Near-Final

I was able to install an older 3COM network card I had in storage, get it to work in DOS, and configure Windows 3.11 for Workgroups with Microsoft's TCP/IP to get it on my home network. It works! I since downloaded Internet Explorer 3.0 and an old Netscape version (Firefox not available for Windows 3.11) from oldversions.com. I also installed old versions of Macromedia Flash plugin, Adobe Acrobat Reader, Goldwave (for audio editing), and JASC's Paint Shop Pro. Everything works lightning fast for Windows 3.11. It's amazing how fast older operating systems can be when coupled with hardware newer than their intended usage. It completes similar tasks in Windows 3.11 faster than a Pentium 4 can accomplish them in Windows XP.

I also ran MemMaker on each one of my four boot configurations to optimize loading modules in upper memory and free more conventional memory. These configurations include gaming with EMS, gaming without EMS, complete DOS (doskey, CD-ROM driver, etc.), and complete Windows (CD-ROM, network, etc.). My gaming configs result in 604KB free of conventional memory, while my complete configs are between 560KB and 580KB of free memory. I took all of my old batch files, custom menus, etc. from my original retro PC and copied them over. Now the PC is complete software wise, but I'll be doing benchmarks and diagnostics on it just to test everything for the long haul. I definitely want to see the exact performance marks for the 266x CF card over UltraDMA-33.

Since many of the old games I have on this system use the default PC Speaker instead of a sound card, I ran a cable from the motherboard's PC speaker pins to the input on the sound card. Only older sound cards such as the Sound Blaster 16 ISA I am using have an input for the PC speaker and support a separate volume control for it. I reduced any possible RF noise on the line with the cable passed through and wrapped once around ferrite cores at each end. The quality of the sound output through Klipsch speakers dedicated to this PC is amazingly clean. Those old Sound Blaster 16 ISA cards had a surprisingly high signal to noise ratio given their date of manufacturing.

I still have acrylic fan grills (tribal-design) on order to replace the metal ones, but here's some pics of the near-final PC. These are both with and without the blue EL lights powered on:

Painted Ribbon Cable

For floppy and IDE cables, the "round" ones out there are not really round, but rather ribbon cables stuffed into a round sheath. Some might actually have individual strands (as with some higher end round cables like Antec), but cheap ones are often just regular ribbon cables that are crushed and compacted into a round sheath. Therefore, I really dislike these cheaper round IDE cables. They just look and feel cheap. The quality round cables are rarely sold in a simple black color; even the pure black ones are not as simple looking as I'd prefer.

However, I've always enjoyed the look of the black, flat ribbon cables Asus includes with their motherboards. They seem so much more elegant and aesthetically pleasing when compared to "round" ribbon cables. Flat ribbon cables also have the added ability to lay flat against a surface, and make folds when changing direction or position. More importantly, flat ribbon cables can be shortened to any length, because adding a connector is simple with the right crimper and connectors.

Unfortunately, it's nearly impossible to buy black ribbon cable that is flat. I've tried to buy them individually, and I've tried to buy it in bulk. The consumer is out of luck here.

After researching and dumping too much of my time into something trivial to others, I found a simple solution that some have already known. Vinyl spray paint. I purchased a can of gloss black vinyl & carpet spray paint from my local automotive store, then pulled out a spare 80-conductor IDE cable to give it a shot. After taping off the connectors with blue painter's tape, I discovered that vinyl paint is really the best way to go! It bonds to the ribbon cable, looks better than the no-gloss black ribbon cables from Asus, and doesn't flake or peel when bent, twisted, and scratched. Whoot!

Here's a pic of the one I painted (80-conductor, dual drive) with Duplicolor Vinyl & Fabric Gloss Black, compared to an Asus cable (40-conductor, single drive). I also painted the various color connectors black with a paint marker, except on their pin side where they won't be visible.




