1 GHz Dual Processor Quicksilver PowerMac G4 Speedup to 1.2 GHz


I recently purchased a Dual Processor Quicksilver G4 running at 1000 MHz, and just finished accelerating it to 1200 MHz. Instructions, observations, and benchmarks follow.
This is the base configuration machine with 256 Megs of CL3 memory. All benchmarks were run under Mac OS 9.2.2 with virtual memory off, running all the 9.2.2 extensions, but no others. The monitor used for testing was an Apple 20" Multiscan (Trinitron CRT) set to 640 x 480 at 60 Hz with 256 colors with the default GeForce4 MX card.

Pictures of the top and bottom of the 1000 MHz G4 daughtercard are shown below with the areas of interest highlighted in yellow.

The MPC7455 has an onchip PLL configured by a set of resistors on the daughtercard (boxed in yellow in the pictures) that controls the bus:proceessor ratio. The bus speed is 133 MHz, and the default bus:processor ratio is 7.5x to give a processor speed of 1000 MHz. I first increased the ratio to 8x for a 7 % speed increase to 1066 MHz. Having observed no problems, I then increased the ratio to 9x for a 20 % speed increase to 1200 MHz. I have not yet tried the 10x ratio with a potential 33 % speed increase to 1330 MHz.

There are two sets of these resistors; each set controls one processor. The resistors are CFG_EXT and CFG_PLL[0:3]. The following table outlines the presence (0) and absence (1) of the 1k Ohm (102) resistors on the pads on the daughtercard for the relevant configurations.

Bus:Processor Ratio

Speed (MHz)

PLL_EXT

PLL_CFG[0]

PLL_CFG[1]

PLL_CFG[2]

PLL_CFG[3]

7.5x (norm)

1000

0

0

0

0

1

8x

1066

0

1

1

0

0

9x

1200

1

0

1

1

1

10x

1330

1

1

0

1

0

There are no markings on the motherboard, so the figure below of the top of the daughtercard contains the necessary labels to the right of the pads. This is an image of the board after it was configured for the 1200 MHz configuration.

The figure below of the bottom of the daughtercard contains the necessary labels to the right of the pads. This is an image of the board after it was configured for the 1200 MHz configuration.


1200MHz modification steps:

Shutdown the computer
Leave the computer plugged in; this leaves the case grounded
Open the case
Remove the two Phillips screws from the back of the case to allow the fan to be removed
Set the fan aside, but leave it plugged in (so you won't forget to plug it in later)
Unsnap the two heatsink springs (four points)
Set aside the heatsink springs
Remove and set aside the heatsink
Remove the four Phillips screws from the daughtercard
Remove the daughtercard
Remove the 1k Ohm resistors from pads PLL_EXT, PLL_CFG[1], and PLL_CFG[2] with a soldering iron (I use two; one on each side) from the top and bottom of the daughtercard
Replace the daughtercard
Replace the four Phillips screws in the daughtercard
Replace the heatsink
Replace the heatsink springs, and snap them in place (four points)
Replace the fan
Replace the two Phillips screws in the back of the case into the fan
Close the case
Start the computer, and be prepared to kill the power if it crashes on bootup

Take a look at Apple System Profiler; it should show 1200 MHz under the System Profile in the Hardware Overview section as shown below for both OS9 and OSX respectively.


Benchmarks

OSX is new to me, so I don't know of many native benchmarking programs yet. Here are the benchmarks I am familiar with. If you have any benchmarks you would like me to run, drop me a line (schrier@mac.com).

MacBench 5 clearly shows a 20 % speedup with the modification to 1200 MHz.

Cinebench 2000 also shows a 20 % increase.

I received a copy of TestbenchX after performing the modification, so I only have data for my machine on the 1.2 GHz configuration, but I also have 1 GHz info from a fellow Mac user.

I received a copy of POV-Ray 3.1g.r2 after performing the modification, so I only have data on my machine for the 1.2 GHz configuration. Apparently the OS 9 version only uses a single processor. A fellow Mac user is looking into porting the linux application so we can give it a spin.

The Mac PPC RC5 client running under MacOS 9.2.2 crunches 25.4 million nodes/second after the 1.2 GHz modification.

The Mac PPC/OS X RC5 client running under MacOS X 10.1.3 (SQ45) crunches 27.3 million nodes/second after the 1.2 GHz modification.

The 20 Photoshop 6 Actions on the Bare Feats web page took 40 seconds under OS 9.2.2 and 38 seconds under OSX. Conditions: 1024 x 768, 85 Hz, millions of colors, 200 megs of memory allocated to Photoshop, VM off, FS off, tested on a fresh restart.


Notes:

The resistors can be removed very quickly with two soldering irons; it probably takes me 1-2 seconds to remove them. I'm confident this will not overheat the daughtercard. If the resistor sticks to one of the soldering irons it may overheat and be damaged. If you damage a resistor, or want a spare one before you start, you can Join the Resistance! by sending a Self Addressed Stamped Envelope to Output Enablers, and we will send you a 1k Ohm (102) surface mount resistor for free.
So far I've had no problems. I've had the machine on for a max of about 10 hours straight, and no crashes yet in either OS.
I have not yet tried the 1330 MHz configuration.
I have not yet explored bus frequency or core voltage modifications.
I will likely try some other thermally conductive pastes to increase thermal conductivity.
I have not yet monitored the increase in processor temperatures, but will soon hook up a thermocouple to the heatsink above the processors.
If you've slowed down your fans to make your machine quiet (How I Reduced Fan Noise in my G4 Tower, Info on Quieter fans for G4 Towers, How I Reduced Fan Noise in a Quicksilver G4 Tower, How I Reduced Fan Noise in a Quicksilver G4 Tower: Part 2, and More Examples of Quicksilver G4 Fan Noise Reduction Mods on the Accelerate Your Mac! Web Page), keep a close eye on things; you may want to undo it for now.

Other Attempts

So far I have not heard from many people who tried the mod, but two people failed at 1200 MHz, but are now running at 1066 MHz without any problems.


Marc Schrier
(schrier@mac.com)