I've read on the internet ( believe everything you read! :) ) - that the iMac G5 can also be "overclocked" by moving a transistor from one place to another and making the bus speed run at 1/2 instead of the normal 1/3.

Edit : Found it - https://www.reddit.com/r/overclocking/comments/28t1gl/overclock_imac_g5_bus_speed/

Has anyone tried this?

[ Edited by MoerBoer 25.04.2022 - 12:54 ]

@MoerBoer

I am not sure that things here are so simple.

Normally (and with G5 too) you set the CPU speed by known System Bus speed and proper PLL setting (i.e. multiplier)
Then CPU speed = Bus speed x multiplier

Note: G5 have a little different PLL tables nomenclature, instead of "multiplier" uses "Ratio". But it is generally the same number. Ratio 3:1 = multiplier x3.

In case of iMac G5 (I assumed here 2.1 GHz, model is not mentioned in your link) we have 700 MHz Bus, and CPU 2.1 GHz, i.e. multiplier x3 (Ratio 3:1)

If you want change Bus speed to 1.05 GHz, you have to: Change Bus speed to 1.05 GHz somehow and change multiplier to x2 (ratio to 2:1).
I wonder if this should be done by change of two resistors. But who knows.

Not to say, that from System bus speed are often derived other speeds (RAM) and Northbridge also have his own frequency limit.

I never overclocked G5, because the overclocking scale is very rough.
Is OK with G4 change multiplier from x9 to x10.5 with System bus 133 or 166 MHz, but the G5 case is x3 or x2, it is 50% difference in frequency! And Bus speed increase in your case from 700 MHz to 1.05 GHz is also 50% step up.

In any case, if somebody want to try it, I recommend at first check iMac G5 Northbridge max frequency, probably change RAM stick to higher speed (but here is also RAM onboard and it can be showstop).
And first of all proof from other sources what exactly are R3012 and R3028 rezistors and how the frequency logic works in iMac G5.





[ Edited by sailor 25.04.2022 - 13:37 ]

@sailor

Thank you for the very informative post and yes, I have the 2.1GHz model.

I won't be doing any soldering on my baby, just thought it was an interesting post, but your information makes sense that this is not as straight forward as it looks.

In G5 models BUS-speed is bound to CPUSpeed (ratio), which is coded into CPU-module.
On a PowerMacG5 1,8GHz you have Bus-speed of half (900MHz). Changeing the CPU-Modules to a 2,3GHz-version, rises up BUS-Speed to 1150MHz automaticaly (Northbridge can do that - working on such a G5 at moment).

So its more a issue of the CPU than of Northbridge.

In iMac there are no CPU module. But of course, it can be done similar way.

But with quick search on web I found, that Front Side Bus (Processor Interface Bus) frequency on block diagrams is 667 MHz for iMac iSight and 600 MHz for older iMacs.

I don't think it will works with 1.05 GHz, but everything is possible. Only me will not do it with my iMac.

But if somebody want to try search for bus overclocking, good start are Northbridges:
Powermac G5 PCI-X have U3
Powermac G5 PCIe U4 (= IBM CPC945)
iMac AGP U3 Lite
iMac iSight some other

Isn't it 700MHz bus as it's running at 1/3 for 2.1GHz?

Quote:


Can you provide benches (numbers and subjective experience to decide wether this overclocking is worthwhile for everday usage or just for the geek factor (which of course is a pretty strong and valid point!).

Most interesting (for me): you also have an iMac G5 (as have I), how does the overclocked Mini compete with the iMac (I myself wonder how close the iMac G5 and the 1.5 Mini are, some parts - for example jpg decoding is faster on the Mini compared to the iMac, but Wayfarer is a good share faster on the iMac).

> MPC7447A HW spec I read before overclocking. But I cannot find anywhere
> MPC7447B ;-) So for simplification I assumed, that only difference was
> higher Maximum Rated Core Frequency - maybe 1.5 - 1.67 GHz ;-)

I don't know what the specific difference is between 7447A and 7447B/C, but it's not the maximum rated core frequency as there have been 7447A chips rated at 1.7 GHz, like the MC7447AHX1700PC (see Freescale/NXP PDF linked there).

> Front Side Bus (Processor Interface Bus) frequency on block diagrams
> is 667 MHz for iMac iSight and 600 MHz for older iMacs.

1.6 GHz CPU <-> 533 MHz FSB
1.8 GHz CPU <-> 600 MHz FSB
1.9 GHz CPU <-> 633 MHz FSB
2.0 GHz CPU <-> 667 MHz FSB
2.1 GHz CPU <-> 700 MHz FSB

So FSB clock rate is always ⅓ of CPU clock rate on all G5 iMacs (as was already written), unless the "ratio" has been tampered with.

> Powermac G5 PCI-X have U3
> Powermac G5 PCIe U4 (= IBM CPC945)
> iMac AGP U3 Lite
> iMac iSight some other ;-)

iMac G5 iSight also has U4/CPC945. Besides, U3/U3H/CPC925 is also used in PowerMac G5 AGP ;-)

Quote:
Of course, this is the most pretty point for me!

Quote:

Yes, I make some benchmarks. Now I am waiting for SSD+adapter, then I registrate MorphOS and after that I will gladly do comparison with my other MorphOS machines (if I solve the overheating of Powermac Quad, it will be also in comparison).

If you have exact aplication what you need to benchmark, please let me know. Otherwise I will do my usual mix of benchs, apps and games.

Quote:

Thanks, the NXP/Freescale qualification record is nice. And if you find somewhere also MPC 7447B or 7447C Hardware Specifications, it will be excellent.
Overclocking is now sucessfuly done, but at least I can store it for future cases.

BTW my 7447B is 1500 MHz marked version.

Quote:


This can be carefully checked, if the iSight have full U4 (CP925) and not "light" like AGP iMacs. If it is full CPC925 and the same revision used on Quad, it can be 1.25 GHz capable (depends on certain version used on certain iMac).
After quick look to CPC925 user manual is another good news, that CPU bus clock and Memory clock are different clock domains, i.e. changing CPU bus speed don't touch RAM speed.

So, if anybody interested in iMac bus overclocking, open your iMac, check exact northbidge type and revision, find according manual and go boldly forward!

I would love to try if I had a spare iMac G5 iSight, but I cherish my baby too much to take that chance.

>> iMac G5 iSight also has U4/CPC945. Besides,
>> U3/U3H/CPC925 is also used in PowerMac G5 AGP ;-)

> This can be carefully checked, if the iSight have full U4 (CP925) and not "light"
> like AGP iMacs. If it is full CPC925 and the same revision used on Quad, it can
> be 1.25 GHz capable (depends on certain version used on certain iMac).

There seems to be some confusion. U4 is CPC945, while CPC925 is U3/U3H.
Changing the ratio on the fastest iMac (2.1 GHz) from ⅓ to ½ would result in 1.05 GHz (as you wrote in comment #19), so testing 1.25 GHz isn't possible this way. What's the maximum frequency a "lite" version of the U4 would be able to run at in your opinion?

> open your iMac, check exact northbidge type and revision

iMac G5 iSight: https://www.flickr.com/photos/htomari/15420568058

It will be better to make new topics, if somebody want do this iMac CPU bus overclocking.


>There seems to be some confusion. U4 is CPC945, while CPC925 is U3/U3H.
>Changing the ratio on the fastest iMac (2.1 GHz) from ⅓ to ½ would result in 1.05 GHz (as you wrote >in
>comment #19), so testing 1.25 GHz isn't possible this way. What's the maximum frequency a "lite"
>version of the U4 would be able to run at in your opinion?

1.25 GHz is mentioned, because we know that 1.25 GHz of certain U4 / CPC945 version is possible ( Powermac Quad have 2.5 GHz with ratio 1:2 ). It means if in Quad is the same chip (with the exact same code) as on your photo, it will be safe from max. frequency point of view.

I don't know, what "lite" version means. I only know, that in older Powermacs is U3 and in older iMacs U3 lite. Similarly there can but cannot be U4 lite version for newer iMacs. Of course, it cannot means limited frequency, but for example bus for only one CPU, who knows.

If this photo is Northbrige from your real iMac, it is possible to find according manual. Codes on top of CPU contains also revision number. Probably revision number is also part of this northbridge code. You need the same manual version (or higher) as Northbridge have. And check frequencies and PLL configs.

And there is also second thing. In link in post #18 there is in the bottom some comments. There is said, that now he have overclocked bus frequency 900 MHz with ratio 2:1. Unfortunatelly it fits to two older different iMac models: iMac G5 1.8 20" or iMac G5 1.8 17" (ALS). So we are absolutely not sure, if resistor numbers in iMac iSight are the same, because this type is much different.

> I don't know, what "lite" version means. I only know, that
> in older Powermacs is U3 and in older iMacs U3 lite.

According to technical reports, U3 lite can interface to only one CPU where the full U3 can interface to two CPUs. Additionally, U3 lite may or may not have a lower clock rate limit compared to the full U3.

> there can but cannot be U4 lite version for newer iMacs.

In contrast to the U3 lite, I never heard or read anything of an "U4 lite", that's why I've been assuming that, opposed to the AGP G5 generation, the PCIe iMacs use a full (i.e. non-crippled) version of the U4 northbridge chip, just at a lower clock rate (maybe even underclocked for product differentiation reasons?) than on the PowerMacs and leaving one CPU interface unused.

> If this photo is Northbrige from your real iMac

This photo titled "CPC945 a.k.a. Kodiak. On iMac G5 (iSight) 17-inch 1.9 GHz" isn't my photo, just one I found through a quick search :-)

> it is possible to find according manual.

