Roll yer own PC Part Deux

Building my own PC wasn't punishing enough the last time, so I decided to build another for myself!  Actually, what I did amounted to a massive retrofit, keeping little more than the case.  Here is the tale of time, sweat, frustration and misery...

Since the original build of the BEAST a few things changed of course.  The SE-440BX2 motherboard was replaced with a BX2V, to support an 800MHz P-III CPU.  128Mb of SDRAM gave way to 768Mb of Crucial (Micron) RAM on the maximum three SIMMs.  The Mag monitor was food-chained and replaced with a 19" Viewsonic PF795, which is about ready for a warranty visit to correct some ghosting (requires shipping to Atlanta, Georgia, egads).  I picked up a Microsoft optical mouse, which is really slicker 'n snot, nevermore a worry about dirty mouse balls, Pepsi Syndrome, etc.  Its scroll wheel is a smidge easier to use but still not perfect.  And of course I had to toss Win98 and install Windows 2000 Professional.  The Maynard 2000 DAT was replaced by a Seagate Scorpion 20/40GB DAT courtesy of Dell and eBay.  The Yamaha CRW6416S was still there as was the CD changer, 3Com 3C905B, the DV300 Firewire card and a flaky Motorola voice modem.  The system still had the original IBM 9ZX 9Gb system drive and I had added a Quantum Atlas 10K 36Gb data drive.  I also added a SanDisk USB CompactFlash reader.

The system was quite reliable for some while, at least after I learned that Win2K had to be installed with the ACPI feature turned off to keep the Pinnacle card happy. Then life was good until a few months ago when one of the Crucial memory chips began to fail. Although I eventually diagnosed and replaced the faulty module under warranty, the repeated crashes and resulting registry damage left my system a little quirky. Faced with having to re-install Windows 2000 and the sixty or so applications and utilities that I can't live without, I decided it was time to make it worth the effort and do some upgrading.

I had to decide what I would keep.  The motherboard, typical of Intel, supplied little more than sound and a couple of USB ports and delivered merely average performance. The CD changer was one of those things that was a great idea but mediocre in execution. Advertised as 16X read speed, it was lucky to achieve 4X, and every time you started Windows Explorer it would take a half a minute for the system to scan all five CD slots on the changer. The 2740UW SCSI adapter was poorly chosen and became a limiting factor on system performance as well, forcing the LVD SCSI drives in the system into the slower SE (single-ended) mode. And the hard drives at the time were half-height rather than the usual third-height, so it was difficult to mount them in a mid-tower system that had two CD drives and a tape drive.

Almost everything was replaced. The new BEAST is built on a Gigabyte GA-8IEXP "Titan" motherboard, an Intel P4 2.53GHz CPU, a pair of 512Mb Geil "Ultra" DDR333 chips, an Adaptec 2740U2W SCSI controller, a Seagate Cheetah X15 (15K RPM) 36Gb SCSI drive with 8Mb cache, a Western Digital 120Gb 7200 RPM 8Mb cache IDE drive, and an ATI All in Wonder Radeon 8500 128Mb DDR. I replaced the old ATX power supply with a new P4/Athlon compatible power supply from Jinco, and I eliminated the CD changer, the modem and obviated the 3Com NIC.

So here we go with the ugly details...


I was very happy with the existing case - a quality mid-tower case supplied by Jinco, unique at the time not just for its tool-less design and slide-out motherboard carrier, but also for its IR window and a nice air filter in front. It's a relatively utilitarian looking case and that suited me just fine - I didn't want anyone confusing it with an iMac.


