Accueil

Dossiers

Fichiers

Liens

Bidouill

Divers

Envie de me remercier,
m'encourager?

RSS

sterpin.net en français Bloc transparent sterpin.net in english

Macintosh powerbook Pismo processor card PowerLogix G3/900

Tête Philippe Helman
Philippe Helman

WARNING : everything below must not be done if you hesitate, just a little!!!

Nobody will be responsible if you try it and damage something!!!

Featuring :

Powerbook G3/400 « Firewire » processor card upgraded to G3/900 (750fx) by PowerLogix.
512 Mb RAM (2x256 PC100 CL2).
Running 10.3.2.

Note : The original card had L2 cache problem. Except that, it worked.
As the G3/900 does not use original the 1Mb cache but its own 512Kb on-chip, an upgrade is allowed.
It made indeed no problem.

The original double heatsink is not used anymore by PowerLogix because the G3/900 installed processor final height is 1,7 mm above original one.
Therefore, PowerLogix provides a 1,3 mm thick single copper plate of 35x65mm size to be glued with thermal paste both on the processor AND under the metal plate covering the processor card.
As I was not fully convinced of the efficiency, I reworked the original heatsink to use it again.
750fx processors have no built-in temperature circuit so no software (TherminDock, Thermometer, Thermograph) can display temperature anymore.
To check the improvements, I made some temperature measurement with a K-type «film» thermocouple inserted between the metal plate and the heatsink.
I did NOT inserted the thermocouple between the processor and the heatsink, because the thermocouple is made of metal and risks of touching the electronics are too big.
I made a comparison between PowerLogix and modified Apple heatsinks, and I also measured with an original 500MHz processor card.
The read values are for comparison only and do not represent the real processor temperature.

The following pictures and legends show the steps of the tests.

1)

Before any test, I measured the height difference.
Here is an original G3/500.

2)

The G3/900 is obviously higher... by 1,7mm.

3)

As received, with the plastic shield to avoid metal contact with the heatsink.

4)

Plastic shield removed...

5)

Shows intermediate printed circuit board. It is amazingly well done !!!

6)

The copper heatsink and the thermal paste provided.

7)

To glue the copper plate on the metal plate.

8)

And the small drop on the processor.
Later, I put Artic silver 3 thermal paste, but no specific improvements were noticed.

9)

Was tested under 10.2.8 and 256 RAM at the beginning.

10)

Carefully remove the original patch form Apple heatsink

11)

And remove the glue.

12)

After unbending the heat-conductive pipe by around 45 degrees, I carefully removed the processor-side plate.

13)

It uses soft-soldering, so quite easy to «unweld».

14)

Fine-tuning the path of the pipe, going above the card but beside the processor, and with the top of the pipe leveling the top of the processor

15)

The 1mm thick brass plate was carefully positioned and welded on the other side of the pipe, using a special fluid to allow a best contact between pipe and plate

16)

It is originally designed for copper welding in central heating piping works, using Ag-Sn (silver-tin) alloy. It contains an acid, so be careful!!!
I used another Sn-based alloy (Sn-Pb) because its melting temperature is far below Ag-Sn.

17)

After cleaning the plate

18)

Thanks to the soldering fluid, the alloy perfectly fills the space between pipe and plate.

19)

Installed with part of the thermal patch.
Fitting and tightening the metal plate will press the plate on the processor.

20)

This is the K-type (Chromel-Alumel) industrial thermocouple.
It has a specific film-designed sensor and is as fragile as expensive !!!

21)
 

The Keyboard was not fully reinstalled for tests.
I did not want to remove the airport card, even if it was then possible to let the wire go through the PCMCIA slot. 

22)

It shows 62 degrees Celsius.
This is a stabilized temperature after a 1 hour work (word processing).
The temperature at the second cooler (the metal visible one on picture before) is around 50 degrees Celsius.

23)

Running several (10x) Xbench raised temperature to 67-68 deg C.

24)

25)

At that temperature, I got a freeze twice!!!
The cooling fan never started.

26)

It shows what it shows...

27)

The temperature sensor installed on the PowerLogix copper plate.

