How to build a reliable PC.

We touched on the topic of reliability last week. I figure I might as well give a more thorough discussion of what makes a PC reliable.
1. Power supply. I see more power supply failures than any other single component. Good power supplies fail without a whimper and don’t damage the rest of your equipment. Bad power supplies take other stuff with ’em when they die. Antec and Sparkle are examples of good basic power supplies. The power supplies that come in InWin and other brand-name cases tend to be fine as well. A notch above that is Enermax, maker of the ultimate in show-off power supplies, with plated finger guards and odd colors. Top-tier is PC Power and Cooling. If I wanted to build a computer and have absolute assurance it would still work in five years, I’d start with a PCP&C or at the very least, an Enermax.

Buy more wattage than you think you need. The power supply will run cooler and last longer if you do. Besides, you never know what you’ll want to stick in the case down the road.

2. Memory. Last time I checked, you could get 64-meg PC133 sticks for under $5. I wouldn’t trust ’em with my archenemy’s work though. Cheap memory may be untested, the PCB may not be a good design, or even worse, it may have chips that were tested and deemed unsuitable for use in PCs (but fine in other less-demanding devices). Unscrupulous makers sometimes buy up these chips and take their chances. It may seem foolhardy to pay $100 for a 256-meg stick from Crucial, but I haven’t just heard horror stories about commodity memory. I’ve seen it with my own eyes. I’ve had more than 1,000 brand-name modules cross my desk. Three were defective. I’ve had fewer than 50 commodity modules cross my desk. More than half proved defective. Some wouldn’t even work–the system would just beep at you. The worse ones appeared to work for a while, but the system was always crashing. Don’t take chances on your memory. I tend to buy my memory over-spec as well. Even if a motherboard takes PC100 memory, I go ahead and buy PC133 CAS2 memory. The chips will run just fine at a lower speed, so I have an overengineered system for a while, and if I ever upgrade I’m more likely to be able to take the memory with me.

3. Motherboards. Buy brand-name boards. I’ve never had an Asus board fail. (Watch one fail next week now that I’ve said that. But I’m happy with the reliability and longevity of Asus boards.) I’ve done well with other brands too, like AOpen, Abit, FIC, and Tyan. I know MSI boards are popular but I don’t have any personal experience with them. Asus has impressed me with their farsighted engineering–in my experience, you’re more likely to be able to upgrade an Asus board in three or four years than others.

Most people know to check the hardware enthusiast sites when researching a board. I urge you to also check the Usenet newsgroups. You’ll find some good advice. Finding very little on a board can be a good sign too-it’s an indication that a board doesn’t have many problems. Years ago, I was researching the Asus SP97V motherboard, because it was dirt cheap, but it was an Asus. I searched on Usenet and found very little about it–maybe a half-dozen messages. Most of it was just idle chatter. One message was talking about various boards, including the offhanded comment, “The SP97V is a good board for the money, BTW. I’ve used three of them.” That clinched it. Nobody was talking bad about the thing. I had one positive, and very little talk overall, which generally indicates satisfaction. Satisfied people rarely talk about stuff unless its quality blows them away.

4. CPU fans. Never go cheap on CPU fans. There’s a humongous roundup of currently available fans. Get a heavy-duty fan, even if you don’t overclock. Remember, the CPU you’re protecting is a lot more valuable than the fan. A good fan will keep your CPU well within its specified operating temperature range, and I’d like to think that the pricier fans will have a longer life. Get a ball-bearing fan rather than a sleeve-bearing fan; a cheap sleeve-bearing fan is quieter but it’s also likely to conk out on you in a couple of years if you leave your systems on 24/7.

Bookmark that site, by the way. Dan’s one of the better technology writers out there today, and he doesn’t take himself too seriously. He’s an entertaining read, explains things well, knows what he’s doing (and he’s pretty open about his methodology), and he’s probably a certifiable genius, but he’s not pretentious. In fact, he seems to enjoy making people think he’s not quite sane. I make sure I pay that site a visit at least twice a week.