The GA-8IEXP is highly regarded and is absolutely the most full-featured motherboard of its kind, and received positively glowing reviews from the folks at Toms Hardware. Along with an ATA133 IDE controller and Promise IDE RAID support giving you four IDE ports supporting up to eight drives, the board sports 4X (1.5v only!) AGP, three Firewire (er, IEEE-1394) ports, six USB 2.0 ports, SP-DIF outputs (optical and RCA), an Intel VE-based 100mpbs Ethernet port AND six PCI slots as well! The only problem with all this gadgetry is that with all the supplied connectors installed on the rear apron, you'd be deprived of the use of three of those PCI slots. The board also sports two standard USB connectors, a game port, the usual legacy ports, and RealTek 6-channel sound. The board has three DDR memory slots and supports a maximum of 2Gb RAM. Unfortunately the current maximum is 1Gb because you can really only use two slots with the big-capacity (dual-bank or double-sided) DDR modules and the 1024Mb modules aren't available yet.

The choice of a DDR266 -based board was in retrospect somewhat questionable, because of the very direct correlation between memory price and bandwidth. Memory for a PC1066 RDRAM system would cost 50% more but would also yield almost 50% more memory bandwidth.


The 2.53GHz P4 had just dropped substantially in price, what else can I say.  Made for a 533MHz FSB and the odd number means nobody will mistake it for 400MHz.  Aesthetics are important in the strangest ways.


I purchased two Geil "Ultra" 512Mb DDR333 (PC2700) chips, chosen for their tight advertised memory timings (5-2-2 1T). Even though the motherboard only supports DDR266, these chips should leave me plenty of elbow room for overclocking.


I found an Adaptec 2740U2W in my parts bin.  Thou still technically not optimal for the Ultra160 drives in the system, to truly exploit the throughput of an Ultra 160 system you would need an Adaptec 29160 card or equivalent and a 64-bit PCI slot to put it in. No well-appointed desktop or workstation motherboards have such slots. But if you work out the numbers, the burst speeds of the current crop of SCSI drives still don't approach the 80mb/s speed rating of Ultra2, so I can't imagine I'll see a noticeable performance penalty unless I'm hanging more than three or more drives off the 2740U2W.

The other nice thing about the U2W is that it can support the slow single-ended SCSI stuff like the scanner and the faster LVD devices without everything on the card being dragged down to SE speeds.


Mixing a 15K RPM SCSI drive and a 7200 RPM IDE drive may seem odd, but no amount of money can buy a 120Gb 15K RPM SCSI drive, and $150 for a 120Gb drive whose real-life performance benchmarks approach those of the 10K RPM SCSI drives I just removed, is a tough argument to beat. For ultimate system performance the SCSI drive is the system drive and all the crap I download lands on the IDE drive. Because of this scenario, the choice of a motherboard with extra IDE ports was particularly useful. The IDE hangs off one of the Promise ports and provides a tidy work-around for that annoying Wintel issue of the IDE drive having to be the boot device in a mixed-bus configuration.

All the hard drives were chosen not just for their speed but for quietness and relatively cool operating temperatures.  The WD1200JB just happened to be an outstanding value.  I like to read The Storage Review for great technical data on hard drives, etc. 

The Yamaha CRW6416S was replaced by the very slick but somewhat ugly Sony DRU-500A multi-format DVDRW/CDRW.  This is a very fast drive, and allegedly picky about media but I have not had problems so far.  It did take Sony a few iterations of firmware to get things straightened out however.

I swiped an LG DVD-ROM from my wife to fill out the drive bays.  The DAT drive, pulled from a Dell 2450, has a black faceplate yet my case is standard beige.  This provides a small distraction from the Sony DVDRW's silly silver tray facade.


For video cards I chose the ATI All in Wonder 8500 instead of the newer ATI 9700 series cards because I'm cheap and it saved me $150. The 8500 GPU is plenty fast and 128Mb of RAM is more than Microsoft Flight Simulator is ever likely to use. The card gives you a stereo TV tuner with closed captioning and offers a really cool "blended" TV display mode where you can have the TV full-screen and your applications run over it, not quite opaque. The ATI card comes with every kind of cable you could possibly need and includes a (X10-based) RF remote control that communicates with the PC via a USB dongle. The ATI, like the 3Dfx, comes with a dongle on a longish cable for access to some of the A/V I/O. But where the 3Dfx supplied both input and output on its desktop dongle for audio, video and s-video, the ATI's dongle only supplies input. For output you have to fish to the back of your PC for connections to the supplied six-inch splitter cable. For that effort you're rewarded with a bonus: a digital audio output (RCA). There's still an analog stereo audio output as well. One of those audio cables must be routed to your sound card's exterior auxiliary inputs if you want sound through your PC's speakers; no internal sound cable connection of any kind is available.