28)

Now with the PowerLogix copper plate.
Temperature raises faster than with my heatsink.

29)

Word processing keeps temperature at 66-67 deg C.

30)

Only one Xbench froze the Pismo.
After restart, a Dropstuff test raised the temperature up to 70 degrees C, then froze again.
Note that with the 500MHz, TherminDock showed once up to 72 degrees C, and the Pismo started to slow down... without having the fan started.

31)

Well, sensor position was not installed between heat exchanger and thermal pad, but above thermal pad, because I did not want to remove this pad. Note that this heat exchanger is made of a kind of black plastic, still with the heat pipe (in copper). This is part of a revision 2 Pismo model from september 2000 (including motherboard revision for Firewire improvements, and a 20GB hard disk instead of the 12Gb on the 500MHz model, from 6 to 10 Gb on 400 ones).
TherminDock was running with the original 500.
And obviously, it was hard to raise processor temperature.

32)

After 15-20x Xbench (CPU and Quartz only, to run processor at 100% load as displayed by Menumeters), TherminDock showed maximum 72 deg C while the K-type sensor showed a heat exchanger temperature of 58 deg C. TherminDock shows a very fast temperature drop when Xbench test stops (from 64 to 52 in 5-10 seconds).
TherminDock showed very briefly 76 deg C but Xbench then quit. I was not able to reproduce it.
And the fan did not go off. I never found the Apple official start/stop temperatures for the fan. Neither on the Web, nor in the service source, nor in the developer pdf. If you have a serious source Maybe ambient air temperature is in cause, at 20 deg C it is fresh for the Pismo.
I guess during summer season, with an air at 30 deg C in the room, the original 500 will freeze. The underside of the Pismo was less hot than with the 900.

You will find in this file the comparative results of various tests in various configurations.

This will not apply to you, you should only compare the time gap for the same task, for example.

Original machine :

Pismo 2x256RAM 2-2-2, thousands of colors, OS 10.3.2, on AC adapter, combo drive (24x CD read) inserted, no battery
Network : all active : Modem ready, IrdA OFF, Ethernet on fixed IP but not connected to hub, Airport card present but OFF
HD : 20Gb/2Mb cache/4200rpm (new), or 40Gb/8Mb cache/5400rpm (new), same files on both

Comparative machine :

G4/533 3x256RAM 3-2-2, millions of colors, OS 10.3.2, 2 optical drives (40x CD read), ATI8500
Network : all active : Modem ready, Ethernet on fixed IP and connected to hub
HD : 1x40Gb 7200rpm + 1x60Gb 7200rpm

Results in seconds

256 RAM
512 RAM
TOWER
REMARKS

Test

500/20
500/40
900/40
900/40
G4/533

Boot

80
57
60
60
64

from pushing button to ready to work

System profiler (speed/cache)

500/1
500/1

900/512

900/512
533/1

Copy 1 file

duplicate 1 file

5
4

4

37Mb .dmg (10.3.2 update)

duplicate 1 file

22
19
28
27
16

189Mb .toast

CDtoHD

120
124

100

280 Mb .mov (Panther_demo)

extFWHDtoHD/return

22/46
22/45

12/13

"

Copy x files

duplicate x files

5
4

4
4

53 files, 36Mb (.pps, .jpg, .mpg, .wmv, …)

duplicate y files

187
177
158
155
145

>2800 files, 580Mb (user library folder)

CDtoHD

166
148

123

78 files, 23Mb (choix redaction folder)

extFWHDtoHD/return

11/13
11/13

8/9

"

