FANLESS COOLING RIB EXPERIMENT & MEASUREMENTs

This document contains information related to the attempted removal of the CPU cooling-fan, and the testing of a copperbased custom made fan-less cooler.

[ Index ]

Introduction:

Add a picture of the rib. Some data. Weight dimensions etc.
Add a picture of the original standard rib, element.
Digital Thermometer used : Oregon Scientific (in/door outdoor type of quality)
CPU running at 233@66 MHz, Vcore 2.5 (0.3 below specs.)
Software cooler used throughout for idle operation : Waterfall.
Load testing s.w. : CPU Stability tester v.6.0

Questions :
Thermal qualities of air. Insulation. Heat transmission qualities. Humidity factors. Look up info.

DAY 1 - Initial measurements

Measurement of the original fan based system.

To get a reference point, measurements where done of the temperature of the initial CPU-cooling system. This was based on a standard aluminum rib element with a cool-master fan. (ref. FAN[2] in equipment used).

Temperature probe placed on the side of the element, out of the air-stream, with some sticky stuff.

Note ! CPU Fan running at 5 V. No thermal regulation. Static rmp.

[1] System idle.

Open-cabinet, ambient temperature of 20.0 C, rib temperature 22.0 C slowly declining towards ambient room temperature. Conclusion. Fan manage to reduce rib temperature to ambient temperature. This temperature will be determined by the room or cabinet temperature at the time.

[2] System load testing.

Same condition as in measurement [1] above. Monitor peak operating temperature after 10 minutes.

* Measurement deviations, possibly due to that test program tests different parts of the CPU. And that these different tests generates more or less heat.

oo - It is interesting to note that the rib temperature here is approx. 7 degrees above ambient temperature. The fan doesn't manage to bring the level down to ambient temperature. This may be improved by increasing the fan speed, and amount of air that is blown over the rib. This indicates the cooling effect of the fan in relation to the same rib element without fan. See next  measurement [3] below.

[3] System idle operation temperature with fan disabled. Same setup as above.

[4] System load testing with fan disabled to find failure temperature.

Same setup as in above measurements.

System failed : 17 C above ambient room temperature. Black screen. CPU running on 2.5 Vcore, 0.3 volts below spec. ( Prolonged operation time possible with increase in voltage. Processor is actually stable in this state on 2.8 V, temperature unknown but max. func. temp. is 70 C.)

Fitting and testing the custom made copperbased rib.

Major fitting problems. To soft and flexible. Problems with cooling plate thickness vs. clip pressure.

Booting up system for the first time with element mounted on processor. Thermal sensor placed on the side of the rib, in a similar location as on the original aluminum rib. Temperatures range from 22.7 C to 23.7 C during boot up.

[1] Finding stable idle temperature. Open cabinet.

Approximate stable idle temperature of  the element is 22.8 C. Possibly lower.

[2] Load testing. Open cabinet.

Ran stable for 45 minutes with open cabinet.

oo - The 18 C difference between ambient and rib temperature, indicates that there needs to be a certain magnitude between them. Even on such a large element. If the relationships are linear, and characteristics can be transferred an internal rib temperature of approx. 48 degrees can be expected with cabtemperature of 30 degrees.

[3] Cooling down - Back to idle. Heat-loss-rate of element.

By measuring the heat-loss rate, heat-dissipation qualities can be found for the element. Question : How many watts are the element able to dissipate at a given temperature difference.

[4] Testing with cabinet cover on.

Bootup. 23.0 C @ 19.3 C Ambient temperature.

Black screen. System failed after approximate 20 minutes of load testing.

DAY-2 - IMPROVED RIB FITTING & ADDED COOLING PASTE

Note ! serious deviations and unexpected results from yesterday.

After tweaking the rib-fitting smoothing surface by removing excess solder on top, and inserting a small strip of copper underneath the clip. A better fitting of the Rib was achieve, with a firmer and more uniform pressure from rib onto CPU core. Also added cooling paste and wiggled the thing around to get a good connection.

Open cabinet measurement : (Fanless, Vcore 2.5 volts )

Stable Idle temperature

[08:20] found to be 23.1 C @  19.8 C
oo Temperature approx. 1 degree C higher than yesterdays measurement. This may be due to relocation of temperature sensor. Or it could indicate a better heat transfer between CPU and rib.

Load testing

System fails at a higher temperature than yesterday. Why ? Compare test time.Possible causes.

- Less uniform heat distribution in element after re-fitting ?
- Measurement differences after relocation temperature sensor ?
- Variations in natural draft in the room ?.
- Variation in air humidity and cooling effect ?

Note ! The temperature sensor where relocated and Rib fitting where wiggled again to improve contact. Maybe to much cooling past o.l. The initial temperature was 28 C [09:58] and fail temperature was measured to 43.0 C at [10:29] system failed somewhere after [10:17].

Closed cabinet measurement : (Fanless, Vcore 2.5 volts )

Stable Idle temperature.

Today's stable idle CPU rib temperature is : 29.5 C @ 20.7 Ambient room temperature.

Load testing.

Error message from windows indicated that test program had failed : Program have performed an illegal operation and will be terminated. Shortly after system performed an soft-reset.

On bootup. Soundcard mixer program failed. "Illegal Page Error" .in module CLS_COM.DLL.

Rib temperature at 02:05 was 42.5 C @ 21.1 C ambient room temperature, falling at a rate of 1 C pr. minute on CPU idle.

Some kind of summary so far.

From the results sofare, stable fanless operation works fine. But not stable on full load.