The only thing I really miss about the 3Dfx card is the stereo FM tuner. I definitely don't miss the poorly written beta Win2K drivers for the 3Dfx however, supplied just before the company went under. I do have to say that I'm a bit worried about ATI now. 3Dfx went down the tubes about three months after I bought their TV3500 card, and Hercules went chapter 11 about four months after I bought their Thriller-3D video card. I seem to be the Video Card Prince of Darkness. If ATI makes it to the spring of 2003 we can all breathe a sigh of relief.


My next step was to get rid of that SanDisk flash card reader. I had to wait a few weeks before the Atech Flash Pro-III became available in retail form. This is a really slick unit that fits cleanly into a floppy drive bay. This purchase may too turn out to be a compromise, since this unit is USB 1.1 and Atech Flash was rumored to have a USB 2.0 unit in the works. But as usual I could not wait, and at least this is only a $45 adventure.

Listening to radio scanners is another hobby of mine, so eBay helped me find an inexpensive Icom PCR-1000, which is a PC-controlled scanner.  It hooks up via a serial port and requires a separate speaker as well.  I knew I hung onto that Radio Shlock CB speaker all these years for a reason!  A side-project is to make a web-based front-end for the Icom and feed the audio to a streaming server, so I can have my own personal scanner wherever there's a web browser capable of streaming Windows Media sound.


The new P4-compatible 380W Jinco power supply was a good choice but not a perfect one. It's plenty quiet and reliable but it does not supply a tachometer lead to connect to the motherboard for power supply fan speed monitoring. If I had to replace the supply I'd probably pick an Antec unit the next time, also well known for quiet running and honest performance.

Putting It All Together

This year marks the era of Fancy Packaging. The Geil DDR is very pretty indeed, with shiny copper heat spreaders that have nice nickel Geil logos brazed on. They came in fancy Plexiglas cases that were nearly impossible to open. I managed to slide the cover off of one, after plenty of cursing, and the other case ended up cracked when my frustration gave way to brute force. The Gigabyte motherboard was delivered in an iridescent burnt orange colored box with a plastic carry handle, and the board itself is a nice shade of blue with a copper top on the Northbridge cooler and color-coded connectors throughout.

When I put the system together, airflow was a major concern. There's a lot of high-performance goodies in there to keep cool. While the do-it-yourself world seems to have gotten a fancy for pretty colored round IDE and SCSI cables for this reason, wise routing of standard flat cables is perfectly suitable too. It's very easy - just fold the cable on itself at 90 degrees toward the side of the case, then a neat 90-degree bend to run it along the surface of that side. A few more simple bends when you reach the desired IDE or SCSI connectors and you're done. The cables stay completely out of useful airflow patterns this way. This isn't anything that Compaq, HP and Dell, et al haven't been doing for years. I also like to twist the power supply cables and tie-wrap them to keep them neat, likewise for the other cables for fans, etc.


I haven't done that much motherboard tuning yet but once I found out about the poorly documented Ctrl-F1 to reach the advanced BIOS screens, I was able to fine-tune memory timing to better suit the Geil DDR modules. In the standard BIOS configuration pages there is an option called "Top Performance", which you can turn on or off. While the effect of this mode change can be measured both in benchmarks and in resulting CPU temperature, there is absolutely no documentation about what motherboard operating parameters are actually changed by this setting.

