If you're looking at a sexy little "Chrome Orb for Socket A 1Ghz" the calculation could be (based on Thermaltake's figures, inclusive of the "high-quality Chomerics T725 compound" - ie the thing you buy has an Rj-a°,C/W of 0.81):

Overclocked Watts = [32.4 ([900/650]{[1.85/1.7] ² }) = 53.1W

CPU temp @ 100% load = [20 + [say] 5] + [0.81 x 53.1] = 68°C [154.4°F]

[OK, so you'll get some case-fans]:

CPU temp @ 100 % load = [20 + [say] 1.5] + [0.81 x 53.1] = 64.5°C [148.1°F]

[get some silver h/s grease on that ChOrb]:

CPU Temp @ 100% load = 21.5 + [0.73 x 53.1] = 60.3°C [140.5°F]

HELP . . . . & this thing is meant to cool a Tbird at 1Ghz? . . . at a default 48.7-freakin'-Watts: ie with an Intake-air temp of 21.5°C that'll run out-of-the-box at a minimum of 60.9°C [141.7°F] under 100% load . . . . .

. . . . even if you turn the air-con down, for every one degree the room's cooler the CPU will be one degree cooler, no more. Playing Quake among the frosted carcasses in a butcher's cold-store at 4°C the 1Ghz TBird cooled by a "Chrome Orb" would still run - according to the manufacturer's own figures - at 43.5°C [110.2°F] minimum.

[Read the 08/04/2000 article by Joe Citarella at www.overclockers.com on his [appalled] findings when testing the "Chrome Orb" on a TBird overclocked to 800. For a Texan view, read the contrasting piece at www.hardocp.com of 08/15/2000, where a "Golden Orb" is said to hold a TBird overclocked to 1.1Ghz [60W] at 45°C].

Example 2: You read a web-review of a GOrb on a PGA366@550Mhz/2.0vcore; it shows the fansink to hold this CPU to 35°C in 22°C "ambient" at full load; this seems to compare real well to a test of an Alpha cooling the same CPU: have you just been saved $15+ here?

If both reviews came from the same otherwise reliable place under the same conditions, you can say it shows the GOrb does the same job on this CPU at this wattage. Strictly speaking, though, it's nonsense & cannot be compared to any other test/review:

Rj-a,°C/W calculated from this above review would be [35-22]÷[550/366]*21.7 = 0.40

Thermaltake modestly claim a Rj-a,°C/W of 0.81 for the GOrb with its own h/s grease; we have tested this fansink out-of-the-box on a 32.5W PGA to have a Rj-a,°C/W of 0.58;

CPU-case = ["ambient" [Intake-air] + [Rj-a,°C/W * Wattage]; so, in the real world, a GOrb on a Celeron366@550/2.0vcore at 22°C Intake-air will run at:

a) Thermaltake's figure:

CPU-case = 22 + [0.81*32.55] = 48.4°C [119°F]

b) Burningissues' figure:

CPU-case = 22 + [0.58*32.55] = 40.9°C [105.6°F]

. . . & I'm afraid that's it - that's the best the thing will do in those conditions; anywhere, at any time. If anything [due to our cautious test-procedure] this is an underestimate of temperature.

[As a general rule: [quality] manufacturers' °C/W figures are worst-case over a wide wattage-range & taken in a poorly ventilated test-chamber; you can expect to improve substantially upon their products' efficiency where using it in a well-ventilated case, lapped, & using top-quality h/s grease. Most web-reviews are astonishingly inaccurate & always err on the low side: a general exception are figures from www.overclockers.com]

Example 3: You want to buy the fastest socketed gaming-CPU that'll work through the summer without having anything in your computer too hot to touch - say 55°C [131F]. You're an avid gamer - play for at least two hours every day - living in Italy [Pisa, say] without air-conditioning: Your local store have 2 Alpha models in stock [hey, this is an example . . .] : 6035MUC & a FCPAL35MU - both with an advertised Rj-a,°C/W of 0.37 - & you have lots of case-fans.

