dean's athlon system

the continuation of a long, tiresome, and ultimately pointless diatribe about my AMD athlon system

 
back to dean's computers
athlon system "bubbler"

Homebuilt Athlon XP 1900+
My Athlon 1.2G system, FrontAthlong 1.2G Side (not much stuffed in the box right now)AMD came out with the Athlon, a sweet CPU, and I was stumbling along with my K6-2/300, which had grown to a 350Mhz system overclocked to ~400Mhz. But it was running out of steam. See now, my original plan when building the K6-2 system was to simply pop in a faster CPU when they became available. While AMD did release the K6-III at up to 400Mhz, and the K6-2 processors up to 550Mhz, neither was a big enough jump in speed for the cost to interest me (a K6-III/400 was ~$140) . In that light, the AMD K6-x solutions seemed to be at a dead-end. I eventually found a deal on a K6-III+ 450MHz, and upgraded to that
(overclocked to 550MHz), but that was the end of the line for the K6.

Of course, AMD wasn't just giving up at the high-end, they positioned the K6-x line as their "value processors," and released their "7th generation" CPU, the Athlon. They did it right, the Athlon is a very nice CPU, and I wanted one. The problem is, you can't just remove the K6-2 chip and plug in an Athlon, they use entirely different plug and electrical configurations, like trying to plug a CD into a cassette slot. (The K6-2 uses the "Socket 7" architecture, and the Athlon uses "Slot A". Socket 7 is a square grid while Slot A is a long thin connector).

So if I wanted to use the Athlon, I couldn't just pop in a new CPU, I'd need to buy a new motherboard as well.

And that's what I did. I wasn't too keen on spending the extra money, but technology was moving on, and I didn't want to be left standing at the station.

I purchased an Abit KA7 motherboard and an Athlon 700 "Classic" that overclocked easily to 820+ MHz. That system did me well for a year, and would still, but, one night, I had the notion to tune it up a bit...

Well, let me cut to the chase, if you fail to connect the CPU fan to a suitable connector, the CPU first gets warm. Then it gets hot. Then really really hot. And then, for seemingly no reason at all, it becomes stone cold.

There is a reason, of course. It's because in it's "really really hot" phase it reverted to it's native silicon in its overheated fury (i.e., It "Burned Up"). Entropy, of sorts.

I needed a new CPU.

AMD had come out with an upgrade to the "Classic" with their "Thunderbird" CPU. (Thunderbird is the code-name for the new processor. AMD officially calls it the"AMD Athlon™ processor featuring performance-enhancing on-chip L2 cache memory". Uh huh. I'll just go with "T-Bird", thanks). Since it's much faster (more CPU cache, a faster basic CPU design, higher clock speeds, etc.) for the same price, it just made sense to replace my old smoldering CPU with this new & better chip. Well, once again, there was a catch. The T-Bird required (wait for it) ... yet another new motherboard design.

Yep, just not the upgrade I'd planned.

The Athlon Classic is largely a dead-end, the "Slot A" architecture was replaced by the simpler and less-expensive "Socket A". So, bullet clenched, I researched the current motherboard market for a good socket motherboard. There were a number of really good candidates out, and I decided on an Iwill KK266. I coupled it with a new Athlon T-Bird 1.2Ghz/266 (FSB), and since have put in the XP 1900+ chip. This setup is wicked fast, even compared to the Athlon Classic 700 it replaced. Oh, in hopes to defer another CPU flameout, I clamped on a Thermaltake Volcano 7 CPU cooler, using some of the trendy Arctic Silver thermal paste.

The system name, "bubbler"
I name my system's on a "budda" theme, for no solid reason outside of my interest in Asian lore. alexa's system was called "baby-budda" because she played games on it when she was a baby. My server is named "sitting-budda" because I had it sitting in the living room at one point. The non-work laptops also have variants on that scheme. So calling this system "bubbler" deviates from my grand plan, but ... so what? Now, where does "bubbler" come from? I'm from Wisconsin, and water fountains (drinking fountains) are called "bubblers". So naming my main system bubbler feeds my blatant self-interest in my Wisconsin heritage. But then again, maybe someday I'll switch over to water-cooling and give merit to the name...

Update, July 2001
I burned up another CPU.

It wasn't my fault this time. Really. In fact, it was because I reseated the heatsink on the T-bird correctly, that it subsequently smoked. No explanation, and it only underscores that AMD CPUs, while speedy, get there by running hot hot hot.

So, I'm replacing the CPU with another T-bird 1.2/266, but while that's on the way, I decided to try an Athlon Duron processor as a temporary replacement. The Duron is the same CPU, largely, as the T-Bird, with less cache memory (but not a lot), and lower speeds. But it's cheap. Dang cheap. I spent $34 for a 750MHz Duron that I've mildly overclocked to 830MHz. Crazy. The Duron's performance is less than the T-birds, but not noticably in day-to-day work. Stability, however, has been another issue.