5. Case fans. It’s a good idea to put a supplemental fan in the machine. Two is usually overkill unless you’ve got some really hot hard drives, and it’ll make your computer louder. You can quiet them by manipulating the voltage. Dan’s Data talks a lot about them too, including how to slow them down. For typical users, a simple ball-bearing case fan is sufficient.

6. Hard drives. IBM currently recommends you not run their drives more than 8 hours a day. So that eliminates IBM from the running. That’s a shame, because they used to make spectacular drives. (I still like their laptop drives better than any others I’ve seen though, and I’m not the only one.) I’ve seen fewer dead Quantum and Maxtor drives than any other brand, although Samsung really has surprised me with their reliability, and the drives are cheap. Seagate has a good reputation but I have very limited experience with their recent drives. Maxtor’s a safe choice at the mid range and high end, while Samsung is tough to beat for the low end.

7. Cabling. The cables that come with brand-name PC motherboards seem to be of good quality, as are the cables I’ve seen bundled in Maxtor retail kits. If an IDE cable looks flimsy, don’t buy it. Problematic cables slow you down due to the need to retransmit data. Also never buy an IDE cable that’s longer than 18 inches. Longer cables are available, but IDE specs state 18 inches as the maximum. Longer cables may work, but it’s questionable. If you have to reach the top bays in a tall tower case, you’ll have to go SCSI. Sorry.

Rounded cables will improve airflow, but be careful. Rounding shortens cables, so the wires inside a long rounded cable are even longer than stated. While a relatively new practice on the desktop, I saw rounded SCSI cables in IBM servers and workstations as long ago as 1995.

04/04/2001

A great hardware site. I found this yesterday when I was searching in hopes of remembering a long-departed name of a hard drive manufacturer. The name I couldn’t put my finger on was Miniscribe. The great site, http://www.redhill.net.au, is a hardware guide, written by an experienced Australian clone shop, that’s unusually straight-shooting. It’s the only mention I’ve seen on a hardware site of the Gigabyte 7IXE4, a low-end Duron/Athlon board that sells for about 80 bucks.

Especially interesting to me is the history. They discuss drives in detail, and though it’s hardly a complete memoir of every drive that was ever on the market, it hits the common ones. Want to know where Western Digital got its sterling (and not very deserved) reputation? Read on. I’m not so sure of their statement that Maxtor was bought out by Hyundai (Maxtor certainly never mentions that), but their history seems about as complete and accurate as any other I’ve seen, and it’s interesting to read the reviews of ancient hard drives. At least to me.

Motherboards and CPUs get a similar treatment. Good stuff.

Monitors. My NEC FE950 finally came in yesterday. It’s gorgeous, and takes up the same amount of desk space that early 17-inchers took. Mine looks like it got pretty banged up either in manufacturing or shipping though, so I’ll have to arrange an RMA. I hate to be picky, but after spending $400 on a monitor, I don’t want something with a beat-up case. It could be a cosmetic flaw, or it could be an indication that this monitor had an incident with a forklift. I’m not taking that chance.

The vendor will advance me a replacement, which is good. I’ll probably opt for that. I hope I don’t get nailed for shipping, but I suspect I probably will. Lesson learned: Order from Staples.

As for the monitor, I can’t tell much difference between it and a Trinitron. I’m not sure if it’s using a Mitsubishi DiamondTron tube or something of NEC’s design (NEC and Mitsubishi merged their monitor operations last year), but whatever it is, I like it. Fabulous monitor, and great value for the money. As far as I can tell, it’s indistinguishable from the FP series other than the FP’s higher maximum resolution, which isn’t comfortable for the monitor’s size anyway. The only fault I can find with this FE is what appears to be an incident with a forklift or some other heavy machinery.

Once again I should emphasize this point: never ever scrimp on a monitor. It might be tempting to get a no-name monitor so you can afford more memory or a faster CPU, but memory and CPU prices drop much faster than monitor prices, and they always have. Plus, their useful life is much shorter. A good monitor can outlive two or three computers, so in the long run, you save money with a premium monitor.

