t be much of an issue now.

I’ve owned 3 AMD powered systems, and now have my Linux box on my first Intel CPU. I found a 1MHZ Celeron bundled with a MB at Fry’s and price was an object, so I bought it. The one really nice thing about it is that Intel chips run cool. At first I had Win2k on the box, and a utility to check CPU temp…it was running about 80 degrees F in a room that was probably 75 (Antec case with good ventilation and whatever heat sink/fan they bundled with the CPU). Aside: I don’t know how system-specific such things are, but anyone know of a good open source CPU temperature utility for Linux?

I hate to wait on slower computers as much as the next guy, but I’ve had several cheap computers bite the dust in 2-3 years. I’d like to have my older computers stay alive and make them into Linux routers or non-gui fileservers or donate them to charity. Tom’s hardware site had an article about how they pulled the heatsinks off Intel and AMD CPUs and the AMD ones melted down, taking the mobos with them. The Intel CPUs just slowed down or shut down with no damage.

I’m inclined to think that Intel and AMD are about equally reliable day-to-day given quality motherboards mated with both. What I wonder about is the long term effects of greater heat, even with better heat sinks for the AMD CPUs.

I prefer choice, and want AMD and Linux to prosper as alternatives to Intel and Microsoft. However, I’m hesitant to buy another AMD CPU until I’m a I’ve learned a little more about the effects or lack thereof of the higher temperatures that they run at.