Yes, searching for the string "41E4244" from the photo above reveals the CPC945 datasheet from August 2006 (there's also a newer one from December 2007, but that should be irrelevant as the iMac G5 iSight was discontinued in January 2006). There in the "Speed Classification" column in the bottom table on page 12 you can see that the "41E4244" part comes out third fastest out of six available.
Now we'd need photos of CPC945 chips of different PCIe PowerMac models to conclude if they are from the same or faster speed classification. Maybe even the 2.1 GHz iMac G5 iSight has a faster rated part than the "41E4244" one of the 1.9 GHz model?
In any case: All listed part numbers, including the one found in the iMac G5 iSight 1.9 GHz, are full versions with two CPU interfaces, so from this I think we can assume that there exists no "U4 lite", at least not in any Mac.

@Andreas_Wolf: excellent detective investigation.

According to CPC945 user manual Table 12-5, PLL1 Clock Settings.

Table shows that CPC945 is capable upto CPU 2.664 GHz with ratio 2:1,
Unfortunatelly there is not mentioned nothing about "Speed Classification" that you find.
There is another minor thing, the CPU busses are double data rate. So for one MHz are two data transports. Bus frequency in Mac specs (everymac.com) should be Bit rate frequency.

Anyway, thanks to Andreas we know, that frequency is not big issue as I was afraid. (If in iMac is lower "Speed Classification" than in Powermac, the difference will not be so big, it is the same chip, omly sorted by quality. We are in same situation like with G4 factory tested to 1.5 GHz, overclocked to 1.83 GHz).

Btw, frequency settings of this northbridge is terrible complex, and I don't want to study it complete. It means, there are no direct PLL signals on board (like Pegasos CPU have) and there is probably Apple logic on top of this that controls the frequencies and can be setup by resistors.

And still remains to identify proper resistors. Have the iSight the same resistors numbers like previous models? Maybe if we look on motherboard there can be some small description except the resistor number.

So I suggest to find a volunteer with iMac iSight, willing to burn to dust his iMac, and risking soldering of resistors according to old type motherboard

After succesfull overclocking I want to make some tests.
But first I have some questions:

Plese, what screen resolution/bit depth do you use on your Mac Mini? I have 64 MB VRAM, but with FullHD/24-bit only two screens fill nearly all VRAM.
What is for you most suitable?

And second question @koszer: here on Mini Benchmarks is mentioned Blender. Unfortunatelly Blender's results are differs with different resolution and mainly with different bit depth. Plese, what screen do you exactly use?

Quote:
I do use FullHD/24bit resolution with my 64MB mini and it's fine for my use. I only keep Ambient in the enhanced mode (double buffered, not triple), but then rest of the screens in the non-enhanced mode, and that lets me to use even 3-4 extra screens with no slowdowns, which is more than enough for me in the most cases.

Quote:

Terribly sorrry, but I don't remember. That was - after all - more than 10 years ago. Maybe if I was Andreas Wolf...
I guess when I had Mac Mini @1,5 GHz it was connected to some XGA display (and I don't have a clue what kind of display was connected to my friends Pegasos 2). But the benchmark was always made on a fixed region so I don't think the screen res would make any difference.

Quote:

Super!
Thanks for tip. It looks like nice compromise.
Other issue with screens on FullHD resolution is, that screens switching on Mini is significantly slower than on Pegasos 2 + Radeon 9800 PRO. But it is a minor thing.


[ Edited by sailor 17.05.2022 - 09:53 ]

Quote:

Yes, Andreas can
In any case, many thanks.
Blender time on fullHD/24-bit and 800x600/24-bit differs in seconds, but with 16-bit the time is about a half on my Mini.

I forgot (or was too lazy) made some benchmarks with CPU 1.5 GHz, an now I have 1.83 and no real comparison.

Please, can any Mac Mini owner send me VRAM write speed?
Sys:Application/Benchmark/GfxSpeed, CopyMem: CPU to VRAM a WritePixelArray is enough, not all tests are needed.

I suppose, that VRAM speed is not touched by overclocking, but want to be sure.
Thanks!

Quote:
You mean in the situations when it needs to swap screens in memory? In a normal situation when all screens fit in the graphics memory and are in the same format, it's so instant that you shouldn't notice a difference in switching not matter if it's FullHD or any other resolution.

Quote:

Yes, swap two open screens.
Regardless of if it is with Amiga+M or with mouse, to swap two fullHD/24-bit tripple buffered screens takes almost one second. Unfortunatelly I don't know how measure exact time.
Lower resolution or lower depth screen swapping is fast. Also swapping fullHD/24 on Pegasos 2 is faster - this is strange, because Pegasos 2 have 1.33 GHz (vs. Mini 1.83 GHz) and have PCI 66 MHz vs. Mini x4 AGP. Only plus of Pegasos is much better graphics card (9800 PRO vs builtin 9200 )

Can it be some monitor issue if it is on the same monitor? Pegasos 2 is connected via DVI + HDMI reduction, and Mini to DVI directly.

[ Edited by sailor 17.05.2022 - 18:14 ]

mmm... very strange.
Just now I tried it again, fist few swaps are slow, and after five or six clicks swapping start to go fast. And afer reboot again slow.

But it is not big issue, just for curiosity.