G4 7448 and 7457 swap

[ervus]

I gave the 150MHz bus mod a try and it seems to work. I set a quicksilver 7450 CPU for 667MHz (5x133MHz) and it runs like it's at 750MHz (5x150MHz). The firmware reports 667. Are the AGP and PCI bus speeds higher too?

The RAM I used is PC133 CL3. I didn't need to try different RAM sticks. Does anyone know how to play with the various RAM timings? Perhaps some firmware settings/scripts could change it.

 

[evanboonie]

I gave the 150MHz bus mod a try and it seems to work. I set a quicksilver 7450 CPU for 667MHz (5x133MHz) and it runs like it's at 750MHz (5x150MHz). The firmware reports 667. Are the AGP and PCI bus speeds higher too?

View attachment 2290780

The RAM I used is PC133 CL3. I didn't need to try different RAM sticks. Does anyone know how to play with the various RAM timings? Perhaps some firmware settings/scripts could change it.

I'm not able to find a datasheet for the C5003 clock chip. The iMac schematic I referenced only shows the full bus speed options. On the Sawtooth/Gigabit, the 120MHz bus speed actually slows the PCI and AGP clocks to 30 and 60MHz respectively. It seems like the PCI clock must be an even division of the bus clock and the AGP clock is double the PCI clock. Looking at the iMac schematic some more, I'm actually not sure if the PCI and AGP clocks are derrived from the bus clock in machines using the C5003. The C5003 itself doesn't have any outputs labled for PCI/AGP. Instead, the "Pangea" chipset has separate clock options and bootstrap pins for the PCI and AGP clock. The Pangea straps set both the AGP and PCI clock to 66MHz, so I don't think changing the bus speed in the Quicksilver will impact the PCI/AGP speed in any way.

Edit: I believe that the PCI clocks are set to 33MHz on the Quicksilver. The 66MHz in the iMac schematic is likely specific to that machine. The Pangea chip allows 33 or 66MHz options.

 

[ervus]

The Pangea chip allows 33 or 66MHz options.

So could I set the PCI slots in a quicksilver to 66MHz? That might be interesting.

1.8GHz lines up evenly with a few different available multiples of bus speed, so this shows back-to-back memory tests using the exact same hardware with a 7447 in a quicksilver at 100MHz bus, 150MHz bus with "CL3" RAM, and 150MHz bus "CL2" RAM. It would sure be nice to fiddle with software settings in software, like they can do on the other side. Isn't open firmware supposed to be better than a DOS BIOS from 1975? So when you put a PC133 RAM stick into a mac at 100MHz and system profiler says PC133, what timings does it actually use?

 

[evanboonie]

So could I set the PCI slots in a quicksilver to 66MHz? That might be interesting.

The schematic does seem to suggest that. I have no idea which resistor it would be on the Quicksilver, but in the iMac it is connected to pin 35 of a chip labeled "FEPR-1MX8", pin T33 of the "Pangea" chip, and pin 6 of the PCMCIA/AirPort slot. The resistor locations are labeled "R193" (4.75kO) for the pull up (33MHz), and "R194" (33.2kO) for the pull down (66MHz). I have never been able to find a schematic for a Power Mac G4, so it might take some probing around on the board to find the proper resistor to change if they aren't labeled the same as the iMac. There are 8 configuration straps for setting the PCI/AGP settings, so they are likely placed close together on the PCB if that helps in tracking them down.

 

[evanboonie]

I managed to get a 7410 running at 600MHz with a 120MHz bus speed in my Gigabit G4. I have attached some photos of MACHSpeed showing the settings and the MacBench results of various configurations. I don't think MACHSpeed shows the cache speeds properly for these cards, as it should always be 1/2 of the core clock. It was pretty unstable at first and even forced me to repair disk on my OS9 volume at one point. I was able to get it running stable by increasing VDD/VCore to 2.05V, changing out my hodgepodge of memory for some matched 133MHz CL2 sticks, and adding an extra capacitor to the blank location near the inductor. I'm not positive that the capacitor is necessary, but I have a bunch of donor cards and it should help with power stability at the higher loads this overclock will place on the VRM. OSX Tiger still Kernel Panics when I try to shut it down sometimes, so 540MHz is probably a more realistic overclock for non-benchmarking purposes. I have also attached a photo of the resistor settings needed to put a 7410 at 2.05V and a 5X multiplier. The card in the photo is a donor 7400 card, so ignore all of the missing components. The settings needed are configured the same as the working 7410, I just didn't want to remove it from my machine again.