Sleeving

As with any PC mod project, sleeving always seems to take longer for me than any other step. May be it's because I put so much love into my cabling, or may be it's because I now solder each connector when splicing or shortening cables. Either way, this step took even longer for this project because the black sleeving I used wasn't 'black' enough for me. It is nice sleeving, but one layer doesn't seem to black-out the color of the cables enough for my tastes. So all of my cables have two layers of sleeving.

Here's the results:

Component Housings

As with any mod I do, I painted the component housings (CD-ROM, FDD, and a junction box) semi-gloss black. This gives it a nice, satin-like finish, without being dull or shiny. I masked off the plastic LED covers of the floppy faceplate, and didn't bother painting the CD-ROM faceplate since it was already black. I did have to later apply some hand glaze to the CD-ROM faceplate to polish a little "wet look" into it to match everything else. For now, here are the painted covers and faceplates:




EL Lights

Since I strongly feel that CCFLs and clear acrylic cases are poor combinations, I'm opting to go with a little bit of EL lighting for highlights. It's a subtle glow around just the light strip/wire itself that adds a nice touch to clear cases. Of course my other alternative is to go UV with UV CCFLs, but I'm really bored with the whole UV craze. I'd much rather have something that looks like it came from the movie "Tron" than something out of a really bad Rave party.

The kit itself, both powered off and on:



Since the inverter for the EL lights emits a high-pitch whine, I decided the encase it inside my own enclosure. The whine seems to be curtailed when the bottom of the unit is sealed by a dense mat or rubber, along with some weight or pressure on the top to keep it sealed. Therefore, I put both molding tape and part of a rubber mouse pad under it, then stuffed it into an enclosure with foam to reduce the shrill sound leakage. I didn't want to use a clear enclosure here since the use of all the foam.




Voltage Conversion

For such a simple issue, this is by far one of the greatest portion and challenge of this project. The problem: many high-end, modular power supplies do no supply a -5V wire (white) to the motherboard, meaning they are unable to generate -5V at all. It isn't needed for newer motherboards, because only the old ISA bus (its controller) utilizes -5V. Some mid to low-end power supplies still offer this -5V output, but I have my mind set on an Enermax, SeaSonic, or PC Power & Cooling. I purchased an Enermax Liberty 400W for this project because of both its good build quality and its use of modular cables. Although I will be cutting and splicing cables as I see fit, I still enjoy the modular output design.

To accommodate my choice, I need to design and implement a separate device to output -5V to the motherboard. After some research, I discovered that an electronic component known as a 7905 voltage regulator will take in -12V (already supplied by a PSU) and output -5V. The only problem is, this is a PC board mount style component with connections that should be soldered. To make it worse, the voltage regulator itself is not enough for reliable operation. It needs capacitors on both its input (-12V) and output (-5V). And if that couldn't get any worse, I was never good at soldering and ditched the notion of it many years ago.

Fortunately, with the help of the community members at Electronics Forum, plenty of online research, electrical design software (KiCad), and many purchases at my local Radio Shack store, I was able to fabricate this voltage conversion unit, and even make it looks presentable in the process. Furthermore, I learned to embrace soldering and pick up hobby electronics.

I first double-checked the simple design in KiCad. It was complicated for me, being an electronics noob, so I wanted to make sure it looked right with those on Electronics Forum first.



I then tested everything in a breadboard before actually soldering together components. Instead of my KiCad plans, I decided to go with 22µF capacitors on both the input and output to ensure stability.



I used a clear acrylic card holder box as the housing of the unit to keep in line with my case's clear theme. I drilled holes in the side for binding posts, and larger holes at the ends to help with ventilation.



Top (component side), bottom (soldering and wire side), finished with lid off, and finished with lid on:




Power Supply

The power supply is one of the last components I'll have completed. The -5v issue, conversion from ATX to Dell proprietary power connector, and even sleeving have delayed it. I found a great reference for converting to/from Dell's proprietary power connector on someone's personal site. Here's what I've done so far with the Enermax Liberty 400W PSU for this build:




System Preview

Here's a quick shot of the front-panel pcb before I modded it. I ended up removing (desoldering) components, cutting down the height of it, and then painting it black as shown later. I did remove the front-panel audio and USB connectors before taking this image.