Why was I stumbling over the name Miniscribe? I was recalling my first-ever building of a PC. I was salvaging parts from a 286 with a blown power supply. I couldn’t get a replacement power supply because it was a Samsung PC, largely proprietary. The power supply had cooked itself because a poorly placed IDE cable totally blocked its vents, so it never had proper cooling. This was 1993, my first year of college. The PC was owned by my fraternity. We went and bought a barebones 386DX-25–just a motherboard in a case–and went to work. The video card and floppy drives and I/O cards moved without a hitch. But the Miniscribe 40-meg IDE drive gave us problems. I couldn’t get it to work, and I doubted I had much future building PCs. I took it into the shop, and they couldn’t make any sense of it either. Their most experienced tech remembered that Miniscribe had been bought out by Maxtor, so he called a contact at Maxtor. The drive turned out to be an 8-bit IDE drive that worked on some 286s but would never work in a 386 or better. They took the drive in trade for a used 40-meg IDE drive using the more conventional 16-bit interface and transferred the data for us. We got a couple of years out of that PC, though it was never a speed demon. But it was functional and cheap.

Of course I recovered from those early stumbles. Within a couple months I was selling PCs, and a couple of months after that I was working as a tech myself.

The wages of spam. And finally, I saw yesterday that Nasdaq suspended trading of PSInet and the company is considering bankruptcy. Excellent. During its not-troubled-enough life, PSInet was frequently accused of operating a safe harbor for spammers, and back in the days when I bothered to try tracking down spammers, I traced large percentages of it to PSInet.

I do not like green eggs and spam. I do not like them, Sam I am.

Hopefully this is the start of a trend. I’ve seen estimates that the traffic generated by spammers increases the costs you and I pay for Internet service by a full $2 per month. That’s just the infrastructure costs our ISPs have to bear and pass on to us. And of course it’s a huge waste of time.

Name-brand memory vs. generic

The difference between brand-name RAM and commodity RAM. I’ve been seeing a lot of questions along the lines of, “Do I really have to buy name-brand memory when I can find memory for half the price on PriceWatch?” on message boards lately. I talked about memory some in Memory-buying secrets, but I didn’t really go into the difference between generic/commodity/broker memory and the expensive stuff.

There are three factors that go into the quality of a memory module: the quality of the chips, the quality of the printed circuit board (PCB), and manufacturing.

When memory chips are made, they are tested. A memory chip that runs at or below spec gets classified as an A-grade chip. Chips with minor defects are classified as C-grade chips and shouldn’t be used in PCs.

Memory manufacturers will also charge varying amounts based on how much testing they do for the chips. Top-tier chips are guaranteed to have a failure rate of .1% or less–we’re talking one in a thousand chips failing here.

The least expensive chips aren’t tested at all.

The only way to ensure you’re getting these best-of-the-best chips is to buy name-brand memory. The best way to ensure you’re getting C-grade or untested chips is to buy the cheapest module you can find.

Inexpensive PCBs use a four-layer design, with signal layers on the outside, and power and ground sandwiched inside. This is inexpensive to produce and easy to repair. Unfortunately, this design leaves you open to signal noise, which can corrupt the data stored in the chip, and lead to unpleasant things like BSODs.

A better approach is to put the signal layers inside, and put power and ground on the outside, protecting them. Better still is a six-layer design, which adds two more ground layers for even better isolation. The higher the memory speed, the more important this extra isolation becomes.

You can sometimes tell the difference between a 4-layer and a 6-layer board by looking at it under a strong light. By turning it slowly, you can isolate discrete layers with the naked eye. However, it’s impossible to tell the difference between power, ground, and signal layers with the naked eye.

Name-brand vendors use high-quality PCBs. Some even proudly proclaim it when they use 6-layer boards. Again, the best way to get a poorly designed 4-layer board is to buy the cheapest memory you can find.

The third factor is assembly. When soldering chips to PCBs, things can go wrong. Trust me. I’m very good at demonstrating. While no one puts together memory modules by hand (I hope), my plumber-like soldering skills make me appreciate good equipment. Quality solder joints are bright and shiny, and they’re applied very quickly. Intermittent solder joints cause problems, and they’re maddening.