 

[indibil]

Hello,

I was reading about the different G4 models produced, and I read that the evolution that replaced the 7448 was the MPC86xx.

I guess the change was as big as the G5. It would be impossible to adapt and use an MPC86xx on a Power Mac G4, right?

MPC86xx on eBay


.

 

[sailorMH]

Unfortunatelly it is impossible.
Classic G4 CPUs upto MPC7448 is CPU core only.
MPC86xx series insist of one or two e600=G4 cores, coherency module, memory controller, PCIe, network controller...

You have made totally new computer design if you want this CPU.
Something like project Pegasos 8641D was.

 

[indibil]

Thanks for answering.

If it had been possible, it would have occurred to someone before, not me, hehehe.

 

[im_to_hyper]

Is this the CPU for the PowerPC laptop design? Or is that an OpenPOWER based design?

 

[sailorMH]

Is this the CPU for the PowerPC laptop design? Or is that an OpenPOWER based design?

No, MPC86xx is a little bit obsolete.
As I know, NXP T2080 was selected. But I don't know, if it is latest version.

4x e6500 core, 2 threads per core, 64-bit, PCIe 2.0/3.0, DDR3

 

[indibil]

Hello! I have been testing removing, reballing and replacing some MC7400 and now I see myself perhaps having the possibility of performing the CPU SWAP of a Mac Mini G4 1.42GHz.

I have a MC7447B-VH1667ZH purchased a while ago. Would it be a direct replacement? Being a 7447 like the original, I won't need a firmware update, right? To reach 1.8GHz I don't know if I will need to increase the voltage, if necessary, how do I adjust the voltage on a Mac Mini?

Thank you!!

 

[sailorMH]

My Mac Mini has original CPU from factory MC 7447B-HX1500QH.
i.e. specification is 1500 MHz and probably 1.24 -50/+100mV ( voltage is probable, because I have no datasheet for this exact "B-1500Q" version, the number is taken from B-1420Q version )

This CPU I successfully overclocked to 1.83 GHz without voltage change.

About your CPU I also have no exact data - 7447B-1500Z has 1.15V -30/+50mV, 7447B-1700Z has 1.32 -50/+100mV. And all 1667 version data I have ( -W, -D, -Y) has 1.28V -30/+50mV.

Simply - if you have no exact datasheet both for old 1.42 and new 1.667 CPUs, you cannot be sure about voltage. You have to try it. ;-)
And B/C CPU versions has max temperature 78 or 85°C.

Please, if you find howto adjust Mac Mini voltage, share it here...

 

[indibil]

Hello! This morning I decided to swap the CPU of a Mac Mini G4 1.25GHz that I had saved for testing.

Before removing it, I measured the CPU voltage on the capacitors next to it and with the tester it was 1.28V. I would have liked it to be a little more, but it wasn't enough.

My preheating equipment is a bit rudimentary, I won't go into details, but it works fine for now. I have removed the 7447A 1.25GHz and installed the 7447B 1.67GHz.

At 1.25GHz it works perfectly, and has passed the Geekbench test. The easiest speed to reach to start with was 1.92GHz, I just had to add a 1k resistor, before removing any pre-existing one, which would take a lot of work to put back.

I assembled everything, I pressed the button and..... "Chang!" was heard, and immediately afterwards the apple, the circle appeared and it arrived on the desk!!! It has booted to 1.92GHz, correct diagnosis and it recognizes it.

To assess the increase, I ran Geekbench, but here the problem appeared, the Mini freezes after 29-31 seconds of execution.

I think that 1.28V is not enough to keep 1.92GHz stable, and surely going down to 1.87GHz will stabilize it, but I always have time to do it.

I don't think anyone has documented it, from what you're telling me @sailorMH, but how would it be possible to increase the voltage to 1.35v? which IC is responsible for it? Maybe there is a small IC near the capacitors, on the underside, but since it is BGA it is impossible to find out which resistors act on it.

Maybe @dosdude1 , @Jubadub or @JoyBed can help me!!! Am I forgetting someone?

