iMac G3 Rev A-D FSB bus overclock

[CC_333]

@croissantking Agreed!

If I sent one of my CPUs out to one of you "in the name of science" (I have three: one Rev A or B, and two Rev Cs), could I get an upgraded one in return? I'd be willing to pay!

A long shot, I know.

I'd do it myself, but I don't really have the skills, nor the equipment.

That said, would a Rev D CPU work better than a Rev. A through C? My guess is that aside from being nominally faster (333 MHz vs. 266 MHz for Rev C and 233 for Revs A and B)

This is just a SWAG, but if one RAM slot stops working at higher clock speed, the reason for the failure may be that there are signal integrity issues at the higher speed.

I wonder, if @indibil used PC100 RAM, would it tend to not work so well?

If it is a signal integrity issue, is it feasible (if not trivial) to recreate the iMac CPU board with the necessary improvements baked in?

c

 

[croissantking]

@croissantking Agreed!

If I sent one of my CPUs out to one of you "in the name of science" (I have three: one Rev A or B, and two Rev Cs), could I get an upgraded one in return? I'd be willing to pay!

Sure, let's talk. Send me a DM.

That said, would a Rev D CPU work better than a Rev. A through C? My guess is that aside from being nominally faster (333 MHz vs. 266 MHz for Rev C and 233 for Revs A and B)

If you're just looking to upgrade the XPC106 chip for 100MHz bus operation and keep the CPU, then a Rev. D daughterboard would be better, since I don't think multipliers lower than 3.5x are possible. So you could potentially leave the CPU on there and run it at 350MHz.

If you're also looking to switch the CPU for a faster one, then it shouldn't matter very much which Rev. it is. For reference, mine was a 233 so Rev A or B.

I wonder, if @indibil used PC100 RAM, would it tend to not work so well?

Looks like they used PC133 RAM from their photo.

If it is a signal integrity issue, is it feasible (if not trivial) to recreate the iMac CPU board with the necessary improvements baked in?

It'd be possible for sure, just not remotely easy without any schematics available for these boards. I'd be more interested in seeing if the originals could be modded.

Didn't the Blue and White G3 have signal integrity issues @LightBulbFun ? I think we saw that mentioned in the schematics that Arbee shared. Maybe we could learn something from the Yosemite board design.

 

[croissantking]

I ran another test. I swapped the lower 128MB RAM module for one of these 512MB ones, which only showed 256MB. It booted up. Then I tried adding the 128MB RAM module to the upper slot, and it still works like before: a chime and an orange LED.

It seems to have the problem with the upper slot. Could it be similar to what happens on the G4 AGP, where a RAM slot stops working when you overclock the bus? It would be interesting to be able to patch this.

So you've tried the 128mb and 512mb modules in the top slot only (with nothing in the lower one), and it doesn't boot?

 

[indibil]

So you've tried the 128mb and 512mb modules in the top slot only (with nothing in the lower one), and it doesn't boot?

Hi, good suggestion. I just tested it by moving the same module from the bottom slot to the top slot, and it works too.

Now I'm wondering if when I tried 128+128, both were 100MHz. I can look for two 100MHz ones this afternoon; I don't have time right now.

Thanks for your suggestions!
P.s. my cpu have a fan

 

[croissantking]

Hi, good suggestion. I just tested it by moving the same module from the bottom slot to the top slot, and it works too.

That’s kind of what I suspected. It’s not an issue with a particular slot, but with driving 2x DIMMs at 100MHz.

The XPC106 and the MPC932 are both operating within design spec so I wonder if the issue is something like non-optimal PCB routing causing jitter or signal loss. I doubt if Apple ever did any 100MHz testing, so we’re in unknown territory.

Now I'm wondering if when I tried 128+128, both were 100MHz. I can look for two 100MHz ones this afternoon; I don't have time right now.

Cool, I look forward to your report.

Thanks for your suggestions!
P.s. my cpu have a fan

Mine is bigger.

 

[croissantking]

very cool! does Open Firmware/thus OS X report the 100Mhz bus speed properly too?

Yes.



I bumped PCI to 35MHz for a 105MHz bus speed. It runs but there are problems with USB.

 

[indibil]

That’s kind of what I suspected. It’s not an issue with a particular slot, but with driving 2x DIMMs at 100MHz.

The XPC106 and the MPC932 are both operating within design spec so I wonder if the issue is something like non-optimal PCB routing causing jitter or signal loss. I doubt if Apple ever did any 100MHz testing, so we’re in unknown territory.