Kingston puts every module through rigorous testing. Other name-brand manufacturers test as well. When you don’t even know who made your module, it’s impossible to know whether it’s had the proper testing. Putting it in a PC and watching the BIOS check is not proper testing. Memory has to be very far gone to fail that test.

Miscellaneous computer memory.
The module on the right has Samsung chips on it, but that doesn’t necessarily mean it was made by Samsung. I’d be fine with putting the HP module directly below it in a PC, though. The same goes for the Kingston module at the top of the image.

Frequently vendors will advertise Micron memory or NEC memory. A Micron or NEC stamp on the chips doesn’t necessarily mean they manufactured the module! Be sure to find out who assembled the module–they usually stamp the PCB, or they put a sticker on the module itself. If you find a so-called Micron module on Pricewatch for half the price that Crucial is asking (Crucial is a division of Micron), chances are it’s a no-name module that just happens to have Micron chips on it.

So, does it really make a difference?

As an IT professional by day, I work on a large network–roughly 700 PCs. I’ll conservatively estimate that farm of PCs has 1,000 memory modules collectively. We buy name-brand memory (Crucial, Kingston, Viking or Simple) exclusively. We buy PCs from Micron, so they have Crucial modules in them. Macintoshes generally come with Crucial or IBM modules.

A typical memory module has 8 chips on it, and it only takes one bad chip to make the whole module bad. I’ve seen two bad memory modules cross my desk–so we’re talking two bad chips out of a batch of 8,000. So if you’re buying a single module for a home PC, your chances of getting a lemon are pretty slim if you get a good name brand.

For my own use, I buy name-brand memory modules. Usually this means Crucial; I got a great deal over the summer on some PNY memory so I bought a couple of sticks. I use high-quality memory, I don’t overclock, and I generally don’t run experimental software. I almost never get a BSOD or an illegal operation error, even when running Windows 95 or 98. I reboot my Windows 9x boxes about once a month on average, and half the time that’s because I installed or uninstalled something and it requires me to reboot. To give you an idea how I use PCs, at the moment I have seven applications running, with 11 windows open between them, and two TSRs running. That’s my idea of moderate use.

I’d rather have 64 megs of Crucial memory than a gig of the cheap stuff. Hopefully now you see why.

Sound card and hard drive troubleshooting

Sound card woes. Gatermann recently ran into some problems with sound cards forcing his Internet connection to drop. It had literally been six years since I’ve seen a problem like that before, but he kept running into it. Finally, it dawned on me: Try changing slots to force it to use a different interrupt. Therein was the silver bullet. The problem didn’t go away completely, but the culprit arose: the Sound Blaster 16 emulation. So I had him go into Device Manager and put the SB16 emulation on a different interrupt, and the problem went away.
It’s been forever since I’ve seen an honest-to-goodness interrupt conflict. This particular PC has every expansion slot filled with something or other, which is why he ran up against it. Keep that in mind: Just because we have PCI and plug and play these days, doesn’t mean you won’t ever see an interrupt conflict. On a well-expanded system, this ancient problem can occasionally rear its ugly head (while Microchannel required their cards to be capable of interrupt sharing; PCI only *recommends* it–so not every PCI device can share an interrupt, particularly if an ISA device has grabbed it. Alas, Microchannel fell victim to IBM’s greedy overly restrictive licensing terms and raw-dead-fish marketing, so as a result we have cheap PCs today but more headaches than we necessarily need. Speaking of raw-dead-fish marketing, I could mention that the Amiga’s Zorro bus had true plug and play and hundreds of interrupts from Day One in 1985, but nobody wants to hear that. Oops, I said it anyway.)

This problem used to happen all the time when people would put their modems on COM4 and a serial mouse on COM2 (or COM1 and 3). Since those ports by default shared interrupts with one another, you got goofy symptoms like your Internet connection dropping whenever you moved the mouse. People don’t configure their COM ports that way anymore, which is what’s made that problem so rare.

I think I finally got that G4 deployed. Wednesday it decided it didn’t want to shut down, and I had to reinstall the OS to fix it. Then on Thursday, it decided it didn’t want to recognize the mouse button anymore. I still don’t know what exactly I did to fix that–I booted off a spare MacOS 9 partition, ran a battery of disk repair tools and a defragmenter, and the problem went away. So while Mac users can snicker about interrupt problems, their machines aren’t exactly immune to weird problems either.

