The Official "Overclocking" FAQ
Comp.sys.ibm.pc.hardware.chips "Overclocking" FAQ version 0.5.2

This FAQ is maintained by HB Papaleonardos, hpapaleo@magnus.acs.ohio-state.edu. Any suggestions, comments, questions, or additions are welcome.

Revisions:
0.6.0 9/27/94: Kept Cameron Spitzer's additions, added Richard Knipe's heat transfer experiment, added much info on fans, etc., and made many minor changes. Crummy wordwrapping. If anyone has a great tool/technique for "reflowing" hard word-wrapped text, please let me know. (editor's note: this problem should be resolved at COMPUTERCRAFT) The revision history section has been removed. If you need revision information, contact the maintainer.

Contents:
I. Introduction
1.1. What is "overclocking?"
1.2. What chips can be "overclocked?"

II. How-to
2.1. How do I "overclock?"
2.2. Where can I get a new oscmod?
2.3. What are typical jumper settings?
2.4. Any other settings to change?
2.5.
2.6.
What are typical performance improvements?

III. Dangers
3.1. What about heat problems?
3.2. What are some other problems?

IV. Solutions
4.1. What can I do to reduce CPU temperature?
4.2. What else about fans and heat sinks?
4.3. Where can I get all this stuff?

V. Miscellany
5.1. Availability & Distribution
5.2. Acknowledgements
5.3. Copyright
5.4. Disclaimer

I. INTRODUCTION

1.1. What is "overclocking?"
"Overclocking" is increasing the clock rate of a processor beyond its rating for the purpose of increasing system speed without buying a new, faster, but more expensive processor. "Overclocking" is a slang term, and not an engineering or scientific term. The correct technical terms are "speed-margining" (more common) and "undertiming" (less common) (Cameron Spitzer).

One can also "overclock" the computer's bus. See question 2.4 ("Any other settings to change?") for details.

1.2. What chips can be "overclocked?"
Just about any chip can be "overclocked." The most popular ones are Intel 386/486, AMD 386/486, and Cyrix 486. The Motorola 680x0 series of processors do not "overclock" to as high a margin as to Intel processors and require a much better memory subsystem than Intel because they do not gain much from a second-level cache. In fact, most Motorola 68040 systems do not have a second-level cache (William Fang).

II. HOW-TO

2.1. How do I "overclock?"
Your goal is to increase the clock rate for the processor. This can be done in one of two ways:
1) Change clock oscillator module
2) Change jumper settings

For #1, your oscillator module (clock crystal) will be a little silver box soldered to the motherboard, possibly strapped to the board with a little plastic fastener.

- If you have a 386, the oscmod will be labeled with twice your clock speed. Example: if you have a 25 MHz machine, the oscmod will be 50 MHz.

- If you have a regular 486 (not clock doubled or tripled) processor with an older motherboard, probably the oscmod will be labeled with the speed of your processor. Example: 33 MHz machine has a 33 MHz oscmod.

- If you have an older but better 486 motherboard, your oscmod will be twice the speed of the processor. The motherboard divides the frequency by two to get better accuracy. Example: if you have a DX-33, then your oscmod would be 66 MHz.

- If you have a DX/2 or other clock doubled processor, the oscmod will show half your clock speed. For example, if you have a DX/2-66, then your oscmod will be 33 MHz.

Just cut the fastener, desolder the oscmod if necessary, and replace it with a new oscmod. You will probably want the 4-pin oscillators (they are often described as fitting in a 14 pin DIP socket). See question 2.2 for oscmod sources. It is recommended that you save your old oscmod in case you have to go back to that speed.

For #2, just change the system speed jumpers. See your motherboard manual, call your computer's technical support number, or see question 2.3 as a last resort to find out what they are.

2.2. Where can I get a new oscmod?
B.G. Micro (+1 214 271 5546)
22.1 24 24.6 28.3 32 40 49.15 49.8 50 86

DigiKey (+1 800 344 4539)
24 25 25.175 28.3 32 40 50 64 66.6666 80 made by CTS
24 25 25.175 28.322 30 32 32.514 35 40 48 50 64 made by ECS
(Michael Picone)

Jameco (+1 800 237 6948)
They carry everything, even up to the 100 MHz module. (AH)

JDR Microdevices
16, 25: $3.99
33, 36, 40, 50, 66, 80, 100: $4.75

Mouser Electronics (+1 800 346 6873)
24 25.175 28.3 30 32 32.5 34 35 36 38 40 42 44 50 64

You will be looking for full-size TTL clock oscillators. Telll them that, or oscmods, but NOT crystals, which are 2 pins only and won't work

2.3. What are typical jumper settings?
I don't know. I have seen motherboards with 3 pin pairs (2x3) and two jumpers. For a DX/2-50, pins 5-6 were closed. Anyone else?

