June 2013 — This is the story of a relatively nice, intentionally low-drama PC suitable for running DOOM and Duke Nukem 3D and playing around with ISA sound cards in plain DOS and Linux. I spent some money to make it faster, quieter, and better looking than the salvaged beige box PCs that have previously served that purpose.
Last modified: 2023-11-19 10:32
Legend QDI PlatiniX 2DI-AL/C motherboard + SL6PP 2.6 GHz P4 CPU. Based on the listing title "LENOVE QDI P2DI-AL/C 478 ISA MOTHERBOARD INTEL 845 2.6G 512MEM," I thought I was going to get a 512 MiB DIMM as part of the deal, but no such luck.
With an ICH2 south bridge and a FSB that is limited to 400 MT/s, this Brookdale board is at the bottom of the pecking order for Pentium 4 chipsets, but it was a better value than either a top-shelf 440BX board or an expensive industrial board. Aside from the ISA slot and the absence of overclocking features, there is nothing especially "industrial" about this one.
2.6 GHz was the limit of the supported CPUs list in the archived remnants of QDI's dead-and-buried web site. (There is no separate page for the 2DI, which is the ISA slot variant of the 2D board.) Nevertheless, the 2.8 GHz CPU was selling cheap on EBay, so once it arrived on the slow boat from China, I tried it and it worked. The new CPU is SL7EY, 2.8 GHz, 512 KiB L2 cache, 400 MT/s FSB, TCASE 75° C.
The fastest available CPUs for 400 MT/s Socket 478 (SL6YH and SL74Q) were selling on EBay too, but even assuming that they would work, another small bump to 3.0 GHz was not remotely worth the asking price.
The Silencer Mk III is extravagant for this build, but is left over from a previous debacle in which two attempts to procure an ISA-age ATX PSU from EBay got me two bad PSUs and a can full of crazy. Hopefully the omitted −5 V rail won't be missed.
Thermaltake Sonic Tower—the only one left in the world that I could find. As the picture shows, this passive heat sink is adequately served by the case fan and the PSU fan.
The first attempt to use it was an overheat failure resulting from insufficient contact. The root cause was that the heat pipes collided with all four corners of the Socket 478 retention bracket, leaving a larger than permissible gap between the CPU and heat sink no matter how much it was tightened down. The instructions say nothing about this. To resolve the collision I reinstalled the heat sink off-center, which was made difficult by the minimalistic mounting hardware that doesn't want to stay put even when perfectly aligned. Since then, the CPU has stayed plenty cool without an additional fan. The biggest remaining problem is that the tower's overhang makes it extremely difficult to check the condition of capacitors near the CPU.
The 1″ screws in the K7 kit of the Sonic Tower are perfect for attaching a fan if need be, but forget about attaching the fan bracket on the bottom of the cooling tower without a complete disassembly. If a fan is installed, it blocks access to the P4 power connector, the CPU fan header, and the DIMM slots.
XFX NV44A GeForce 6200, PCI ID 0x0221, with 256 MiB of DDR2 GPU RAM, a 350 MHz GPU clock, and a 532 MHz memory clock. Listed as "Best Data Products NVIDIA GeForce 6200 (GF6200AGP256) 256 MB AGP 4x/8x..." with a stock picture and stats of the wrong card.
Whatever. Any 6200 would do to get a DVI output and reasonable performance without adding a fan or generating tons of heat. Gigabyte's GV-N68128DH GeForce 6800 would have been a contender, but none were listed.
For starters, I'm trying out this CT2800 Vibra, FCC ID IBACT-SBV16S. Evidently a "Value" Sound Blaster that lacks bass and treble controls, it nevertheless has a genuine OPL3 (YMF262-M) and Yamaha DAC (YAC512-M).
ACubic Art Black Widow? Maybe? Or just in the wrong box? Listed as "Athena Power CA-GSB01DA Black 0.8mm SECC Steel ATX Mid Tower Computer Case - OEM" (Newegg).