I stuck with Windows 2000 - XP is still too new for my taste, too many new quirks to discover just yet, the Professional version is rather expensive, and I absolutely don't care for WPA (Windows Product Activation).  Since the dusty old days of Lotus 1-2-3 I've emphatically promoted this very mportant idea:


I cannot trust software or a publisher that does not trust me.  These days of the DMCA and its myriad abuses are making me that much more of a believer.  It's time to fight back and the wallet is where it hurts 'em most.  Don't buy anything that uses copy protection or any of the intrusive, inconvenient, tyrannical implementations of DRM (Digital Rights Management).  If nobody buys into it, it WILL go away.

Setup was not without problems. I could not get the Pinnacle software to work properly with my DV camera until I removed the drivers for the motherboard's built-in IEEE-1394 ports, then disabled them in BIOS, then de-installed and re-installed the DV300 software. It took me DAYS to figure this out and Pinnacle was of little help.  I have not yet tried re-enabling the motherboard's ports and I was subsequently told by Someone Who Knows that Windows, any version, simply doesn't like mixing firewire buses.  Although I cannot verify that information in any knowledge bases or databases so far, it seems to be self-evident.

I originally hung the IDE hard drive off the Promise port, assuming it might offer better performance.  Later testing proved that the performance of the ATA133 Promise IDE controller was not measurably better than Intel's ATA100 IDE controller except perhaps during cached bursts, and I ended up moving the IDE drive onto the Intel port because either the Promise controller itself sucks, or the drivers suck, or both.  When the IDE drive would spin down due to inactivity, a subsequent spin-up would halt all processing on the box - even mouse movements - until the drive was responding again.  The Intel controller exhibited no such problems.  As usual for Giga-Byte, the company was completely unresponsive to my queries about this problem.

Another annoyance has more to do with the Microsoft and the software industry in general being completely out of touch. The All in Wonder supplies composite video input, to which I have attached a cheap and outdated but effective Panasonic EggCam. Unfortunately NetMeeting only supports devices that use the very old Video for Windows standard, and CuSeeMe 5.0 doesn't fare much better. The former doesn't even think a video capture device exists, and the latter thinks it sees the video capture device but can't seem to actually acquire video from it. ATI takes the moral high ground and won't support VFW, insisting video applications use only WDM and DirectShow.

Initially I could not get the Atech reader to work at all using the internal cable. Then when I tried the external cable it seemed like it was about to work but the driver installation hung the machine beyond the capabilities of Ctrl-Alt-Del. This is the first thing ever to have hung my PC, and I had to hit the reset button. On the next boot the external cable seemed to work OK but the internal cable was still no good. I whipped out a multimeter and the cable seemed OK, but then it struck me that the fit of the header connector onto the motherboard was a bit loose. I tightened up the contacts by pressing them down with a jeweler's screwdriver, plugged it in and everything worked perfectly!

The reader shows up as four drive letters - one for each flash memory slot - and as an additional nice touch, each letter has a unique icon in Windows Explorer. Like most drive letter accessible devices in Windows 2000, you can use the Disk Manager to re-assign the drive letters for each slot to more convenient or meaningful letters.

Now here's where things are less than perfect. The Giga-Byte motherboard supplies connectivity to its four rear apron -mounted USB 2.0 ports via a set of cables using two connectors to the motherboard - two USB ports to a motherboard connector. In order to plug in the Atech internally, I have to sacrifice not one but two of the rear apron USB ports.

The Atech also supplies a Firewire/IEEE-1394 extension facility on the front panel, and includes a cable to connect from a six-pin connector on the back of the Pro-III to header pins on a motherboard. Two problems present themselves here. First, the 1394 connector on the front of the Pro-III is the four-pin type, where usually you have six-pin connectors on the host devices. So get ready to pick up different cables to connect your peripherals. The other problem was slightly unique to my situation - I wanted to extend the internal 1394 port from my Pinnacle DV300. But that's not supplied on a header on the DV300 - it's ever so thoughtfully supplied on a six-pin 1394 connector! So yet another new cable will be required. Granted, if I wished to connect the Pro-III to the motherboard instead of the DV300 I would have no problem.

