PowerMac G3/233 and G3/266 Speedup


Well, I got a PowerMac G3/233 a while ago, so I've been playing :)

The first info out was on Takashi Imai's PMG3 Page The info there maps out most of the possible configurations for changing the bus, pci, and processor speeds. Unfortunately that is about as far as it goes; there is no mention on which configurations might actually work.

I've tried several configurations now, and it looks like I'm maxed out with a 66MHz bus (default), the 512K L2 backside cache running at 150MHz (half the processor speed(default)), and the G3 running at 300MHz. At many other faster configurations I could not get my machine to boot or monitor to sync. I've even tried a standard 14" monitor with some of the faster configurations and still it's a no go, even when I try to run the cache at 2.5:1 or 3:1, so it does not appear to be a cache limitation hindering 333MHz since the backside cache would be running at 120 and 100MHz respectively under these conditions.

So following the table on Takashi Imai's site you can increase your processor speed, and then there is one more thing to try. PowerLogix (cache@powerlogix.com) has modified a program they wrote for their G3 Accelerator cards so it will work with the Apple G3 machines. Their G3 Cache Control 1.5.1 allows you to change the frequency ratio your L2 backside cache runs at compared with the frequency of the processor. The PowerMac G3/233 and G3/266 backside caches both run at half the processor rate, so 117 and 133MHz respectively. Instead of a 2:1 ratio (processor:cache) you can set the ratio to 1.5:1.

To accelerate these machines to 300MHz you will need to remove the sticker that says "VOID Warranty if seal is broken". I feel stupid saying it, but in this legal day and age I ought to remind you that this means your warranty is void so if something else, totally unrelated breaks within the Apple warranty time period, then Apple is not responsible for fixing it. If you just want to go to 266MHz from 233MHz you can install a jumper without removing the seal and jumper block. Although this too probably voids the warranty it is less obvious.

So the first thing is to see if you can get your machine up to 300MHz by moving around the jumpers so they are at positions 1,5,6,and 7 (jumper 1 is where the 1 is printed on the board). If things are going well, then you can try the G3 Cache Control 1.5.1 and see if your L2 backside cache can run at 200MHz instead of the 150MHz it started at; I could not get mine to work with this 1.5:1 configuration; it was too fast. If it does not work just restart the machine holding the shift key to disable the extensions and remove the G3 Cache Control.

If 300MHz is too fast, then drop to 266MHz by moving around the jumpers so they are at positions 2,4,5,6, and 7 (jumper 1 is where the 1 is printed on the board). If you started with a 266MHz machine this will not be a problem, but if you started with a 233MHz machine you need another jumper, so just use a little piece of wire. Or ignore the jumper block altogether and use individual little jumpers like you get with new harddrives. The odds are super high that 266MHz will work for the 233MHz machines, so now try the G3 Cache Control 1.5.1 and see if your L2 backside cache can run at 178MHz instead of the 133MHz it started at; although I could run my machine in this configuration like this for a few minutes, that was it and then it would crash. If it does not work just restart the machine holding the shift key to disable the extensions and remove the G3 Cache Control.

And if 266MHz does not work (or if you don't want to clock chip your machine), then you should at least try the G3 Cache Control 1.5.1and see if your L2 backside cache can run at 156MHz instead of the 117MHz it started at. If for some reason it does not work, just restart the machine holding the shift key to disable the extensions and remove the G3 Cache Control.

If you don't have much luck with the G3 Cache Control 1.5.1, you might give the G3 Control Strip a try. I've heard that G3 Control Strip loads later in the boot process than does G3 Cache Control 1.5.1, and this may avoid some crashes.


Here are some MacBench 4.0 files:

PowerMac G3@67/117/233 (Base 233MHz machine)* PowerMac G3@67/156/233 (233MHz machine, boosted backside cache)* PowerMac G3@67/134/267 (Base 266MHz machine)* PowerMac G3@67/150/300 (300MHz machine)*

Next you will want to install the LibMoto shared library extension from Motorola. These libraries will dramatically enhance your FPU performance for programs that have not already incorporated them.