A couple shots of the partially-installed PC in the modded Sunbeam UFO case:

I grew up with the really old games for the Atari, NES, Tandy TRS-80, and 8086, 386, and 386 systems. Some of those memories are quite fond, while others are hair-pulling nightmares of compatibility, performance, and configuration problems. I want to avoid the nightmares and focus on the best memories of these golden days by building an old PC to keep DOS 6.22, Windows 3.11 FW, and a host of games installed. I am sentimental about DOS, as I spent many years of my youth becoming a guru of DOS batch files and memory tweaking.

More importantly, I want to challenge the notion that old hardware architectures are obsolete. To do this, I'll be bridging the gap between multiple generations of computer hardware to use classic computer games and other software. By bridging, I am specifically taking new PC components, and adding them to older architectures that include the motherboard, CPU, and memory. Older hardware from the mid to early 90s is commonly known as "vintage" and difficult to find, even on eBay and with vintage hardware resellers.

In the future, as more and more vintage components suffer failures due to age, it will be even more important to support the means of bridging new and old components. Some may be more than content in waving goodbye to vintage architecture. Others, such as myself, hope to preserve some vintage hardware. There are those who would argue that emulation is the most reasonable solution to running vintage software, but I find little value in this because it degrades the experience of running vintage software on its original hardware architecture.

How will I accomplish this, you might ask? It's impossible to take on the entire responsibility of preserving vintage hardware and software myself, and I know there are plenty of individuals and organizations already doing this by creating their own 'computer museum'. Typically, computer museums do not focus on operational equipment, and when they do, they try to keep entire systems in tact. This is a flawed ideal, as many components inside a vintage system were never designed to remain functional for more than five years, let alone a decade or two. My solution takes this into account by offering the concept of bridging the hardware generational gaps. I will be hunting for solutions to run old hardware (specifically the motherboard and processor) with new hardware.

For example, this project of mine will take a vintage Dell P100c XPS computer and install it in a new case, with a new power supply, hard drive solution, CD-ROM, floppy drives, cables, cpu cooler, and a fan. There are many hurdles I've encountered, but nothing that has prevented me from making everything work so far.


Proposed System Specs

Case: Sunbeam UFO clear acrylic ATX (ACUF-T)
PSU: Enermax Liberty 400W (ELT400AWT)
Mobo: Dell Socket 5 XPS P___c, last rev. (AA 649849-601)
CPU: Intel Pentium 100MHz, Socket 5
Cooler: Modified Intel stock Socket 478 heatsink
RAM: 4x 32MB (128MB) 60ns EDO 72-pin NEC modules with Motorola chips
Video: Number Nine S3 Virge PCI
Audio: Creative Labs Sound Blaster 16 ISA (the original 92 vintage card)
Optical: Sony Optiarc IDE CD-ROM (CDU5225)
HDD: Kingston Ultimate 266x 2GB Compact Flash w/ Syba IDE to CF adapter
FDD: TEAC dual 5.25"/3.5" floppy drive (FD-505, Dell part #79788)


Orignal Dell XPS PC

Here's what I have to start the project. It's a Dell P100c XPS that has a Pentium 100MHz CPU, 16MB of RAM, a 2GB hard drive (originally a 400MB drive), a 4x CD-ROM, 3.5" floppy, a cruddy 15-bit ISA sound card, and a really nice Number Nine Imagine 128 video card. I will be completely removing everything useful from this beige case, installing it in a new case, and purchasing new hardware to keep the system running as long as possible.

I originally purchased this old system for about $20 from my local university's salvage center (Penn State Salvage), and was able to find one in which everything works. I dumped all of my old software from archived CDs onto this PC's hard drive, and configured it to run it all. After a few years, I'm now ready to give this old PC some love by rebuilding it.

The first step is to take all of the existing data off of the old, noisy hard drive (before it fails), and transfer it over to a Compact Flash (CF) card that attaches to the PC via an IDE to CF adapter. This will ensure that the data is preserved in digital format for longer than is possibly with a mechanical hard drive. I purchased an adapter and a 2GB Kingston Ultimate (233x speed) CF card. I partitioned it into two partitions, one 500MB for the DOS, Windows, applications, and utility software, and the remaining space into another for games.



This is what is left over after I've gutted the system:




Case

I purchased a Sunbeam "UFO" case, which is a clear acrylic, cube-like case. The motherboard and expansion cards lay flat on the top section of the case, while the power supply and drives are mounted on the bottom. This provides both the best isolation of heat zones, as well as reducing strain on the motherboard compared to mounted vertically in a tower-style case. It provides the benefits of a desktop case, but without the large footprint.

Of course I selected clear acrylic since this is a collector's PC that is meant to have its hardware visible. Why spend all of this time on restoring old hardware if I can't occasionally view it for a trip down memory lane? For this case, I first needed to drill my own holes for the motherboard standoffs (case is standard ATX, but motherboard is a proprietary AT version). I also needed to move the entire rear bracket that holds the expansion card slots' mount by making a new cutout, moving it, drilling new holes, and finally patching over the existing ATX cutout with more clear acrylic. Once this was done, I even needed to flip one of the corner mounts so that it would fit next to the new layout of the rear bracket. All-in-all, a tedious but rewarding process.




Motheroard

The motherboard of this system is quite a rare find. It's a top-notch and reliable socket-5 Dell board that includes an Intel chipset, 4 banks for 30-pin RAM (70ns regular to 60ns EDO), and even a removable cache module. This was built when the term "external cache" for processors really meant that it was 'external'. Many other boards at this time implemented cache with small IC sockets with retention contacts to hold memory chips. These were always a pain to both install and remove. The cache module is a proprietary solution, so it is far more expensive, but it is a welcome feature that makes the task of swapping cache extremely simple.




CPU Cooler

Fortunately, the heatsink mounting system for Socket 5 is extremely simple. A straight, metal bar fits through the middle of the heatsink and clips onto the socket housing directly on two sides with clips. This is a simple and easy method for older heatsinks which were little more than flat pieces of metal with grid-style fins. However, mounting new heatsinks this way is a chore.

I had to drill a large hole through the entire heatsink, then cut out a small section below this hole on one side to get it installed. The clip now runs through the heatsink to clip to the socket housing instead of over it.




Motheroard

After researching the modder trick of painting a motherboard, I felt confident enough to try it on this vintage PC. I used non-conductive paint (flat-black modeler's enamel), marked off all contacts with painter's tape and pin headers, then painted it using a model paint brush. Everything seemed to go well, and I tested it in between painting different sections to make sure it worked. I let the paint thoroughly dry each time. Unfortunately, the last section I painted near the chipset caused the motherboard to fail on the new power up test. It either forced a short, or forced an existing trace to never make the electrical connection it needs. I can't be sure which, but this has discouraged me from ever trying this again.

It was still an enjoyable experience, and now a nice piece of art to hang on my wall.



After hosing my original motherboard, I had to find an exact replacement (newer revision, but same board) to keep my original layout plans. I was very lucky to find one on eBay (the only one I found!) that was guaranteed against DOA. Lucky me.

I also purchased 128MB of RAM (new stock from Memory Ten) to max-out the board's installed memory, and another couple of Pentium 100MHz CPUs. I purchased pretty much everything individually again so that I am really using nothing from the original Dell XPS system anymore. The original CPU, memory, and cache are all on hand as back-ups in case the ones I have fail in the years to come.

