> 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.

If it takes a second, then it sounds that the screens don't fit into the graphics memory, and they must be swapped between RAM and VMEM when you switch between them. Probably triplebuffering and enhanced mode screens consume too much memory to fit in the gfx mem on your setup.

When I have the FullHD enhanced mode Ambient + other non-enhanced FullHD screens (which consume wayy less memory than enhanced/buffered screens), switching between screens take only fraction of a second... it's immediate and you can't say it would take any time in human eyes. This is because there's enough graphics memory to store them all.

Maybe your 9800 PRO has more graphics memory and it's faster because of that (all screens fit into the graphics card's mem)?

Quote:
You're probably on the limit that if they all fit in the graphics memory or not. For enhanced mode screens every open window etc will consume memory more than they would do on non-enhanced display.

@jPV
thanks, it is very good explanation.
It is the case. Swapping od fullHD doublebuffered + fullHD non-enhanced is fast.

Sorry, I have to again get use to 64 MB VRAM

Can I disturb you and kindly ask you of your Mac Mini GfxSpeed / CPU to VRAM a WritePixelArray?

I get 215 MB/s with those tests.

Many thanks,
I have 215 MB/s too.

I was afraid, that I corrupt something during overclocking. But is seems to be normal with Mac Mini G4.
It is a little, because my Pegasos 2 ( again this machine! ) have 232 MB/s.

I assumed much more here, as Mac Mini have AGP x4 video bus and Pegasos only PCI 66 (~ AGP x1).
So Pegasos have speed 87% of theoretical bus maximum, and Mini 20% of theoretical bus maximum? Strange.

I always suspected all-in-one computers of inefficiency
A am not expert on MorphOS, only power user, but I supposed, that these benchmarked functions are the fastest for write to gfx card memory.

And please, why CPU to VRAM and WritePixelArray are allways nearly the same numbers?
Original AmigaOS and AmigaOS 4 have big diferences here, because COPY to VRAM is without DMA. Does it mean, that MorphOS use for both functions DMA and different names are for compatibility resons? Or I am totally lost?

Thanks to all ( especially to cyfm and jPV ) for help and advices.
Here is link to my article with final result. Unfortunatelly only google translated.
And here is original in Czech language.

Very cool article! Nicely done sailor.

Now do an Imac G5 iSight :)

[ Edited by MoerBoer 26.05.2022 - 11:51 ]