——-

From: “Gialluca, Tony”

question: RE Optimizing Windows and Temp files

Hi Mr. Farquhar,

In you book on page 112 you discuss placing temp files on a ramdisk. On this page you show an example where:

Set temp=ram disk letter:\temp Set tmp=ram disk letter:\temp

Shouldn’t you also include changing

[HKEY_LOCAL_MACHINE\Software\Microsoft\Windows\CurrentVersion\explorer\Volum eCaches\Temporary files\folder] to “ram disk letter:\temp” also ??

Per the description

([HKEY_LOCAL_MACHINE\Software\Microsoft\Windows\CurrentVersion\explorer\Volu meCaches\Temporary files\description]) says: “Programs sometimes store temporary information in a TEMP folder. Before a program closes, it usually deletes this information.\r\n\r\nYou can safely delete temporary files that have not been modified in over a week.” The only potential pitfall that I can think of is if windows or programs (say during installations) need this area to remain persistant through reboots, even though the files may be of
a temporary nature…

Your thoughts would be appreciated …

Respectfully,

Tony

———-

To be perfectly honest, I didn’t know that registry key existed (nor did the book’s technical reviewers, evidently). That registry key, too, should be changed, yes. Thanks!

You are correct that if a program does a hard reboot (rather than just exiting to real mode and reloading Windows), you’ll lose the contents of the ramdisk and thus the temp folder. Fortunately, most programs seem to use the temp directory the way they’re supposed to–for temporary, fleeting things. Now if they’d just learn to clean up after themselves…

Of course, this also applies to my advice on creating a temp partition, on page 62.

Thanks much; this is very good information.

———-

From: “Gary M. Berg”

Subject: Maxtor hard drives

Since you’ve been talking about WD and Maxtor hard drives…

I heard rumors just after Win2K SP1 came out that the service pack had problems with machines with Maxtor hard drives. I’ve not been able to find much of anything else on this. What have you heard?

———-

That’s a new one to me. Maybe another reader has heard something, but it sure seems odd. I can’t imagine Microsoft didn’t test SP1 on the major drive manufactuers’ drives (Fujitsu, IBM, Maxtor, Quantum, Samsung, Seagate, and Western Digital), and with Maxtor being one of the Big Two in retail….

Once I get my current big project off my back this weekend, I’m half-tempted to try it just to see. Unless someone already has…

Western Digital hard drives

Apparently it’s possible right now to get WD hard drives dirt cheap at certain warehouse clubs in St. Louis. How cheap? One person wrote in and told me $30 after rebates for a 10-gig drive. He asked me what I thought of the deal. It’s a great price, sure. My problem is, if I bought one, I’d be tempted to actually use it.
I’m very down on Western Digital. At my previous employer, we had about 600 PCs, with a variety of drives: a small number of Seagates, and roughly equal representation of IBM, Maxtor, Western Digital, and Quantum. We had maybe a drive a month go bad on us (ours was an aging fleet). I saw about as many Western Digitals go bad as all the rest–combined. I’ll buy an IBM, Maxtor, or Quantum drive without flinching, but I stay away from WD.

At my current employer, we have fewer problems (newer equipment), but I still see about as many WDs go down as anything else. Here we have mostly WD, Samsung, IBM, and Seagate drives, since that’s what Micron tends to use. Again, I see about as many WDs go as all the others. The last WD to go out happened when I took a half-dozen PCs to a convention in New Orleans. It was the middle of registration, with tired travelers all around, and the machine kept locking up. Finally, one time the drive just didn’t come back. I located a computer store, paid an outrageous price for a drive (unfortunately, another WD because it was all they had), and managed to get the drive in with only a couple hours’ downtime. But after failing me when I most needed dependability, I vowed to never buy another WD. Whenever I spec a drive for work, I get a Maxtor. I find them more reliable, faster, and they’re just as easy to find as WDs. And the CompUSA down the street always has a good deal on them.