2.4. Any other settings to change?
If you have a good CMOS you can change other system speed ratios. You may need to add wait states to your memory and/or decrease bus speed. For instance, if you were using a processor which was 40 MHz but now 50 MHz, your bus speed might have been CLK/5 (8 MHz). Now, if you have device problems, you might want to change that to CLK/6 (8.33 MHz). You may wish to see the "BIOS Survival Guide" on Comp.sys.ibm.pc.hardware.chips for more information.

Bus "overclocking" may not be as dangerous. You go into your BIOS and change your CLK/x to a higher ratio (a lower x). If you have problems, either try to eliminate slow devices, or raise the x value. It only takes one device which can't handle the higher speed to make your system unstable. Incidentally, ISA standards exist for bus speeds as high as 16 MHz, [if you know of a published timing specification for "16 MHz" ISA, please mail me, cls@truffula.sj.ca.us (Cameron)] even though the most common is only 8 MHz. Again, I recommend that you see the "BIOS Survival Guide."

2.5. What are speed changes are typically made?
Very common is 25 to 33 and 33 to 40. Apparently AMD 486- 40's easily go to 50's, because practically everyone is doing it. It seems that DX/2-50's are not very likely to run reliably at 66 for some reason. It is also common for DX/2-66 to become 72 and 80 (lucky). There are some amazing cases, such as 486SX-25 running at 50 for two years, no problems.

There are no intentions to add "real-life stories" of "overclocking" to this FAQ; such stories may be appropriate for another document.

2.6. What are typical performance improvements?
[Specs wanted here.]

III. DANGERS
3.1. What about heat problems?
There are two main types of problems caused by heat, one is fatal, one is not.

The non-fatal one is due to timing. Processors are designed and tested so that their internal parts will all be ready about the same time, according to specifications published by their manufacturer. As the heat of the processor increases above specifications, the internal paths slow down. Some paths slow down more than others, and eventually there becomes a significant difference between when something is expected to happen and when it actually does happen. At this point you may get false information (such as 0+0=1), system lock-ups, or spontaneous resets. This behavior is usually a signal for you to decrease processor speed or temperature (see next section).

One fatal possibility is called electromigration. From Cameron Spitzer:
...overstressed ICs can be slowly destroyed by electromigration. The combination of heat and electric fields cause metal atoms to wander around under the passivation layer.

They tend to grow little whiskers at any sharp corner or irregularity along a trace. Whiskers at different potentials tend to grow towards each other, much as stalactites grow towards stalagmites in a limestone cave, because the sharp point accentuates the potential gradient. In a cave, limestone columns eventually form from floor to ceiling. On a chip, you get a short circuit.

(Heat is only a secondary factor in electromigration. The primary factor is current density in the presence of an electric field. In most digital ICs, the internal clock signal is distributed by a conductor, usually aluminum, which is sized carefully for its load. If the clock switches more often than the designer sized it for, then the clock "trunk" is overloaded and subject to premature failure due to electromigration.)

The other fatal possibility is simply burning out the bond wires that run from the pins on the outside of the package to the silicon die of the processor.

BTW, these problems with heat have nothing to do with the die melting. According to Nick Paizis, most often silicon remains functional well above 125 degrees C. The key is to not exceed the maximum recommended operating temperature.
[Operating thermal range data on any processor is welcome!]

3.2. What are some other problems?
Just because your processor will run satisfactorily at the speed it is running doesn't mean that the rest of your system will. In fact, it is more likely that, if your system becomes unreliable, it is because other devices are overstressed by the higher clock speed. Relatively big ticket items, such as video / disk controllers and memory, may fail because there is not enough time for them to do their work when bus frequencies are increased (See question 2.4). You can try decreasing bus speeds, and increase memory wait-states, or you can get faster devices. Generally, if you get lots of strange errors, such as "No ROM BASIC installed", try slowing down memory and bus speed.

The other device-based problem is with your motherboard. Every cheap little component on the board must function reliably at the higher speed or you will likely experience reliability problems. The only reasonable solution to this problem is to get a faster motherboard.

IV. SOLUTIONS

4.1. What can I do to reduce CPU temperature?
You can try either a heat sink, fan, cooling unit, or a combination. A heat sink attaches to your processor (covering up the fancy logo ;) and helps dissipate heat into the surrounding air. A fan spreads the heat around inside the case (where the case fan should get rid of it). A cooling unit is somewhat more complex (John Loo):
[...] a marvelous invention called the Peltier junction. It has fantastic cooling specs. I have one that is 1.56"x1.56" and it is awesome. It can cool a piece of metal till frost forms, or up to 50 watts of heat removal.
[...] this device cools by moving heat from one side of the device to the other.
[...] approx. 1 watt is needed to remove 1 watt of heat, and it takes at least 5v to run it at high power.

They are rather expensive. You still need a fan, and probably a heat sink, when you use one.

