Whether you use your Linux box as a server or as a single user workstation, you will find that the CPU is very often idle. This information is provided by the /proc/loadavg kernel interface or by running a utility that will signal the CPU load (which itself just reads and interprets /proc/loadavg). Even on a mildly used Web server, load average rarely goes beyond 10%. What does the CPU do when it is idle? Running Linux in its i386 flavour, the CPU when idle will execute a HALT instruction and wait for an interrupt (I don't know about the other architectures). You can check the code in the Linux kernel source under /arch/i386/kernel/process.c.
The problem with Cyrix 6x86 (and AMD K6) CPUs is that they generally run much hotter than their Intel Pentium counterparts. Cyrix used to sell their 6x86 CPUs with huge heatsinks and fast (and very noisy) fans. The following graph displays maximum power consumption for Cyrix/IBM 6x86 CPUs under normal operating conditions:
However, a clever feature of the Cyrix CPUs is to enter a special power-down mode when executing a halt instruction, that will cut its rated power dissipation from 20 W or so down to around 250 mW (or even less than 150 mW on the newer 6x86MX chips). So, if you don't use your Cyrix box to heat your house, it would clearly be the Right Thing to enable this feature. Note that it imposes absolutely no speed penalty on the CPU, i.e., the CPU wakes up from a HALT instruction just as fast with power-down mode enabled as when it is disabled.
AMD K6 processors will also consume less power when HALTed (< 2 W). This does not need any special setting, but the power savings are not as significant as with 6x86MX processors.
Beware: using the 6x86 power-down feature does not mean one can do without a CPU heatsink.
I recommend a good heatsink/ball bearing fan and a paper-thin layer
of silicon thermal compound applied on top of the CPU. Ball bearing
fans (versus sleeved fans) are a must for long operating life and low noise.
A
13 mm (½") high, aluminium black anodized heatsink coupled
to a ball-bearing fan will provide adequate heat dissipation in most cases.
A metal clip should be used to firmly press the heatsink against the CPU.
In my experience, a 6x86L CPU running at 133 MHz will function perfectly
in ambient temperatures up to 45 °C, if adequately cooled.
I have had good results with CoolerMaster heatsinks. The SB-TI5-4515C2 is a ball bearing fan, 13mm (½") high, aluminium blue anodized heatsink adequate for normal operation of most 6x86L and 6x86MX CPUs. The SB-TI5-5020C1 is a similar, but larger, taller (19mm) model with a larger and more silent ball bearing fan, providing adequate cooling for high clock rates CPUs (JPEG image below). These are Taiwan-made, relatively inexpensive (< $10) heatsinks, good for at least 2 years of continuous operation. To get the coiled wire, just wind it tightly around a pen or similar cylindrical object.
A small marking on the side of the fan indicates the direction of vertical and horizontal air flow. The vertical air flow is always downwards, and the horizontal air flow should be directed to the on-board voltage regulator heatsinks. The fan can always be unscrewed and rotated to obtain the correct air flow. Note the pass-through power connector.
The need to use silicon thermal compound arises from the fact that metal surfaces are not perfectly flat i.e. a very thin air gap exists between the CPU top metal cover and the heatsink. The problem is that air, short of total vacuum, is one of the best heat insulators. This is clearly undesirable.
Silicon thermal compound is a white paste which you can find at most Radio Shack stores in small tubes of 10g, for less than $2. Silicon paste by itself is not a good heat conductor; the addition of Zinc oxide powder provides it with the desired thermal conducting characteristics, and also gives the paste its white color. You should use as little paste as possible to evenly cover the CPU's top metal surface with a thin layer.
A much neater solution is a grey polymer found on the underside of some CPU heatsinks (e.g. Cyrix labeled heatsinks), in the form of a very thin 38x38mm (1.5x1.5") square . This heat conducting polymer molds itself to the CPU top cover, hence eliminating any air bubbles and providing a nearly perfect heat junction. It also avoids messing with the silicon paste. If your heatsink has this polymer, do not use silicon compound.
You will find two kinds of voltage regulator circuits on Pentium/6x86 motherboards: linear and switching. Switching regulators are more efficient and theoretically should never have any overheating problems, but linear regulators may shutdown when trying to handle the high current loads and temperatures generated by 6x86 CPUs. If your motherboard has a linear voltage regulator with a small heatsink and you can feel it gets excessively hot, try adding a small fan directing the air flow to the voltage regulator circuitry.
Another issue that may plague users of the most recent 6x86MX (and AMD K6) CPUs on older motherboard designs is that the linear voltage regulator circuitry will simply not handle the transient current peaks generated by the newer processors, no matter how well cooled it may be. The only solution in such cases is to upgrade the motherboard, preferably to one equipped with a switching voltage regulator capable of handling such current loads (switching regulators usually exhibit better dynamic behaviour under heavy current loads).
Electrolytic capacitor aging phenomena
Another problem that you may encounter is that on some machines, ramdom crashes may begin to appear after a year of use or so, for no apparent reason.This may be due to electrolytic capacitor aging, a phenomenon that does not affect high quality motherboards built with tantalum capacitors. This aging effect (which translates into lower capacitances and consequently the appearance of decoupling problems) is greatly accelerated by higher internal case temperatures eventually caused by 6x86 and AMD K6 processors.
These decoupling problems are the most frustrating for the average user, because they are not related to either short term temperature peaks nor instruction execution: they look like totally random failures. Since Intel parts usually draw lower currents, they are not as sensitive to decoupling issues as the other CPUs, and will not fail like their 6x86 and K6 counterparts. However, this has nothing to do with CPU quality, it is rather a problem of long-term reliability and quality built in the motherboards.
Replacing electrolytic with equivalent high-quality tantalum capacitors will usually solve the problem, but even though the total cost of a dozen tantalum capacitors may be < $10, this may be a hassle to the average user: it involves disassembling the CPU case, removing the motherboard, identifying and unsoldering the electrolytic capacitors, soldering the new tantalum ones and putting everything back together again.
Again, the only other solution in such cases is to replace the motherboard.
Last updated on January 5, 1998.
Copyright 1997 Andrew D. Balsa