In use, the Atech reader is a bit weird.  CompactFlash works perfectly OK but several times I experienced system instability when writing to a Lexar 128MB SD card.  I did not test any other memory formats yet.  I think the drivers need some work.  In addition, there is no way to tell in Device Manager which slot is which device, so if you wish to alter the drive letters assigned to each slot, it can be a hit-or-miss proposition.

Need for Speed

In actual use, the new P4 system is lightning-quick! The boot-up and shut-down times are blazingly fast and application performance is absolutely remarkable, i.e. Adobe Photoshop 6.0 starts initially in about 5-1/2 seconds, and subsequent starts after closing it take just under 2 seconds thanks to Win2K's aggressive disk caching. Flight Simulator 2002 Professional still seems to take ages to load (20 seconds from program load to ready at the runway) what with all the terrain data and textures and so on, but with all the realism, 3D effects, display quality and detail settings set to their absolute maximum at 1280x1024x32, I'm getting frame rates of 20-35 fps while touring around the densely detailed New York City in a Beech Baron. This is quite acceptable!

The first benchmark was casually made by Pinnacle's DVExpert, which said that my C: drive could read and write up to 70 Mbyte/sec, with an average throughput of 45 Mbyte/sec. The IDE D: drive naturally didn't fare quite as well in the averaging figures - 37 Mbyte/sec, but the alarming number was the maximum throughput - 90 Mbyte/sec read and 87 Mbyte/sec write! This sounds awfully impressive but further inspection of DVExpert's test graphs revealed evidence that the WD's 8Mb buffer was largely responsible for those incredibly high peak numbers. Just the same, this speaks very well for IDE's potential versus far more costly SCSI devices.

The next benchmark was SiSoft Sandra 2002 Pro. This is where we really get down to business. Since we're getting serious now, I thought I should mention that the benchmarks are running in a production environment with about fifty applications installed, not on a squeaky-clean Window installation. For example, the system tray shows the motherboard monitoring utility, the sound chipset utility, Norton Anti-Virus, pcAnywhere, and the Palm Hotsync application. The machine is a member of an Active Directory domain, so a certain overhead goes with that as well.

In spite of the high-speed hard drives and a total of six fans running full-time (two on the power supply, one on the case, one on the Pentium, one on the video card AND one on the Northbridge), the system is actually substantially quieter than the PIII-800 it replaced.  Or that was, until the Northbridge fan's bearing began failing a few months later.  Since GigaByte didn't deem fit to reply to me about how to get a new fan under warranty, I tossed the fan in favor of an $8 Zalman heat sink.

The GA-8IEXP allows for plenty of over-clocking options including a very important "asynchronous" capability to change CPU and memory bus speeds without affecting the PCI and AGP bus speed. But the BEST thing about the Giga-Byte boards has to be "EasyTune 4". Using EasyTune, you can fiddle with most over-clock options (but not memory timing parameters) on the fly, instead of having to endure reboots between each change. I immediately upped the system bus to 142MHz resulting in a 284MHz memory clock and just shy of 2.7GHz processor speed, with perfect system stability and only a 1C rise in CPU temperature. Then I got a little greedier and set the over-clocking to "linear", which brought the AGP bus to from 66MHz to 71MHz and brought the PCI bus up to 35MHz. Then I thought things were working OK until I ran WinAMP and it stuttered terribly. Back to asynch and things straightened out just fine.

My greed for speed persists, so I'm still upping the system bus 1MHz at a time. The memory is supposed to be good for 333MHz so the CPU is the only thing I should need to worry about.


Entire contents Copyright (C) 1994-2015 Brad Berson and Bytebrothers Internet ServicesAnim Plug
Page updated February 12, 2009.  See Terms and Conditions of use!