 

[indibil]

Maybe this is the IC that adjust the Vcore?

https://www.insidegadgets.com/wp-content/uploads/2016/10/MAX1993.pdf

I don't understand how does it work.

 

[dosdude1]

Maybe this is the IC that adjust the Vcore?

https://www.insidegadgets.com/wp-content/uploads/2016/10/MAX1993.pdf

I don't understand how does it work.

Yes, it's the MAX1993. Its output voltage is controlled via a resistor divider feedback loop. There will be two resistors on the board that make up the voltage divider, though I haven't looked into it enough at this point to figure out which they are.

 

[indibil]

Yes, it's the MAX1993. Its output voltage is controlled via a resistor divider feedback loop. There will be two resistors on the board that make up the voltage divider, though I haven't looked into it enough at this point to figure out which they are.

Thanks for answering.

This morning I ran tests. I have been able to locate Rc, which measures 100ohm, but I have not been able to locate Rd. But since Rd is connected to ground, I have measured between Rc and ground and it gave me 110ohm. So I have ventured to carry out a test. I have provisionally soldered a 2k2 resistor between Rc and ground, resulting in a resistance of 104ohm.

I have booted the Mac Mini at 1.92GHz and measured voltages between 1.34 and 1.37V. I have managed to increase the voltage!!! It has started and when running Geekbench it freezes again after 32 seconds.

I'm starting to think that the problem may be temperature and not voltage. But I don't hear the fan running very fast.

On the Mac Mini, the fan does not reach high speeds? Or maybe he doesn't know how to manage excess temperature? Could I speed it up through software?

I will appreciate your instructions.

 

[indibil]

I have done more tests.

What I have done by adding the resistor is to add another resistor of a higher value in parallel to Rd, to decrease the value of Rd and thus increase the Vcore. Since I have not located Rd, and it connects between Rc and ground, I have played with a resistor in parallel to lower the value and obtain a higher voltage. Look at the screenshot I attached of the MAX1993 datasheet.

This afternoon I did more tests. At first I thought the problem was temperature, so I managed to power the fan externally and get it to work at maximum. But again the same result, frozen at 33 seconds of Geekbench. So maybe the voltage was still not enough, so I tried raising the voltage higher. Initially it was 1.28v, now I had it at 1.34v, and I have gone up to 1.43v (I measure it on a CPU capacitor). Even increasing the voltage, it always freezes after approximately 32 seconds.

If I use the normal computer, use the internet and others, it also freezes after a while.

I have measured the CPU voltage all the time and I do not observe any drops, nor high peaks (activating MAX on the tester).

I'm baffled, I'm afraid I'll have to go down to 1.83GHz, and resign myself to not being able to maintain 1.92GHz.

 

[indibil]

Hello! I'll tell you news!!! A person who has skills with overclocking and swapping CPUs has helped me, without his instructions I would not have been able to do it.

I was doing the voltage increase wrong. DO NOT ADD ANY RESISTANCE IN PARALLEL AS I INDICATED BETWEEN Rc AND MASS!!

I have managed to have the Mac Mini G4 stable with an MPC7447B 1667@1.92GHz:

First I will remember that I swapped the original processor for an MPC7447B-1667:

The correct scheme of MAX1993 is this, although the datasheet values do not correspond to reality:

On the Mac Mini PCB, R3 and R5 are located here:

The value of R5 measured is 41k. First I added a 330k resistor in parallel with R5 to get 36k. With this value I have gone from 1.28-1.31V to 1.34-1.38V. It was still not stable.

Then I changed 330k for 220k, obtaining an R5 of 34k and reaching 1.38-1.41V. It was still unstable, with both Geekbench tests freezing at 30 seconds. This has made me deduce that the problem is not voltage, the problem must be current. The MAX1993 can also limit the current.

R3 and R4 are responsible for managing the current limit. Then I added a 330k parallel resistor to R3 (34k) to obtain 31k, and with this value I managed to raise the upper current limit, and make it totally stable at 1.92GHz!!!! I have passed Geekbench two times and PowerFractal three times.

The fan is heard faster, but nothing alarming, it is not turn at 100%.