For the memory, 128MB of RAM was almost unheard of in the days of the Pentium 100, because it was far too expensive to own. What might have been up to $2,000 for 128MB of 60ns EDO memory back then is now $5 a stick for a total of $20. I had to purchase another CPU, because the original one I have is almost permanently fused to the heatsink. Since I want to use my own heatsink (from a socket 478, stock Intel cooler), I needed another CPU that was bare. In those days, Dell believed in attaching heatsinks to CPUs via an extremely difficult to remove sticky thermal pad. Removing it would risk damaging the CPU itself.

The "secret" to the future of gaming lies in the fact that bandwidth means little if one is unable to move that much data fast enough through the bus. This is comparable to other technologies such as UltraDMA-33, 66, 100, and 133, as well as SATA 1.5Mb/sec (150-183KB/sec) and SATA II 3.0Mb/sec (300-366KB/sec). Or even network speeds 10Mb/sec, 100Mb.sec, and 1Gb/sec. The bandwidth may be large, but the bandwidth may only be reached a fractional minuscule of the time data is being moved across it. Every single SATA II drive out there right now (except may be SSD drives) wouldn't be able to move data any faster than if it were regular SATA 1.5Mb connection. Of course burst transfers may peak past this, but it really doesn't affect overall performance.

So, with that being said, PCI-Express 2.0 will suffer from the same limitations in the near future. Someday it will be useful and GPUs will utilize some of that extra bandwidth, but for now, it's just future-proofing. This is only indirectly related to quad SLI/Crossfire, because having more bandwidth and additional GPUs doesn't change that all of the video data traffic still has to go through the PCI-Express bus. If you saturate the bus (only xx number of PCI-Express lanes are available), additional graphical processing power won't help much.

This is why PCs still have bottlenecks. The front side bus is a killer for Intel CPUs, because it's a serious factor for bottlenecks, including the Qxxxx quad-core processors from Intel. If we we weren't limited by the FSB, our rigs would scream with speed! This goes back to the old comparison of latency vs. speed. Always put latency as a higher priority than speed, because latency is your biggest bottleneck that will make even the fastest rigs appear slow for mundane tasks. For a hypothetical example, if data takes 2 pico-seconds to reach a CPU, but the bandwidth can support 128 of these bits of data at once, it still takes 2 pico-seconds for the data to arrive, regardless of how many pieces of data are traveling at one time (up to 128, with additional data taking extra clock cycles).

So if a relatively small chunk of data needs to be processed, it will take the same amount of time to move around a 1,333MHz bus on a 3.6GHz CPU as it would a 2.4GHz CPU.

This is only one of the many reasons why there are diminishing performance gains to multiple CPU cores, multiple GPUs, and multiple video cards. On to trend forecasting...

If you're interested, you might want to check out the history of the PCI bus and how it overtook the VESA local bus addition to ISA. It has a very similar tech evolution when comparing PCI-Express and PCI/AGP. Look specifically into the reasons for moving to PCI, and compare it with what actually happened in the market and why ISA was still used for a long time after for simple expansion cards, such as sound and I/O.

For quad SLI/Crossfire to be successful, the market switch from PCI-Express to PCI is moving far to slowly. Manufacturers may try to market and push for this configuration, but it will not be nearly as profitable as it could be with our current limitations.

Imagine the possibilities in the future:

If you have a motherboard with all PCI-Express 16x slots (ie: 6 to 8 of them), both SLI and crossfire become a thing of the past provided drivers mature to handle all multiple GPU configurations across the bus only (possible with the increased bandwidth of the PCI-Express 2.0 standard. Imagine being able to place any number of video cards along this bus and have them talk to each other, designate one as a "master", and output rendering frames to a monitor utilizing the processing power of multiple GPUs. Imagine if it didn't matter what board you have, as it would perform this functionality on video cards with any manufacturer's GPU.

That could be the future of high-end gaming, not SLI or Crossfire per say.