Possible remedies. Increase Vcore voltage to specs. (2.8 V). System has been tested stable on this voltage for a long time without fan before, using the standard cooling rib. However the rib became so hot that it wasn't considered healthy for the processor. But maybe. A test on this Vcore level would be interesting. Acceptable peak temperature 50 C. Other alternatives are to improve the air-circulation within the cabinet.But when the open cabinet solution failed, this would have to include an active fan, which brings us almost back to scratch.

Other observations : A fanless cooling system is much more susceptible to factors such as draft and air-humidity than a fan based system. A fan-based system takes the temperature of the cooling rib down close to room/cab temperature on idle. A fan-less system is dependant on having a certain (and variable) temperature difference between ambient temperature and rib temperature. So far this diff. seems to be approx. 16 deg. C for this element. Which exceeds the stable operating temperature range on the under-voltaged CPU (-0.3 from spec.).

[ prime this drooling later ]

OO - Adding mass to the thermal element without increasing the surface area, only defers the time of failure. If you can buy enough time it's fine. But to achieve a stable solution the surface area and heat dissipation must be increased.

Moving on : different states to test.

1. Chew into the sour apple, and increase Vcore to 2.8 to find the stable rib temperature here.Drawbacks are that this will increase the heat generated inside the cabinet, both from CPU and MB voltage regulator. The power-drawn from the PSU will increase, and PSU fan will work heavier ... making more noise.
But there is something to be learnt about the heat dissipating properties of the element from this. Maybe I'll even find that temperature is on the border of beeing stable, and that some minor improvement in surface area will make the element stable at the 2.5 Vcore voltage.

2.  Improve cabinett ventilation by adding an internal fan with purpose to improve air circulation. Drawback is that more noise will be added, but it will only need to be active at heavy loads and could be acceptable. Would be interresting to see the effect of this.

3. Learn something about the thermal properties of air. standing still. and moving. principles of fan/air cooling.

: From what I know : The thermal insulating properties of air increases with the dryness of the air an vice versa. As mentioned before air humidity is an important factor. A certain difference in ambient temperature and element is required to creating natural ventilation. Similar to the air circulation generated by a bon-fire or a stove. This means that the higher the internal cabinet temperature get, the higher the CPU temp. will have to be inorder to create the same draft/ventilation.

4. Dig up some more detailed info about the CPU. power-consumption, heat-generation, working temperature range. life-time expectancy.

: Pentium 233 MMX, CPU External I/O 3.3 Volts, CPU Internal voltage (Vcore) 2.8 Volts. Maximum power consumption 18 Watts (Guess). Average powerconsumption (Guess) : 8.8 Watt. Operating temperature range : [0 .. 70] C.

Day 3

Executing point 1 from above. Vcore increased from 2.52 V to 2.8 V. PSU Fan appears slightly more noisy. But definetively more noisy.

Measuring over time to find stable idle temperature, closed cabinett.

Shortly after boot-up rib temperature 32.5 @ 22.1 C Ambient room temperature. Time 18:35. Waiting. ....

Conclusion : Stable idle temperature at 2.8 Vcore, is 30.2 C. Normal operating temerpature is around 34 C.

Starting load testing : find peak temperature.

Concludes that system is stable at max. load on 51C @ 22C Ambinent temperature after 1 hour and 20 minutes of tesing. Also Hard disk temperature stays well within operarating range of [0..50]C ($2B equals 37.3C). Noise level acceptable.

This conclusion hangs in thin air, as there is nothing to compare it with after the Vcore voltage was changed. Does this fancy custom made copper cooler have any effect beyond it's cool ? Comparative measurement with standard element is needed.

Anyway. Results are fine. 50C temperature on element (and assumed to be operating temperature of CPU core) is fine.

oo - Since the system now is stable. The element is able to dissipate an effect of 18 Watts to the surrounding air. Assuming an cabinett temperature of 30C, and ignoring other factors such as air humidity and PSU ventialtion, the element is able to do this at a temperature difference of 20C.

oo - The 2.5 Vcore voltage was prefrable to 2.8. PSU more noisy on max. load, and also when system is idle. A thermostat regulated CPU cooler fan of decent quality would be a better solution. But but. Wait for the results from the modified version of PSU(2).

oo - 95 % of peak rib temperature reached within 15 minutes of testing. Indication for future testing.

oo - The software cooler have an dramatic effect on cpu-temperatur, when compared to a fan-based.system where the difference isn't that noitcable.

oo - Heat loss has an exponential relationship with (delta) temperature (rib vs amb.)..

Update : Measurments of standard element w/o fan.

Some missing data on the standard CPU cooling element.[ Late January 2001 ]

Idle temperture, measured from the side/top was 30.3 C @ 22.7 C after 40 minutes of sw cooling.

Normal temperature, measured from the top of the element in the airflow, without swc, running windows idle, was measured to 62.9 C @ 23.5 C. Measurment was aborted, as the the temperature was still climbing steeply at this point. Rate of change was 0.8 C/min at 60 C. Estimate that internal temperauture already have exceeded 70 C, as there is some latency between the measured airstream temperature and the internal CPU temperature.

No Load (peak) measurments where possible as no stable state where reached within CPU specs.

An alternative, and closer to CPU temp. peak measurment where also done of the custom element. Performed inside one of the cooling tubes that where sealed off. Temperature after 1:30 (hh:mm) was 57.9 C @ 23.0 C. Approximately 3 degrees higher that the previous temperature measured from the side (when adjusted for differences in ambient temperature).

The custom element has a much more effective cooling effect than the standard element at high-temperatures. But at idle load the difference is negible.

The same test above could be done for the standard element of system(2), but as it looks; only of the idle temperature.

All mesurments done in a closed cabinett presumably ?

[ Index ]

Document is yet to be concluded. February 2001.