I think it's because I've had two motherboards, three CPU's, and any number of add-in cards on one install of Windows 98 that the system started coughing up errors, crashing, and hanging. I'd boot and it'd complain about a program that somehow exceeded it's bounds. Or I'd get hangs, even black screens. I didn't think the hardware was the problem, heaven knows I tweaked and tinkered with it enough. I figured the OS was getting cranky after so many changes. And, CPU's that are going up in smoke are known to do rude things to disks and the data and OS files on them.

So I did the 2nd most logical thing, I upgraded to Windows 2000. Now, the first most logical thing is to format, reinstall a clean Win98 system, then reinstall all of your apps. But I've got a LOT of apps, and, simply, I didn't want to do that. It's not just installing the things, it's configuring them to how you want, registering them, and applying patches and updates. It's a pain.

So, bolstered by a couple Tsingtao's, I popped in my Win2K Pro CD, and let it go. After a couple of hours, and a few more hours bringing in new drivers, updating configs (you can never get away from that), the system's been entirely stable. And, the Duron's fast, despite it's "meager" sub-gig speed.

So, Win2K seems good for now (I've run it since it first came out on other systems, and I've always liked it, but I had planned to keep one Win98 system at home). And the Duron is dandy. I'll put the T-bird in when it arrives, but things are going good as they stand right now.

Update, October 2001
The replacement T-Bird 1.2/266 came back in July, I swapped it in, and all was well. I eventually built another system for Alexa from the Duron, more than enough for her.. AMD went and upgraded their processors, now to the "Athlon XP," with ratings going to 2100+, although these are proving to be conservative, with the lower-clockspeed XP beating Intel's higher-clockspeed Pentium 4's. These clock-wars are getting old, fast. Maybe it's just time to get rid of clocks altogether.


After watching [the system] painfully crawl up to the point where I could finally log in, I started probing around to see what was the matter. It only took a few minutes to realize -- my D: drive was... gone.

So a 750MHz Duron's starting to look old now, and even my treasured T-Bird is getting wimpy. But I'll stick to my guns that the system's still far faster than I need, and I think the next upgrade will be to a dual-XP system, when more and better SMP mobo's are released, running DDR memory, with a nVidia "GeForce 4 XP" video card, yadda yadda yadda....

And, with Windows XP out, I considered "upgrading." However, I was a beta-tester, and while some features of XP were nifty, overall the improvements from Win2K aren't "inspiring." But mostly I'm opposed to the heavy-handedness I see from Microsoft, with their product activiation schemes and relentless push to have you sign up with MS Passport, etc. Granted a guy who writes this much about his computers and family, and posts it on the net isn't the most concerned about privacy, but I don't like to have it taken away from me. I'll hold off on XP, maybe see where Lindows is going (vaporware, to date).

The road to RAID -- January 2002
Well, just before I took off to Oregon to visit family for Christmas, I was (ohwaddasooprise) tinkering around on my system one night. It started acting up, hanging for a bit, apps flaking out, so on. So, this is Windows, and even though it's Win2K, what's the first thing you do?

Reboot, natch'.

Twenty minutes later the system finally booted back up. Twenty minutes, much of the time seemingly doing not a thing, and I was scratching my head the whole while. After watching it painfully crawl up to the point where I could finally log in, I started probing around to see what was the matter. It only took a few minutes to realize -- my D: drive was... gone.

I partition my systems into a C: drive, with the OS and program files on it. The D: drive is especially important, it holds essentially all of my data, pictures, movie clips, mp3s, email, websites (this on included), documents.... you get the idea. About an hour later I'd downloaded and tested my D: drive with IBM's Drive Fitness Test (DFT). The DFT gave me the verdict: "The drive is corrupted, it must be formatted to continue." Yikes! Not one to take catastrophe at first pass, I ran DFT a couple more times, only to get the same message. So I clicked "Yes," and away it went.

The format and subsequent verification took, oh, about 1/2 hour, and in the end pronounced the drive broken, with an error code I'd need to get this still-under-warranty drive RMA'd by IBM. I jotted it down, cursed, and surfed again to IBM to get an RMA via their web forms. I got it, but not without a fight -- part of their web-processing asks if you ran DFT, and if so, what was the error code? I put it in and the next page proclaimed this an "Invalid Code" and dumped me out. I eventually snuck around this by saying I didn't run DFT, and it let me get my return number.