It turns out that a totally plain mid-tower ATX case is hard to get. I managed to avoid PSU at the bottom, sideways hard drive cage, all-5.25″ drive bays, and drive bays with hinged covers. Fortunately, the tool-less stuff was removable.
One of several variants with the model number XJ-HD166S, this kind of optical drive is new enough to read DVD-Rs reliably, yet still has navigation buttons for playing CDs though the sound card without any software. In later years, optical drives first lost the navigation buttons and then the DAC; many of them have putative audio-out headers that actually aren't connected to anything.
Noise control was a huge success. The case fan just makes a slight buzzing noise.
Memtest86+ 4.20 reported memory bandwidth of 857 MB/s with the 2.6 GHz CPU and either 1 or 2 256 MiB DDR-266 2.5-3-3-6 non-ECC DIMMs (mixed brands). With the 2.8 GHz CPU it says 859 MB/s.
The on-board 6-channel AC'97 sound doesn't work. Drivers detect and operate the device but no sound is output. This was tested before an ISA sound card was installed.
The case's front panel USB and audio connectors don't fit the headers.
ISA DMA works.
The CT2800 Sound Blaster is one of those sound cards where you can set the volume at 100% and not get any distortion. However, the synth is relatively quiet, requiring voice to be turned down for balance.
Connecting the Sound Blaster's PC_SPK input required an improvised 4-to-2 pin cable. (Sure enough, it goes "beep.")
Power consumption ranges from 55 to 115 watts with a power factor of 0.96 or better. With the 2.6 GHz CPU the range was 49 to 102 watts.
With the CT2800, there were occasional sound glitches in DN3D that have not occurred with other motherboards.
Later I found that Yamaha OPL3-SA3 (YMF718 and YMF719B) sound cards are very glitchy with this motherboard. Both PCM and OPL3 are affected.
Lots of things work well enough, but it seems clear that the PCI-to-ISA bridging is not equivalent to what Intel included in earlier chipsets.
The first motherboard I received, a model QDI P2DI-AL/U, had a bad case of bad caps and was replaced under warranty.
The replacement board is QDI P2DI-AL/C instead of QDI P2DI-AL/U, came with an older BIOS (both are from 2002), and lacks the Fujicon caps. The Nichicons, however, are the same kind, so it seems I am doomed to a bunch of recapping anyway.
IBM Deskstar 120GXP model IC35L060AVVA07-0, 60 GB, 7200 RPM, ATA-100, lot of 2.
On the first one tested the spindle motor was remarkably loud. SMART reported nothing out of the ordinary except that Power_On_Hours was 86717. With a date of manufacture of September 2002, that would mean that the hard drive had been run 24/7 for nearly 10 years. The other one was quieter and reported 41843 power-on hours (4 3/4 years).
However, the showstopper was some kind of firmware conflict. Two different drives of this model caused two different computers to hang during boot. If the BIOS was configured to ignore the drive, booting would proceed normally from other devices, and the drive could then be accessed from Linux with no hint of trouble.
To avoid further delays I ended up pulling a working drive from another PC. The salvage drawer is full of 4 and 6 GB drives, but Slackware alone uses more than 6 GB.
Lot of 4, as-is. The first one was DOA. Another one initially tested OK but over time proved to intermittently corrupt and destroy floppies when writing to them.
Award Modular BIOS v6.00PG
| First board | P6I845 PlatiniX-2DI BIOS V1.7PMC Aug.16,2002 |
| Second board | P6I845 PlatiniX-2DI BIOS V1.4PMC May 13,2002 |
| Mentioned in a lost Lenovo archive | P6I845 PlatiniX-2DI BIOS V1.8PMC Dec.06,2002 |
Can't find an update anywhere. The Wayback archive of QDI's web site has no BIOS listing for the 2DI boards. The closest thing there is for the 2D boards, but it's not the same. So when the 2.8 GHz CPU arrived, I copied out the BIOS from the first board and flashed it onto the the good board before switching CPUs.
