The same trend carries over into, you guessed it, computer hardware purchases. When building a new system from scratch you'll probably cut a few corners here and there, sometimes without even noticing it. You'll opt for a somewhat smaller case, a slightly less powerful speaker set, and most likely, that 24" Sony monitor won't be finding its way onto your desktop anytime soon either. Almost by definition, we are all, as humans, inherently creative in our own unique ways, and by looking around at the methods used to cut costs when buying computer parts, there is a definite level of creativity that goes in to every single purchase where cost is a consideration. The only real area where there has been a general lack of creativity in terms of "cost management" (aka not buying that 136GB hard drive you've always wanted) is in the high-end workstation/server market. The reason here is simple as well, high-end has always meant high-price, and if a company can get away with charging an arm and a leg for a "high-end" solution, then you better believe that they're going to charge an arm and a leg for that very solution.

Luckily, there happened to be an individual that managed to discover one of the very rare ways of being creative with cutting costs in the high-end workstation/server market. His name? Tomohiro Kawada of Kikumaru's Technical Laboratory. His claim to fame? Discovering and successfully implementing the modifications necessary for allowing Intel's "low-end" (and thus, low-cost) Celeron processor to run in a multiprocessor configuration. Unfortunately, Kawada's method for enabling multiprocessor support on the original Slot-1 Celeron and Celeron A processors was a bit over the edge for most users who didn't want to risk damaging their processors to try the neat trick. The trick involved physically "shaving" off, or drilling into, the actual PCB (Printed Circuit Board) of the Celeron processor as well as a number of other modifications to the processor in order to achieve success. While your average engineer would get a kick out of Kawada's method, it was quite impractical for the everyday user to explore.

It was obvious, from the success of the original Dual Slot-1 Celeron project, that Intel had not included an incredible level of protection from running the Celeron processor in a dual processor configuration (4+ slot-1 processor configurations are not possible), and if one were to successfully implement a dual Celeron 300 configuration, clocked at 450MHz, you would theoretically have a dual 450MHz "high-end" workstation at a cost lower than a single Pentium II 450. The problem here was that overclocking two Celeron 300As to 450MHz was considerably easier than manipulating the processors to support operation in a multiprocessor configuration.

Intel originally chose to disable multiprocessor support on the Celeron line of processors as an attempt to more distinctly separate their processor brands. If the Pentium II was supposed to be the high-end flagship, and the low-end Celeron solution was allowed to run in dual processor workstations, customers would begin to wonder why they were spending a premium on the dual processor Pentium II systems when the Celeron could do the same thing. Most AnandTech readers already know that the Celeron 300A, when overclocked to 450MHz, offered, at the time of its discovery, unparalleled performance for the price. However, for a graphics artist, software developer, or engineer, even a single Celeron running at 450MHz would be considered slow in comparison to a beefy dual processor system that most professionals of that sort are used to. Luckily for Intel, the Celeron was kept away from the graphics artists, software developers, and engineers who reluctantly shelled out the extra cash for dual processor Pentium II systems.

With the release of the Socket-370 Celeron, many wondered if a motherboard manufacturer would be bold enough to release a motherboard equipped with two Socket-370 CPU sockets as to encourage more experimentation with dual processor Socket-370 Celeron solutions. The rumors of the possibility quickly died out, and the idea of a dual processor Celeron system remained out of the reach of everyone but those with the skill and the courage to attempt the basic dissection inspired by Kawada's method.

Celeron Processors & Cards Provided by Azzo Computers

How is it done? AnandTech took a look at the process, and attempted to simplify it as much as possible, so without further ado, here is a step-by-step guide to modifying a Socket-370 to Slot-1 adapter for use in a dual processor configuration.

• 2 Socket-370 Celeron processors of the same clock speed, preferably of identical steppings/revisions.

• 2 Socket-370 to Slot-1 adapter cards, AnandTech used the Microstar MS-6905 cards provided by Azzo Computers. Some adapter cards will differ in the modification process, so you may have to consult the technical documentation on your specific parts if using anything other than the MS-6905.

• Tape or Soldering Iron

• Scissors, preferably an exacto knife or anything able to make small cuts out of the tape for hard to reach areas

• Wire, the smaller the better, ideally you'll want to have insulated wire although it is easier to setup with bare wire

• Anti-static wrist strap (optional)

Although this is a step-by-step procedure, it is recommended that you read through the entire procedure before beginning.

Step 1)

First things first, be sure to either put on your anti-static wrist strap, or ground yourself by touching a part of your computer's case. After doing so, unpack the first Socket-370 to Slot-1 adapter card and turn it face down so that the Socket-370 CPU socket is facing the ground and the back side of the card is facing you.

Step 2)

Test both of the cards individually, making sure they work properly with your Socket-370 processors. Also test both cards/processors out at any overclocked speeds you are hoping to run them at. After the testing, remove the cards from your motherboard and be sure to ground yourself once again if you're not using an anti-static wrist strap.