Stuff 1 file

std 802, dmg to sitx

164
165

113
120

37Mb .dmg, no compression result

Stuff x files

std 802, folder to sitx

183
182
125
128
139

566 files, 68Mb > 29Mb

Unstuff 1 file

std 802, sitx to dmg

106
105

70
88

37Mb .dmg

Unstuff x files

std 802, sitx to folder

127
124
87
85
104

566 files, 29Mb > 68Mb

Unstuff x files

std 703, sitx to folder

124
127
-
-
102

566 files, 29Mb > 68Mb

Launch soft only

Netscape 7.1

13/10
14/10
12/6
11/6
14/7

1st/2nd lauch

Excel X.1.5

6/3
6/3
-
4/2
5/3

1st/2nd lauch

Photoshop 701

21/13
21/13
16/10
15/10
15/11

1st/2nd lauch

PS701

67Mb file open

32/26
23/19
-
20/17
22/18

1st/2nd lauch

67Mb filter

32
32
-
23
15

gaussian blur 25 pixels

iTunes 4.2 encode

AIFF to MP3, 192k

3.3x
4.2x
5.8x
8.0x
8.0x

42Mb AIFF file on HD (4'12")

Tests and compatibility

Xbench 113

CPU speed/cache size

500/1
500/1
900/512
900/512
534/1

cache speed

200
200
360
360
267

CPU

34,2
34,2
61,3
61,1
63,7

disk

53,5
57,7
59,7
58,8
90,6

graphic Quartz

61,1
61,2
89,4
89,7
102,5

graphic Open GL

67,9
68,2
111,5
113,3
65,2

CPU Director 1.3

CPU speed/cache size

900/512
900/512
533/1

cache speed

900
900
267

iChatAV support

DV FW camcorder

-
No (1)
-
No (1)
Yes

(1) Yes with APE 1.4.1 and iChatUSBCam 1.1.2

iSight

-
No (1)
-
Yes
Yes

IMPORTANT UPDATE April 2004

I installed the 900 MHz in another Pismo, running 10.3.3.
And, after 1 hour working hard in a hot car (with Airport connected to a hotspot), I heard the fan coming ON.
It ran for 20-30 minutes.
This was reproduced several days.
I can't say it is the change of Pismo, or 10.3.3 (all tests were made under 10.3.2).
I might suspect 10.3.3 because I read Apple changed the fan start temperature level in the software, after people complaining about heat feeling on PowerBook Aluminium G4 12" mainly

Conclusion :

I would not trust the PowerLogix heat exchanger under heavy processor load!!!
Even word processing raised the temperature, no matter how I carefully cleaned and put new thermal paste at each installation.
My modified heat exchanger stabilized the temperature but it is still not reliable under continuous load.
Nevertheless, I have NEVER met a freeze with the mod when I normally work (processor under various and varying loads) : playing with Xbench or Dropstuff is not a matter of hours
Furthermore, using Artic Silver 3 thermal paste slowed a little the temperature raising speed under heavy load : heat transfer is faster to the heat exchanger.
I made some test by downclocking the processor to 500 MHz thanks to CPUDirector v1.3.
Continuing text editing showed the temperature dropping to 57-58 degrees C.
Xbench was not able to dangerously raise the temperature.
But what, I have a 900MHz, I want to fully use it !!!
...
SO
...
WHAT???

Well, too hot is too hot, and temperature margin is too small for me.
I asked one of my colleagues, whose hobby is electronics, to help me.
The cooling system specs sheet included :

  1. use original fan (5 Volts, 0,5 Watt) but not motherboard circuit
  2. control circuit must be OFF when Pismo is OFF
  3. control circuit must be 5 Volts powered and as low consumption as possible
  4. triggering must include an hysteresis (fan starts at X deg C and stops at Y=X-Z deg)
  5. Start (X) and stop (Y) temperature must be adjustable, at least on prototype
  6. Temperature sensor must be as closed as possible to the processor
  7. The accuracy of the sensor should be +/- 1 degree C
  8. Printboard must be maximum 5 mm thick, components included, anb will be located above the hard disk
  9. 5 Volts may be taken from hard disk power supply
  10. circuit must be protected by a fuse (or a similar electronic system with automatic reset)

We rapidly forgot about Peltier effect cooler, their power consumption being far too high.
Surprisingly (for non-electronic fan), he needed 5 minutes to design the circuit !!!!
Parts cost was estimated below 50 Euros.
Of course, handmade printboard is time-consuming, and if you have to pay for it, forget it!!!!
On the other hand, if you are careful and have the appropriate tools, go for it!!!

As time is missing, the above temperature control will be part of a future release.

Back Haut