Fortunately, the BIOS does support manual reservation of IRQs and DMAs for non-PnP ISA cards.
| JAV | BIOS flashing enable/disable |
| JCC | Clear CMOS |
| JFUSB | Front USB wake-up enable/disable |
| JKB | Keyboard password power-on function enable/disable |
| JUSB | Rear USB wake-up enable/disable |
| JSD | On-board sound enable/disable |
Ironically, the motherboard has a PC-PCI (a.k.a. SB-Link) header to allow a few of the early PCI sound cards to support DOS games without a nasty driver.
All dollar figures are bottom line costs including taxes and shipping (estimated when items were part of a larger order) and subtracting rebates.
Revised 2013-08-04 to account for second bad floppy drive.
| From storage | ||||
|---|---|---|---|---|
| Items | Cost | |||
| PC Power and Cooling Silencer Mk III Series 400W Modular Power Supply | $59.99 | |||
| CT2800 Sound Blaster | $11.50 | |||
| 2 round 6″ IDE cables | $10.12 | |||
| Round 10″ floppy cable | est. $3.99 | |||
| 2 256 MiB PC2100 (DDR-266) 2.5-3-3-6 non-ECC DIMMs (salvage) | 0 | |||
| Maxtor 6E030L0 DiamondMax Plus 8 UDMA/133 30 GB PATA hard drive (salvage) | 0 | |||
| New purchases | ||||
| Items | Total cost | This project | Write-off | Other assets |
| Legend QDI PlatiniX 2DI-AL/C motherboard + SL6PP 2.6 GHz P4 CPU | $72.00 | $65.50 | 0 | $6.50 |
| Thermaltake Sonic Tower heat sink | $54.70 | $54.70 | 0 | 0 |
| ACubic Art Black Widow case | $31.98 | $31.98 | 0 | 0 |
| IBM Deskstar 120GXP hard drives (lot of 2) | $14.99 | 0 | $14.99 | 0 |
| 3.5″ floppy drives (lot of 4) | $16.80 | $4.20 | $8.40 | $4.20 |
| XFX PV-T44A-WANG GeForce 6200 256 MiB | $16.50 | $16.50 | 0 | 0 |
| XJ-HD166S DVD-ROM | $10.99 | $10.99 | 0 | 0 |
| SL7EY 2.8 GHz P4 | $6.50 | $6.50 | 0 | 0 |
| Shop supplies | ||||
| Supply | This project | |||
| CR2032 battery | est. $3.53 | |||
| Arctic Silver 5 | est. $0.75 | |||
| Screws, cleaning supplies, audio cables, misc. | est. $5.00 | |||
| Totals | ||||
| Total project cost | $285.25 | |||
| Total new purchases | $224.46 | |||
| Total waste | $23.39 | |||
2014-08: Since last year the supply of high-end ISA motherboards on EBay completely dried up. Fearful that the supply of Socket 478 CPUs would soon follow, I ordered another 2.8 GHz one (now $10.60, no big deal but still a 63% year-on-year increase) and got on with trying to salvage the motherboard with the leaking caps.
After replacing twelve 3300 µF Nichicon HMs and the two failed 1000 µF Fujicons with Panasonic FRs, it was an epic struggle to get the CPU socketed. The pins were only slightly bent, but it was enough, and most attempts to straighten them only made things worse. I now hate Socket 478.
Shockingly, the board and CPU both survived, and the hasty build turned out better than expected.
Memtest86+ 5.01 reports only 747 MB/s. Eh? Memtest86 v4.3.6 says 763 MB/s. Memtest86+ 4.20, the version used last year, says 865 MB/s. Aha! Results from different versions are not comparable.
The AC'97 audio doesn't work on this board either. Weird.
A 6800 GT puts out a lot of heat for a closed-front case, but this was the only case I had on hand and the 6800 GT was the only AGP 4×/8× card left in the drawer. Fortunately there's enough air from the side fan to avoid a meltdown. The bottom half of the case gets warm but not hot.
Apart from that and the extra fan noise, the only negative on this build is the lack of a side window for keeping an eye on those caps.
White... beige... It matched just fine until I took that picture. Ignore
that Athlon XP badge too.