Cool, I look forward to your report.



Mine is bigger.

View attachment 90553

Okay, you win, yours is better, but try putting it in an iMac tray

I tried putting in two 128MB DIMMs (actually two 250MB ones, with only half of each recognized), and it worked. It seems the limit is 256MB. And the OS9 system profiler doesn't show the speed correctly.

 

[croissantking]

Okay, you win, yours is better, but try putting it in an iMac tray

It sounds like an aircraft taking off, and that’s at 7V. At 12V it’s like a hurricane.

I tried putting in two 128MB DIMMs (actually two 250MB ones, with only half of each recognized), and it worked. It seems the limit is 256MB.

I’m hoping it’s just having issues with those particular 512MB DIMMs you have, and not that 256MB total is a hard limit. Can you try some different large (>128MB) DIMMs?

And the OS9 system profiler doesn't show the speed correctly.

It’s irritating, isn’t it? Acknowledge our achievement, please! At least Mac OS X does.

 

[croissantking]

I got the bus running stable at 110MHz – that's with a 36.66MHz PCI clock.

To do this, I set the PCI clock generator (SC608) to its 35MHz setting and swapped its reference crystal for a 15MHz part in place of the stock 14.31818MHz part.

 

[croissantking]

It has the Lombard EEPROM installed, 2.5:1 ratio, the processor is running at 450MHz but the L2 cache is running at 225MHz!!!! 2:1 ratio!!! Does the ratio depend on the bus?

Looking back at your earlier post, it appears it does.

66MHz bus = 3:1 ratio
83MHz bus = 2.5:1 ratio
100MHz bus = 2:1 ratio

I'm not sure if this has ever been discussed before.

 

[herd]

https://68kmla.org/bb/index.php?threads/imac-g3-rev-a-c-g4-cpu-upgrade.5043/post-564134

 

[croissantking]

I borrowed the 512MB PC133 module from my iBook. It works fine on its own (but detected as 256MB of course) and also works with either my 32MB or my 64MB DIMM as long as it’s in the lower slot. (The other way round causes a kernel panic or lock up during boot.)

So, am running stable with 320MB RAM and a 105MHz bus.



I’m hoping to get hold of a couple of low density 256MB modules and see if I can boot with 512MB.

 

[CC_333]

Sure, let's talk. Send me a DM.

OK, will do! Kind of been going through a mental funk for the past few years and haven't had much energy for my Mac collection, so I haven't been super prompt.

I'm hoping that will change soon, though.

If you're just looking to upgrade the XPC106 chip for 100MHz bus operation and keep the CPU, then a Rev. D daughterboard would be better, since I don't think multipliers lower than 3.5x are possible. So you could potentially leave the CPU on there and run it at 350MHz.

If you're also looking to switch the CPU for a faster one, then it shouldn't matter very much which Rev. it is. For reference, mine was a 233 so Rev A or B.

I think I only have Rev A/B and Rev C machines, so it'll probably want a CPU swap.

Looks like they used PC133 RAM from their photo.

I see. I'm kind of rusty at recognizing the different sorts of RAM.

c

 

[croissantking]

I’m having success running with 512MB RAM at 100MHz bus speed.

The modules are a matched pair of low-density 256MB modules.

 

[adespoton]

I used to use unmatched low density 256MB PC133 SODIMMs at 100MHz without issues as well.

 

[croissantking]

I used to use unmatched low density 256MB PC133 SODIMMs at 100MHz without issues as well.

Which machine was that on?

 

[adespoton]

Which machine was that on?

That was on a 1999 iMac DV SE, so not that different from a Rev D.

 

[adespoton]

That was on a 1999 iMac DV SE, so not that different from a Rev D.

Ah; faulty memory; the DV SE was regular DIMMs, not SODIMMS, wasn't it?

 

[CC_333]

Ah; faulty memory; the DV SE was regular DIMMs, not SODIMMS, wasn't it?

TL;DR: Yes.

The Slot Loaders (the DV SE) used normal desktop DIMMs, whereas the Tray Loaders used laptop-style SODIMMs.

In a basic sense, they're not too different (same form factor, similar specs), but internally they're quite different, as the difference in RAM style demonstrates.

The Slot Loaders also didn't have a separate CPU daughter card like the Tray Loaders did – The CPU and supporting circuitry was integrated directly into the Slot Loader's main logic board.

c