Watts = [CPU Temp °C - Intake-air temp °C] ÷ Rj-a,°C/W [fansink + H/S grease]

You can expect the local temperature to be 38°C [100°F] for several days of the year; so the maximum Intake-air temp you must allow for is 39.5°C.

MaxWatts = [55 - 39.5] ÷ 0.37 = 41.9

The highest-wattage FC-PGA available as of September 2000 [1 Ghz cB0 stepping] draws 33W; the TBird 850 uses 40.2W, the 900 44.6W.

The highest-speed TBird many folk can use throughout the year air-cooled by a copper-insert Alpha is the 850, with the same cooling, you can use a PIII 1Ghz. It's your call which of these you reckon to be the faster gaming chip.

[Please note: none of the above is intended as: "Anti-AMD" - but it is clear that 'bargain' CPU's may well need non-bargain cooling if you put your system under anything like full loads. Also note that both example calculations were based on the manufacturers' own figures: we have not yet tested these exact models of fansinks & have not seen a reliable review]

Calibrating your system-monitoring utility:

The best of these utilities - like MBProbe & MotherBoardMonitor [both free] allow the option of resetting the temperature measurements from the various sensors on your mobo upwards or down. Sadly, one of the most interesting [CPUCool - cheap, with an inbuilt HLT cooler for W9x, a SOFT-FSB-like utility & with really good graphical logging], cannot yet be negatively calibrated.

To give an example of how necessary calibration is; I've had to reset the readouts from the onboard Winbond monitoring chip on a Soyo6BA+IV to read 7°C higher [for the CPU] & 5°C lower [for the "system" temp]. These kind of errors - 12°C in total if used to contrast case-air temp & CPU temp - are fairly typical of the sort of thing you can expect from an uncalibrated utility, which, sadly, are generally used as the source of the absolute temperature measurements in cooling widget web-reviews.  Even after calibration, you'll possibly find that your motherboard's sensing hardware is disturbingly non-linear, & will be passably accurate only over a very small range - typically 5C, from observations on several modern motherboards.

Cooling range:

Any fansink has a range of watts within which it will work to a tested °C/W efficiency -even a model which uses its tiny 9.5CFM fan with the efficiency of the Molex FC-PGA has a limit beyond which it simply cannot cool stably. The maximum wattage it will cool is determined by several factors: radiator-size [area of fins/pins/vanes swept by the fan]; internal conductivity [how quickly heat is wicked from base to radiator]; airflow management [how efficiently what flow there is scrubs heat from the radiator]; & lastly fan-size/fanflow. To some extent, simply putting a larger/higher-flow fan on an existing fansink can improve its °C/W - but this appears less true the more efficient the basic design of the fansink is - ie the better the above factors have together been optimized.

For a very brief look at this topic, see the section in our Radial Cooler Review where we examine the cooling density of three fansinks of very different design - ie how much cooling work can be done within a given volume occupied by the fansink.

Accuracy:

Despite all the proviso's made in this piece, I believe that temperatures & °C/W measures constructed from them using the techniques outlined here will be accurate well within 5% - enjoy . . . . . & please share your knowledge - there's not enough fact out there.

Notes on hardware used in this piece: From our test-data I believe the Thermaltake line to be at their efficient limit at around 40W, & the Molex FCF-PGA at just over 30W - the HP turbocooler 'orb' was designed for .35 process RISC processors running at up to 80W [with a large contact-area & very firm bolt-down clamping between CPU & cooler].

If heat cannot be conducted away quickly enough from the area of contact, through the body of the fansink to the fins/pins/vanes sitting in their local breeze, the fansink will literally choke on its own internal resistance. Playing vast quantities of air over poorly designed heatsinks made of insufficiently conductive materials is as fashionable as it is silly; our next cooling review will concentrate on case-cooling & practical solutions to getting an adequately cool environment for your system to work in.