Also, since I was leaving the next day, and it was getting mighty late, I clicked over to newegg.com (btw, you may be wondering how I was surfing on my broken system. I wasn't, I was using my K6 server by this time) and ordered two more disks, again IBM, one a 40G and one a 60G. My idea was that I'd replace the 40G straight-away, and stuff the 60G into the server so, in the future, between CD backups I'd simply copy the Athlon's data over to the larger 60G drive on the server "just in case". And, I had another plan kinda-in mind that I'll talk about in a bit. (Oh, despite the failure of this IBM drive, it's the first such failure I've ever had with an IBM, so I decided to stay loyal).

The drives arrived the day after we came back from Oregon, and I did just what I planned, popping the 40G into the Athlon, and the 60G into "Sitting-Budda", my server (the Athlon's name, btw, is "Bubbler"). Then, over the next week, I endured the pain of scraping together the latest bits & pieces of my data from old CD's, and from my IBM laptop (which, thankfully, had my most current email, although I lost a lot of my older archives. Dang). Obviously I could get my websites back by pulling them from their hosts, and I that's just what I did.

Now, I've been in the storage (high-end, arrays) for a long time professionally. As you may or may not know, disk arrays, "RAID" (Redundant Array of Independent Disks) is a technology where groups of small disks are arranged to work together and look to the host as one larger disk. RAID offers many advantages and, depending on configuration, they may work faster, become more reliable, or both. The arrays we sell are spendy, and, again depending on configuration, it would be cheaper to buy a new Lexus LS 430, or even a couple Ferrari 550 Maranello's. Or a house in the Boise foothills, with a view. I was feeling bad about losing my data, but not that bad.


But the thing is, RAID 0 isn't redundant. In fact, if you stripe two disks in RAID 0 you double the risk of data loss because a failure in either drive results in complete data loss.

Fortunately, there are more reasonable alternatives for RAID controllers, many of which cost less than the cupholder in my Volvo ($125. It's plastic. But it's genuine Volvo Plastic).

So, during my tear-filled week sorting through CD's and lost data, I poked around to see reviews of PC-based RAID controllers. What I was looking for was simple: a low-cost controller for Bubbler (remember? the Althon system's name) that would let me stuff two disks working as a RAID 1 array. If you didn't follow the link (or even if you did and for some reason thought it was a good idea to come back here), simply put, RAID 1 is known as "mirroring." Mirroring is a technique where, in a simple example with two disks, the array controller writes the same data to each disk all the time. The disks are then identical copies. Write that letter to mom and save it? It's on both disks. Download an mp3 of "Sinnerman"? One copy is on each disk. Invent a cure for cancer and write it down? Saved twice, one on each disk.

Now, astute readers will realize this provides, essentially, a continuous backup. Both drives get the same data, all the time. And, should one drive fail, the other will continue to spin merrily along, with all of your data! (Actually, there's no guarantee that both drives won't fail together -- if there's a house fire, for example, and the entire PC gets bbq'd. Or if some wacky government decides to try an EMP experiment over your house. Or just Bad Luck, and both disks simply fail at the same time. It can happen, although, with MTBF's of 500,000 hours or more, it'd have to be pretty-darn-bad luck to have them both fail at the same time. I've seen it happen, but not, of course, to me! If this is what happens to you then you probably have bragging rights to a couple of lightning strikes, or the little old lady you let skip ahead of you in the supermarket line bought your winning lottery ticket. If your luck's like that, redundancy's the least of your problems. Build a bomb-shelter, and stay in it).

Where was I going with this...?

Oh yeah, so RAID 1, which is the simplest redundant RAID level, provides protection by duplicating data to 2 or more mechanisms (you can have a bunch of 'em, but 2 provides pretty good protection). Now, having duplicate data isn't much good if you can't get to it, but even, ah, inexpensive PCI/IDE RAID controllers let your system continue happily along by taking the failed drive offline, and using the good one. Seamless. And these el-cheapo controllers generally ship with monitoring software that tells you (pop up on your screen, or email, SMTP, so on) when one mech fails. This is good because once you lose one disk, you're vulnerable to complete data loss should the second disk fail at some point. So you want to replace a failed mech as soon as possible, tell the RAID controller to copy all the Good Data to it, then use it in RAID 1. Safe again.

And even more-astuter readers will be thinking, "Yeah, but if everything's duplicated, don't you lose a lot of disk space?" Yep, in the case of two disk RAID 1, you get only half the total space. Two 40G drives together total 80G. But put 'em in RAID 1? 40G total. Such is the penalty for redundancy (although other RAID levels, such as RAID 5 spread the redundancy among disks in a different way, and a simple RAID 5 system with 5 disks effectively loses only 1 disk to redundancy (80% efficient). But I don't need 5 disks, or even three, and RAID 5 controllers cost more. Today you can pick up a 60G 7200 RPM disk for $115, so the cost of "wasting" one full disk to redundancy isn't very high.

I went with the IWill SIDE-100 RAID card, capable of RAID 0, RAID 1, and RAID 0+1 with up to four IDE "Ultra ATA/100" disks. FWIW, RAID 0 seems to be what most "PC enthusiasts" use because you can get better performance and don't lose any space -- the disks are "striped" together and two 40G disks give you a fast 80G drive. Even the manufacturer of the RAID chip (HighPoint) only uses RAID 0 when showing wonderous performance comparison graphs to convince you their chip's got it goin'. What's not to like about RAID 0? Why didn't I go with it then? Remember, my goal was data protection: I didn't like losing years of data nor the hassle of piecing it back together. But the thing is, RAID 0 isn't redundant. In fact, if you stripe two disks in RAID 0 you double the risk of data loss because a failure in either drive results in complete data loss. Bummer. RAID 0+1 solves this by striping & mirroring the data, giving you fast access with redundancy. But, everything's a tradeoff, and RAID 0+1 requires a minimum of four disks to work. I don't want to spend the money on what, to me, is a minor speed boost. Besides, I don't have enough disk slots in my case for all those drives. The IWill provides an additional safety feature, one that I didn't use. You can configure a RAID 1 array with two drives, but hang a third drive on the controller that, well, essentially does nothing most of the time. The third drive is a "hotspare," a spare disk that the controller will automatically bring on line should one of the two main drives fail. That way, the array automatically rebuilds redundancy and doesn't leave you with the possibility of having the other drive fail until you can swap in a new drive for the failed one. That's a good idea, but I don't want to spend the additional bucks, and I feel pretty safe with just the two. See? RAID has all kinds of safety hooks!

I changed my plan, slightly, from the original. Idecided I wanted more space, so with the new 40G & 60G drives, I bought yet-another drive, another 60G. I coupled the SIDE-100 RAID card with the two 60G drives, mirrored them, and copied all my data from the 40G to the new array. I put the 40G into the server (a couple times -- it matters which way the IDE cable plugs into the motherboard) and was done with it.

So now, I'm redundant and happy about it. The software that comes with the IWill controller seems fine. Perhaps a bit unpolished, but it provides notification upon failure of a mech. I tried to set it up to send me email, but doing so and pressing "Test" pops up an error box that it didn't work, and I've never gotten any email. There doesn't seem to be anyway to troubleshoot it, but I'll keep tinkering.

Last, I ran SISandra on the array to see what kind of performance I'd get. Technically, a RAID 1 array should give better read performance because the controller only needs to read from one of the disks, and if it's smart, it picks the one that's closest to the data (or, in practice, tells them both to get the same data and simply takes it from the disk that gets there first). IWill's documentation claims "Improved performance over single disks due to advanced and proprietary seek algorithms", or some such, so I was enthused to see how it panned out.

The Sandra RAID 1 score was just over 23K. On par with a standalong disk, perhaps a bit slower (I admit I didn't take the time to test the new 60G IBM standalone, but my old 40G once measured over 28K in Sandra). I believe the lack of performance increase is in part due to the fact that the RAID functions for this card are actually done in software, in the card driver. So there's overhead. "Real" RAID products are hardware-based, and you can buy them for PC's for a few hundred dollars. I didn't care about that, and the IWill lists for around $75, but I got it for $50. Dedicated hw working on your RAID is faster than pushing the job off onto the OS. But the IWill serves my purposes.

Good enough.

update, summer 2002

well, it seems not "good enough" is how it turned out to be. I dumped the RAID card, and moved to one Western Digital 120GB disk. See the current system's specs here. why? Well, another IBM drive failed, and I found the RAID card was just user un-friendly in managing actual failures. several times the RAID card reported "a" disk failure, but nothing I could find in the software or BIOS for the card told me which drive had the problem. I had to flip a coin, and after a while I decided to: 1) Get rid of all the IBM drives, and 2) Get rid of the RAID card until I could get one that not only reported problems, but gave you some idea how to fix them.

Also, in the interim I changed motherboards, and the new one came with a built-in RAID, but I wound up using that just as an extra EIDE controller, and after THAT motherboard blew up (I believe I didn't fully seat a memory chip and fried a regulator supplying memory) I moved to YAMB ("Yet Another MotherBoard"), an ASUS K7N8X that didn't have RAID built-in. Oh, and between that I simply gave the RAID controller to a friend.

I'll move back to RAID at some point, but I just wasn't happy with the one I had. So I'll wait. In the meantime I back up a lot more.

And the information about my upgrades above is now old, I continually move to the next level as I see fit (I never buy the "latest" technology, always one or two generations back. I used to, but the cost is always too high, and it doesn't last. Currently if you buy two generations back on AMD CPUs you'll pay a third of the cost of the cutting-edge, and the cutting-edge is perhaps 20% faster. Doesn't make much sense, but it took me a lot of years to realize that!) So, for my current specs, go here. It's probably where you came from anyway...

(Back to Top)