Step 3)

If you're going to be soldering all connections then you can skip this step. Place two strips of tape on the PCB. Place one along the left edge of the back of the Socket-370 CPU socket and place the other above the Slot-1 contacts as shown in the following diagrams:

Step 4)

Cut approximately 2" (5cm) of wire. 2" is more than enough for a single modification, but it's always better to have more than to have less. You don't have to pay a visit to your local electronics shop to grab some wire, you can use pretty much any wire you could think of. If you have an old fan that isn't being used, a single wire from the power connector would do, although the thickness of the wire is going to be a difficult thing to work with. The best case scenario would be using a wire that is about the thickness of one of the traces (etchings) on the PCB itself. If you're planning on soldering all of the connections (the recommended method, although it does require some prior experience with a soldering iron) then it is best that you use an insulated wire.

Step 5)

Locate pin AN15 (7th pin from the bottom, located in the first column) on the back of the Socket-370 CPU socket as illustrated on the diagram below and solder (or tape) one end of the wire to this pin. Be sure the wire does not touch any other pins or part of the adapter card, remember that even the slightest contact between the exposed wire and the adapter card can cause the board to short and could potentially ruin much more than your $20 investment. You can use strips of tape to prevent any exposed wire from making contact with the adapter card. Any part of the wire that happens to be insulated won't cause any problems and shouldn't be of any concern to you.

Step 6)

Locate pin B75 on the connector portion of the adapter card, it is the second pin from the left of the notch separating the SEPP connector into two parts. Solder or otherwise attach the remaining end of the wire to this contact. This connection is a tricky one as it may shift or move as you install the card. AnandTech had a unique experience with the Microstar cards, as one card allowed for a wire to be placed through a hole which was connected to a trace which led to B75 while the other card had this hole sealed. Hold your adapter up to the light to see if any light shines through the hole, if so, then simply run the wire through that hole, otherwise you'll have to make direct contact with B75. You'll have to toy with the connection between B75 and your wire quite a bit before you can truly be sure that the wire won't shift. AnandTech did have an experience where the wire shifted and made contact with an adjacent pin, although it didn't harm the adapter card or the processor on the board, it could have, so be very careful.

Step 7)

After checking and double checking your work, turn on your system with the first modified Celeron in place. If everything goes ok, your system should boot properly after detecting a Celeron processor. If you don't hear any fans spin up, or the system doesn't boot, immediately unplug the power to the system and re-examine your work. Bad signs to watch out for are no fans spinning, no disks spinning up, and of course, smoke ;)

Step 8)

Repeat steps 1 - 7 for the second processor, being sure to test the second processor by itself, and then with the first processor. If everything worked ok, your motherboard's BIOS should report an Intel Celeron processor running in dual processor configuration. Most BIOS' indicate a multiprocessor system by placing the number "2" after detecting the processor type or something of that sort. If only a single processor is detected, you'll want to double check all of your connections, while being sure to check each card individually before re-attempting the dual processor setup.

Step 9)

If everything finally works out ok (don't worry, you're not expected to get it 100% right on the first try, I had a nice little time getting the second processor card to work properly), cover up your system and start up your favorite multiprocessor operating system. Remember everyone, Windows 9x does not support multiple processors, your second Celeron will go wasted. Be sure to read AnandTech's other multiprocessor articles about the true benefits of and for background information on multiprocessor systems.

Step 10)

Enjoy!

Setting up the AnandTech test bed, the Epox Dual Processor BXB-S Intel 440BX based motherboard was used as the heart for the Dual and Single systems. A single 64MB Mosel Vitalic Memory Man SDRAM DIMM was installed in the first memory bank, then upgraded to 128MB, and in 128MB increments afterwards up to 512MB. The rest of the configuration was as follows:

• 9GB IBM Ultrastar Ultra Wide SCSI-3 HDD
• Matrox Millennium G200 (16MB) AGP Video Card
• Microsoft Windows NT Workstation 4.0 with Service Pack 4 Installed
• The latest device drivers as of March 18, 1999 were installed
• All tests were run at 1024 x 768 x 16-bit color at a 75Hz refresh rate

The performance of the Dual Celeron 450 solution managed to outdistance the competition by a large enough margin to make the relatively small time investment worth the while in all cases with the exception of Adobe Photoshop where the larger L2 cache of the Pentium II begins to shine over the smaller albeit faster 128KB on the Celeron.

Celeron Processors & Cards Provided by Azzo Computers

Just as we originally discovered in the first Multiprocessor evaluation, the upgrade from 384MB to 512MB of RAM produced absolutely no increase in performance what-so-ever, indicating the high-end "sweet spot" for workstations is in the 256MB to 384MB range.

Although the steps can be a bit tricky, the end result is a fairly cost effective dual processor system.  The long-term stability tests AnandTech ran on the modified dual 450MHz Celeron system indicated that the solution, most likely due to its overclocked nature, was not suitable for do-or-die server situations, however it was reliable enough to handle the majority of the stability tests AnandTech tossed at it. 

There is escape for those of you that still don't feel comfortable with performing the modifications to the Socket-370 to Slot-1 adapters, Microstar will soon be shipping revision 1.1 of their MS-6905 boards which will feature jumpers that allow the user to modify the voltage going to the Socket-370 processor.   At the same time, a company in Taiwan has already agreed to sell modified versions of the revision 1.1 boards that are going to ship dual processor ready with the trick already performed.  For those of you that can't wait, and are looking for something "daring and new" to try out (there is only so much you can do with computer hardware that's "daring and new"), give the trick a go.