4.2. What else about fans and heat sinks?
There are different heat sinks and CPU fans for different CPUs (size difference). Make sure you get the right one. Optionally you might get a heat sink that is too large for your chip, but you wouldn't do the same with a CPU fan, as it clips onto the chip itself. You should use some sort of heat sink compound to "glue" the heat sink to the processor. The correct heat sink compound GREATLY improves heat transfer. In contrast, ordinary glue is a HORRIBLE heat conductor.

If you are concerned about heat transfer efficiency in your combo heatsink and fan, perform this simple experiment (Richard Knipe):
1) Unplug the heatsink's fan and turn on the computer.
2) Touch the heatsink with your finger. It should get too hot to touch.
3) Plug the fan in and then touch the heatsink after a couple of minutes. The heatsink should now be luke warm.

If the heatsink never gets hot without the fan running, then either your chip doesn't generate much heat or the thermal contact between the heatsink and chip is poor--and you better find out which, else... bad luck! If the heatsink gets really hot without the fan running, the thermal contact is good.

There is some room for concern regarding the CPU fans (the ones that clip directly onto the chip). Some suspect that the vibrations from the fan may damage the processor, or slowly cause a bad connection between the chip and its socket (Natsuki Ishizuka). In this case it may be better to mount the fan somewhere else in the case. But be careful!

4.3. Where can I get all this stuff?
Act-Rx Technology Corp
11F, No 44, Sec. 11, Hsin Seng Road
Taipei, Taiwan, ROC
Tel: 886-2-5683678
Fax: 886-2-5682689

Their product is marketted under ARX CPU Cooler. The part I'd recommend is ACC4846-23/231 and ACC4846-30/301 (with slightly better cooling power) for 486 and ACC5856-30TC and ACC5856-40TC (again with slightly better cooling power). (WH)

Alpha & Omega Computer
+1 714 774 5670
Heat sinks equipped with Peltier junctions. (VW)

Data Products DPS
+1 800 669 8194; +1 303-667-8260 Orders (M-F 8-5 MST)
+1 303-667-6245 FAX
"486/Cooling Fan" $15. Snaps onto 486, 1.5 W, ball bearings, 50000 hours MTBF. Presumably floppy power connector, like JDR fans.

DigiKey
+1 800 344 4539
+1 218 681 3380 (FAX)
Wide selection of oscillators, fans, and heat sinks. (SH)

JDR Microdevices
+1 800 538 5000; +1 408 559 1200 Orders
+1 408 559 0250 (FAX)
"486-FAN" $29.95. Snap-in installation, in-line floppy power adapter. "486-FAN-R" $49.95. Same as above, but with "active electronic refrigeration unit" (probably Peltier junction).

Melcor
1040 Spruce Street
Trenton, NJ 084648
+1 609 393 4178
+1 609 393 9461 (FAX)
The part I got is CP1.4-127-045L, which I now realize is an overkill. You can get a catalogue from them, which contains comprehensive technical specs. (WH)

PC Power & Cooling
+1 619 931 5700
I had a cheapo that got loud in 2 months, then bought this one, still going quietly after about 7 months, computer on 50% of the time. Supposedly they claim their bearings are better. (PB)

Vemaline
+1 800 227 0254; +1 401 739 7600
Well-designed heat sinks and fans; no Peltier junctions. (VW)

[Additions to this section are always welcome!]

V. MISCELLANY

5.1. Availability
Comp.sys.ibm.pc.hardware.chips newsgroup (now posted monthly)

FTP: ftp.rahul.net:/pub/cameron/PC-info

npaizis@sedona.intel.com (Nick Paizis)

cls@truffula.sj.ca.us (Cameron Spitzer)

bbs.kohler@tsoft.net (John Loo)

ishizuka@fl.lab.shionogi.co.jp (Natsuki Ishizuka)

int877w@lindblat.cc.monash.edu.au (William Fang)

wil@shell.portal.com (Ville Walveranta)

mpicone@gandalf.rutgers.edu (Michael Picone)

accsah@vaxc.hofstra.edu (Steven Henry)

skr106@psuvm.psu.edu (Senthil Kumar)

a-hassan@uiuc.edu (Aslam Hassan)

knipe@lobby.ti.com (Richard Knipe)

pieterh@sci.kun.nl (Peter Herweijer)

eng30174@leonis.nus.sg (Weng Hong)

paul_braren@cornell.edu (Paul Braren)

5.3. Copyright
The information contained in this FAQ is the property of the FAQ maintainer (me). I grant permission for free reproduction of it as long as it appears, unmodified, and in its entirety. It may be reproduced for a fee, so long as the fee covers only the cost of reproduction.

5.4. Disclaimer
The maintainer and contributors of this FAQ take no responsibility for any damage caused by following the suggestions contained herein. Absolutely no warranty is supplied by anyone associated with this FAQ.

Last modified: