Due to the number of issues covered and amount of reader feedback I've decided to create a permanent page covering the cooling and temperature issues of the G3 systems. The following comments and notes are from my 6/2/98 news page:

Apple G3 Temperature Variations: Last night I finally decided to try and determine why the temperatures reported by the G3's vary so much. One thing I noticed is that the clip that holds the CPU heatsink on seemed to be offcenter of the CPU, not putting a lot of pressure on the small raised center of the G3 CPU chip which is the only area of thermal transfer to the heatsink. (The CPU is not mounted in the exact center of the ZIF socket, but the Heatsink band is centered on it since the ZIF socket was originally designed for a direct CPU mate only, not for a card with other components like the G3 has.) I added a bit of thermal conductive paste (avail. at Radio Shack and other elec. parts stores) to the CPU/Heatsink contact area and reattached the heatsink (trying to center it best I could). Don't use too much compound - it's conductive and flows a bit so you don't want it spreading too far. It is only there to help provide a uniform contact area for heat transmission and to make up for a lack of perfect flatness in the heatsink contact surface.

Junction temperature is now down to the low 30's instead of 50C.

G3 Temp Reporting Utils: Larry Riddle comments on a post on the PowerPC G3 temperature reading issue:

" Some comments about the MPC-750 CPU junction temperature readings. On 5/18 you posted a comment from PowerLogix:

'The PowerPC 750 has a thermal register that we poll. The processor chip has a thermistor built into it, we are just reporting what is in the register. We aren't doing anything except looking up data in that register and displaying it; all other software does the same thing.'

This puzzled me, and I looked into how the Motorola junction temperature device works. If this is really all that the PowerLogix and other boards do, then I doubt that the indicated temperature readings are correct.

According to the Motorola MPC-750 User's Manual (Section 10.3, p. 361 of the pdf file downloadable from their web site) the Thermal Assist Unit (TAU) includes a thermal sensor, two registers (THRM1 and THRM2), a DAC, a comparator, and a batch of related components. The purpose is to provide a means of sensing when the temperature rises above a set point (loaded into one of the THRM registers). This event generates an interrupt which can be used by the CPU board to take some actions such as reducing the clock speed to protect the device. This is intended for laptop computers where the thermal dissipation is limited. "

Many CPUs now (Pent. and PII also), even for desktops have this feature to enable turning on fans for cooling, etc. Most Intel PII motherboards have 3 jacks to connect fans under control of the bios/cpu (they need it!).

Powerlogix reports on temperature tests between the old and new Powerbook G3/250:

" Enclosed are screen shots of PowerLogix SpeedMeter 1.1.1 running on an old style Powerbook G3 250 and a new style G3 250. Both machines were on for about 10 hours straight without sleeping. You can see the new G3 runs considerably cooler.

Old PB G3 250: 120 C, 248 F
New PB G3 250: 76 C, 168 F

BTW, PowerLogix G3 Cache Control v1.0.1P worked on the new PowerBook G3 to up the backside cache speed to 166 Mhz (up from 125.)

Both of these softwares are available at http://www.powerlogix.com/support/software.html .
[Update: Speedmeter and Cache Control have been replaced by PL's Cache Profiler-Mike] Robert Jagitsch"

PowerPC G3 Operating Temperature Spec: Several readers sent info on the G3 temperature specs:

" Mike,
Regarding the discussion about G3 junction temperatures, the PDF file on the G3 from Motorola's site (750_hs) defines the operating temperature range for the G3 to be between 0 and 128 deg C.
Howard Mckenzie"

Jeffrey Gay sent the following specs for the G3 CPU based on Motorola documentation:

• Storage Temperature Range: -55 to 150 degrees C [Not operating]
• Die Junction Temperature: 0 to 105 degrees C
• Thermal Sensor Range: 0 to 128 degrees C

Ryan Dumperth sent a link to a Motorola document stating recommended junction temperatures are 0-68 degrees C. Some reported temperatures exceed that figure. The document header says preliminary, so it's not a finalized document.

http://www.chips.ibm.com/products/ppc/documents/datasheets/750/750_hwspec_rev30.pdf

These junction temperatures ratings are much lower than I suspected based on other commercial electronics components I've seen at work.

More G3 CPU Heatsink Feedback: Mitch Brandwin writes he saw the same thing I did last night (today's earlier post):

" Hi Mike,
Checked my G3's cpu temp and found it was 80c. So I took a peek. I found the following 1) the clip that holds down the heatsink is offset. The pressure point is supposed to sit on top of the cpu, mine was opposite. The heat sink uses pressure only to draw heat from the cpu a weak clip or misinstalled clip can cause high cpu temp. I added heatsink compound to the cpu/heatsink junction to enhance heat draw from the cpu. My cpu temp now reads 28c.

A warm heatsink shouldn't equal a 50-60c cpu temp. (A 50-60c heatsink temp would burn your hand) The problem is not cooling the heatsink, its efficient removal of the heat from the cpu to the heatsink."

You can't directly relate heatsink temp to internal junction temperature - the 50C noted was junction temperature inside the chip. Thermal efficiency of the heatsink, size, conduction/contact area and ambient temperature and cooling all affect the temperature of the heatsink. However, all things being equal, airflow over the heatsink makes a big difference in cooling efficiency - we see it here at work all the time. Even a small amount of airflow helps. It can't make up for a poor conductive path at the heatsink/cpu junction but it will help. Any airflow over the heatsink increases it's efficiency dramatically.

The purpose of the heatsink is to radiate heat away from the CPU, if it's in a confined, stagnant enclosure it is not as effective as it could be. Usually in confined enclosures with poor airflow either a fan is added to the heatsink or the size of the heatsink (and fin shape) is increased to allow more surface area to dissipate heat (conduct it away from the CPU).

Just a warning to G3 owners thinking of opening up their case and playing around with the CPU/heatsink. There is always a potential for damage when you work inside the G3 case. Make sure you discharge yourself by touching the metal power supply case and wear a static discharge wrist strap if possible. Any work like this may void your Apple warranty if damage is done (not sure what Apple's policy on this is).

My G3/266MT ran for months with no problems at 52-56C and even the users reporting 80C were not having any problems so if in doubt follow the old " if it ain't broke - don't fix it " rule if you're not confident working inside the machine. Ultimately you are responsible for any modifications you do to your Mac.

PowerBook G3 CPU Temp Over Spec?: Just a note that Powerbook G3 10-hour temperatures reported earlier today exceed the recommended (in the case of the new Powerbook 250) and the maximum (in the case of the old Powerbook G3) CPU junction temperatures for the PowerPC 750 accord to the Motorola specs (see PDF file link posted earlier today).

I'm still not convinced the software utils are accurate in their reporting, since it would mean the Powerbooks are operating beyond the recommended temperature range for the CPU according to Motorola's preliminary PowerPC 750 specifications document.

Updates: Here's the latest feedback from readers:

Cooling Results:

" Hi Mike
Again, I want to thank you for such a wonderful sight. I downloaded the Powerlogix Speedmeter 1.1a2 and tested my cpu heat.
[Note: Speedmeter is no longer available - Powerlogix's cache profiler replaced it-Mike]

First reading was 80C
Second reading after installing 384 mb of RAM went down to 57-60 C.
third reading after pulling off the heatsink and putting some heat sink compound in place is now 44 C.

I suspect your page just added alot of life to my computer.
Thanks again
Gregory Santilli"

" Mike, At an 80F room temp., my G3 300 had been reporting about 48C (correspondingly lower at lower room temps).

My clip seems to be on backwards also. I removed my heatsink, put a thin film of thermal grease on it, put the clip back on it the SAME WAY IT WAS (apparently backwards). The temps are about the same.

Tonight, I'm going to try reversing the clip to change the pressure/pivot point, as you noted.

Thanks for BEST site on the Web.
Ed"

" Hello Mike.
Just to let you know that the heat of the CPU of my All-In-One G3 PPC 750 is praticly always of 88 to 89 celsius (+/- 190 Farenheit). This seems to be really high, but I'm not surprise since this model more compact... Any other clues about the All-In-One? With a 89c I beleive I'll don't try to overclocking my computer!
Au revoir,
SÈbastien"

" hi mike!
as i wrote in earlier mails, i'm driving my apple g3/266 @ 333MHz (since months, without any trouble). to avoid any problems, i've added a big heatsink with fan. now somebody started the discussion about the temperature-measurement-utilities. i was curious if the software works correctly. here at work, we use a FLUKE 51 High-Precision THERMOMETER for measure the CPU-temps. all i can say is: the software works correct! when the fan ontop the heatsink is running, the tool reads 40 degrees celsius, the fluke says 37. (i'm sure the little difference depends on the measuring-point, i had to measure between the clip that holds the heatsink and the heatsink). when the fan is off, the software says 72 degrees, the thermometer reads 71. (check the 2 screenshots) [Attachement had to be deleted due to my mailbox being too full to read - I'll ask him to resend.-Mike]

hey mike, you're doing a really GREAT job with your site!
greetings, michael "

" I read through this documentation a while back (I wanted to write a utility to report the temperature, but couldn't because the registers in question are only available in privileged mode, and I don't know how to fool the MacOS into letting a user program into privileged mode). While it is true that there is no direct temperature readout, it *is* possible to use the TAU to get one. There are two possible interrupts, one which is triggered by temperature rising above a set point (so that the OS can slow the CPU down), and the other which is triggered if the temperature drops below a second level (intended to let the OS speed the CPU back up again).

As it turns out, you don't have to turn on the interrupts these events generate. A program which wants to figure out the CPU temperature can do successive approximation by starting the upper and lower bounds at the extreme ranges of the sensor and stepping them inward, checking each time whether the flag bits which indicate high or low temperature have been set.

The built in feature for slowing down the CPU to save power is pretty slick. Instead of slowing down the clock, it's done by simply not issuing instructions all the time. Under normal operation the instruction fetch unit issues an instruction to the execution units every time it can. But it can be set to do this at a lower frequency. I don't remember how fine it can be tuned, but it's something like skip one out of 10 clocks, 1 out of 9, 1 out of 8, and so forth. Programs run slower, but the chip will consume less power since, like most PowerPCs, the 750 automatically shuts down inactive sections of itself to save power.

In conjunction with the temperature high and low sensing, this makes a pretty effective system for regulating CPU power without the downsides of the "processor cycling" software method Apple has used in the past. I don't think Apple makes use of this feature, however.

About airflow: I've noticed in the past that if you run the G3 desktop with its drive cage rotated up, the CPU will heat up a lot and will eventually crash from overheating (Apple actually warns against doing this in one of their tech documents). As slight as the airflow under the drive cage from the power supply fan may be, it's critical to maintaining a low CPU temperature; convection and radiant heat transfer are not enough. I think there's a very good chance that everybody with high temperature readouts has something like a ribbon cable blocking airflow across the CPU heatsink.

Like you, I took the heatsink off, added some thermal goop, and put it back on. I also added a 1" fan to blow through the heatsink tines, which now lets me run with the drive cage up if I want to.

About the off-center heatsink clip: it should apply pressure directly on top of the CPU die if installed in the correct orientation. The heatsink also has a correct orientation, since there is a black thermal conductive pad on the bottom which should sit directly on top of the CPU die. I have heard of some machines being supplied from the factory with the clip installed backwards.
Timothy A. Seufert"

" Heatsink clip & cpu are both off center, heatsink has a dry heat transfer pad (also offset). The pressure point on the clip should be towards the rear of the machine where it sits directly over the cpu. The heatsink is nearly impossible to mismount as it has only 1 groundstrap screw hole.

BTW, Cpu junction temp range is 28c (266mhz) to 32c (333mhz) after correction & addition of heatsink compound.
Mitch Brandwin"