So here's another MacBench 4.0 file with the LibMoto extension installed:

PowerMac G3@67/150/300 (300MHz machine)*

*Here's the background for these MacBench 4.0 files:

HD: Quantum Fireball ST4300A reformatted (with a bit of trouble) with FWB HDT 2.0.6 into multiple partitions.
Virtual Memory: Off
Disk Cache: 1024K
Extensions Manager: PowerMac G3 Base
Video: 16 bit, 1152 x 870, 75Hz
Memory: 32Megs
Video Memory: 2Megs


And now that I have played a while and optimized things, here's what I've optimized, and some new benchmarks with these settings:

HD: Quantum Fireball ST4300A reformatted with FWB HDT 2.5 into multiple partitions.
FWB Driver Options: Enable write cache (that's it; the other defaults are optimized)
Virtual Memory: Off
Disk Cache: 6144K
Extensions Manager: PowerMac G3 Base and the LibMoto extension
Memory: 96Megs
Video Memory: 6Megs

The video benchmarks depend strongly on the video settings. Here are some at 1152 x 870, 75Hz:

Because of this strong dependence on video settings, if you wish to compare a machine to a standard MacBench 4.0 file you need to use 8 bit video in 640 x 480, 67Hz, so use this file as an optimized configuration for a G3@300MHz:


Now that I've played a bit more, I've learned a bit more.

Jumpers 7,8, and 9 set the bus:

Bus (MHz)

J7

J8

J9

60.25

 

X

 

66.82 (norm)

X

 
 

70

 

X

X

75.15

X

X

 

83.3

 
 
 

Jumpers 1,2, 3, and 4 set the processor multiplier (with respect to the bus):

Ratio

J1

J2

J3

J4

2

X

 

X

X

2.5

X

 
 

X

3

 

X

X

X

3.5 (233 norm)

 
 
 

X

4 (266 norm)

 

X

 

X

4.5

X

 
 
 

5

 

X

 
 

5.5

 

X

X

 

6

 
 

X

 

6.5

X

 

X

 

7

X

X

 

X

Jumper 5 seems to set the PCI multiplier (with respect to the bus) (I thought this was the case, but it does not seem to be true, so I'm lost now...)

Ratio

J5

.4

 

.5

X

And this needs to be set at 30, 33, or 35, so

Bus (MHz)

Ratio

PCI (MHz)

60.25

.5

30

66.82

.5

33

70

.5

35

75.15

.4

30

83.3

.4

33.2?

Jumper 6 is always installed, but I'm not sure it is needed in all cases (I don't know what it does)

So the important ones are

Bus (MHz)

3x Ratio

3.5x Ratio

4x Ratio

4.5x Ratio

5x Ratio

60.25

 
 
 
 

300MHz, slow bus

66.82

 

233MHz norm

266MHz norm

300MHz

 

70

 
 
 

Crash on boot

 

75.15

 
 

300MHz

 
 

83.3

Unstable

Unstable

 
 
 

I'm not exactly sure why, but the overall benchmarks for the 66.82MHz bus seem to be faster than the 75.15MHz bus benchmarks even though one might expect the faster bus configuration to be faster when the two processors are both running at 300MHz. I don't think it is because of the PCI bus (33MHz for the 66.82MHz and 30MHz for the 75.15MHz), but maybe it has to do with the video or maybe an additional wait state is needed for the memory. For now I'll stick with the 66/150/300 configuration.

Here are the two new benchmarks. The PowerMac G3@67/150/300 8 bit, 640 x 480, 67Hz is basically the same as the one above, but these two are new, and I know my drive is fragmented, so it's probably best to compare these two with each other instead of the older files generated before I fragmented my hard drive.


A while ago it got a bit warm out, and my machine became unstable, so I installed a fan. I have not collected a full set of fan data yet, but here are some of the observations. Any temps will be relative to room temperature, and since I'm not yet reporting a full suite of data, here are a few of the observations.

I've played with two fans on a thinner, black anodized aluminum heatsinks I've milled out so I could use the original Apple heatsink clip. The 12V, 0.18A sleeve bearing was 4°C cooler than a 12V, 0.10A ball bearing fan. The processor also runs 4°C cooler with the case open, but I don't recommend leaving the case open. I'm now running with my case closed with a 12V, 0.18A sleeve bearing fan, and my processor temperature hold pretty steady at 36-40°C. This is in a desktop model, and I also have two other drives in the case. These temps are from the G3 Control Strip and I believe they are based on resistance changes in a RTD. I'm not sure where this RTD is located, but the temps do not really matter; it's all relative. If the RTD reported temp is really 10°C higher than the core temp that's fine since we don't have the core temp. The important thing would be to determine what the temp is before you do anything at say 66/233, 66/266, and 66/300. Then redo those configurations with the new heatsink/fan combo with some additional thermal grease. I suspect the 66/300 temps with the heatsink/fan will be very close to the original 66/233 with just the heatsink.


And for more info check out Andreas Kann's (andreas.kann@dortmund.netsurf.de) G3 Page


Marc Schrier
(schrier@mac.com)