I have to test if it would be stable at 1.28-1.31V with the current limit increase. But today I don't have any more time, another day.

p.s. The post in my blog:
https://rosysumenteinquieta.blogspot.com/2023/12/mac-mini-g4-192ghz.html

 

[LightBulbFun]

thats a very interesting find on the current limit, I have known others with Mac Mini's to hit the same 1.83Ghz wall as you did, despite even some people I know changing out VRM inductors etc

are you going to try for 2Ghz? that would be fun to see


@dosdude1 I wonder if this might be worth looking into for the 7448's that would not clock well perhaps they where tripping a current limit rather then a clock limit


although I do wonder what the current limit is, and how much it is safe to exceed before VRM mosfets and the such like start to complain

 

[dosdude1]

thats a very interesting find on the current limit, I have known others with Mac Mini's to hit the same 1.83Ghz wall as you did, despite even some people I know changing out VRM inductors etc

are you going to try for 2Ghz? that would be fun to see


@dosdude1 I wonder if this might be worth looking into for the 7448's that would not clock well perhaps they where tripping a current limit rather then a clock limit


although I do wonder what the current limit is, and how much it is safe to exceed before VRM mosfets and the such like start to complain

Yeah, I’ll definitely have to look into that... Never even noticed that these buck regulators have current limit settings LOL. Will definitely have to check for that on other systems.

 

[im_to_hyper]

No, MPC86xx is a little bit obsolete.
As I know, NXP T2080 was selected. But I don't know, if it is latest version.

4x e6500 core, 2 threads per core, 64-bit, PCIe 2.0/3.0, DDR3

Ah nice! Do you know if they have selected an open GPU design yet? Would be interesting to see if they would go with Radeon/Arc.Xe/RISC-V for that solution

 

[indibil]

I have add a copper heatsink in my Mac Mini 1.92GHz

https://rosysumenteinquieta.blogspot.com/2023/12/mac-mini-g4-192ghz-con-disipador-de.html

 

[LightBulbFun]

I have add a copper heatsink in my Mac Mini 1.92GHz

https://rosysumenteinquieta.blogspot.com/2023/12/mac-mini-g4-192ghz-con-disipador-de.html

I look forward to the 2Ghz attempt I think I have only ever seen one Mac Mini G4 hit 2Ghz, and that was with a massive heatsink coming out the top of it, I guess they managed to keep it cool enough to stop it tripping the OCP...

I do really want see what could be done with a 7448 now the OCP has been figured, if you use a Last gen Mac Mini G4, PowerMac10,2 then you dont need to worry about firmware patching, as the 4.9.x firmware supports the 7448 natively

you will need an NVRAMRC script to boot MacOS, but thats not too hard to do

"fastest" flash, and using a pencil to overclock.

Another huge update: I FULLY FIXED the NVRAMRC script, and it can now be loaded on boot automatically without issue! I have attached a text file containing the updated script; just set it as NVRAMRC using Terminal in Mac OS. Then you should be able to set use-nvramrc? to true as well, and the...

forums.macrumors.com

I wonder if one could find a copper heatsink with the same fin-stack density as the original, for maximum improved cooling?

 

[indibil]

There is no need for more fin density, copper is capable of displacing heat much better. The original has more fins but they are very thin and made of aluminum, surely the upper half of the heatsink does not even get hot when in operation. This copper one is 2mm thicker at the base, and has enough fins. The fan now runs at very low rpm, it doesn't need to blow harder like the original, it has more temperature range.

I will think about whether I dare to go up to 2GHz, but I don't know how forced the power stage is now that I have raised the OCP limit.

My model is not one of the latest, it had a 1.25GHz, it would need a patch and NVRAMRC update, but in Spain I have difficulties finding a 7448, I would like to...

 

[incc74656i]

I have quite a lot of those salvage 7448s and 7457s. The 7457s are great, and haven't had any issues so far as I've used them (on MDD cards). The 7448s, however, are a bit more finicky. A couple of the ones I have ended up having very small cracks in the ceramic interposer, making them completely useless. The ones that did work (most of them) were able to run at 2.0 GHz with some careful VCORE voltage tweaking. I also recently reverse engineered the system firmware of some later PPC Macs, and was able to successfully patch them to add 7448 support (see https://forums.macrumors.com/thread...g-a-pencil-to-overclock.1973903/post-29596069).

im looking to mod my QS dual card. i was thinking of 7455B chips for better OC at pushing 1.6ghz or so. do you feel thats a better route than an interposer adapter and some other chip? i dont want to loose cache of course so im looking for ppc's with cache channels.