Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Wild Hare

macrumors regular
Original poster
Mar 29, 2021
124
123
Top.png
Acquired this with an actual truckload of older Macs years ago and almost yearly now… I trot it out and try it again. (In Quicksilvers and other G4s.) *Pardon the heatsink paste still on the processor. (7447A)

CPU.png

BTM.png

Have tried innumerable dip switch settings / combinations and even some machine firmware updates (supposedly for use with this little time-sucker) all to no avail thus far. Even have settings provided from suppliers that once sold this thing and nothing has ever proved functional.

So it now comes down to someone possibly having an identical one to provide their specific settings and the machine application for it.

Anyone?

Thanks.
 
View attachment 1850209
Acquired this with an actual truckload of older Macs years ago and almost yearly now… I trot it out and try it again. (In Quicksilvers and other G4s.) *Pardon the heatsink paste still on the processor. (7447A)

View attachment 1850210

View attachment 1850211

Have tried innumerable dip switch settings / combinations and even some machine firmware updates (supposedly for use with this little time-sucker) all to no avail thus far. Even have settings provided from suppliers that once sold this thing and nothing has ever proved functional.

So it now comes down to someone possibly having an identical one to provide their specific settings and the machine application for it.

Anyone?

Thanks.
@bunnspecial has a working Newertech 2.0Ghz G4 CPU. Maybe he'll chime in.
 
  • Like
Reactions: Wild Hare
Thanks. Would be absolutely great to have this perpetual mystery finally solved.
 
@LightBulbFun has one also.

Mine is currently deep in storage, but I'll see if I have a photo that shows the dip switches. Several years ago, I took mine apart because someone was claiming that their Powerbook had a 7448 and at the time there were few if any photos of 7448 dies in circulation.

With that said, mine is in a Digital Audio, which is more like a Quicksilver than different.
 
View attachment 1850209
Acquired this with an actual truckload of older Macs years ago and almost yearly now… I trot it out and try it again. (In Quicksilvers and other G4s.) *Pardon the heatsink paste still on the processor. (7447A)

View attachment 1850210

View attachment 1850211

Have tried innumerable dip switch settings / combinations and even some machine firmware updates (supposedly for use with this little time-sucker) all to no avail thus far. Even have settings provided from suppliers that once sold this thing and nothing has ever proved functional.

So it now comes down to someone possibly having an identical one to provide their specific settings and the machine application for it.

Anyone?

Thanks.


you can find a good picture of my card in my thread here :) https://forums.macrumors.com/threads/powermac-g4-fun-part-2.2246307/

did you touch the dip switches above the CPU? those are what control CPU voltage and can lead to physical damage if set incorrectly, so its good to know if they have or have not been messed with

(from your photo they thankfully look to be in their stock setting)

also do you have a picture of the 7447A showing the markings on its die? this will tell me what its factory speed is, and then we can figure out what the dip switches are for that and then go from there :)
 
I've not tried it in a Sawtooth (yet).
But I think I have one "partially resurrected", downstairs.
Later this week can test.

DIPS.png

Chip.png
 
alright cool thats a 1.25Ghz 7447A


so this is what id do, see the dip switches on the bottom of the CPU right under the Transistor marked "Q15"?

from left to right (1 2 3 4 5 6) I would set them like so

ON OFF OFF OFF OFF OFF

down is off and up is on, in other words push switch 1 up to ON, and push all the other switches down to OFF

this will set your CPU multiplier to 9X which in a Sawtooth/GigE/Cube (100Mhz FSB) will set the CPU to 900Mhz or in a 133Mhz system (Digitial Audio/QuickSilver) will set the CPU to a still safe speed of 1.2Ghz, both of which will be safe stable speeds for testing regardless of which system you have on hand


once you have done this try the CPU in a system you know you have patched the BootROM on and verified STILL works with its Stock CPU

then we can go from there :)


looking at your pictures it looks like the CPU is currently set to a 17x multiplier so in a 133Mhz system that would be trying to run at 2.26Ghz! which obviously wont work with a 1.25Ghz rated 7447A so that may be your issues all along :)

(and in a Sawtooth/100Mhz system thats 1.7Ghz which is a fair old overclock for a 1.25Ghz rated part)
 
Thanks. Hopefully will test with QS again in the next few days.;)

(Boot ROM version 4.2.8f1 - in the Quicksilvers.)

Have also downloaded (again) NewerTech Firmware Update 3.1b5 & 2.0F1.1 Enablers - yet still question as to whether either is necessary, as some of these were supposedly P&P.

Will also check if “CPU Plugins” file was removed from the System Folder/Extensions/Multiprocessing folder, in previous failed attempts to install here.
 
I've got one too! It has the 7448 and it's the 1.8ghz version. @Wild Hare I have mine in my sawtooth and it works great in it.

@LightBulbFun do you have any info for the voltages dip switch? I've tried to run mine at 2ghz, it's rather unstable and locks up my whole system when under load. I also want to try and push it a bit past 2ghz, since the 7448 supposedly can go up to 2.5ghz
 
  • Like
Reactions: Wild Hare
@Wild Hare you will need the NewerTech Firmware update 3.1b5 as this a 7447A upgrade, they are not plug and play and do need BootROM patching

tis only 7450/7451/7455 based CPU upgrade cards that are plug and play

@LightBulbFun do you have any info for the voltages dip switch? I've tried to run mine at 2ghz, it's rather unstable and locks up my whole system when under load. I also want to try and push it a bit past 2ghz, since the 7448 supposedly can go up to 2.5ghz
I dont have any info on hand on the voltage dip switches but I very much plan to look into it them at some point, because like you, id like to see just how far I can push mine :) (2.4Ghz is the highest I have heard of out of a 7448 but I have never actually seen proof of that, highest I have actually seen is 2.13Ghz but that was just out of a 1.4Ghz part so who knows what a well binned 1.6Ghz or 1.7Ghz part could clock to!)

with regards to yours, mine was also sold as a 1.8Ghz Upgrade but overclocked to 2Ghz just fine, im curious do you have a picture of the actual 7448 installed on your Card and what its stock speed is?

it may be worth trying to play with your RAM sticks shuffle em about, remove 1 or 2 or try others, as I found when I installed my 7448 upgrade it made my Sawtooth Reject a good amount of my SDRAM that I had floating around, so in the end I just decided to treat it to 4 Brand new Fresh 512MB PC133 sticks maxing it out at 2GB and solving my stability issues :)
 
@LightBulbFun Alright cool! I don't have any pics but i do plan on re pasting the cpu this weekend, so i'll be sure to take some photos of the die. Thanks for the tip about the ram, I'll give that a try and see if it helps some. It's weird- at 2ghz on tiger it's very unstable and freeze prone, on leopard it's more stable but doing high cpu tasks like running geekbench seem to crash the system. Oddly it seems perfectly fine on linux at 2ghz. I've tried lubuntu 12.04 and 16.04 and don't recall ever having the system crash on me.
 
My thanks again to all responders here. Regretfully this item is once again going back into storage. (Installed a 933 MHz QS CPU instead and still seek an Apple 1GHz DP QS CPU.);)

Quote from OWC back in 2018: “It should go up to 1.6 GHz, but that was 12 years ago. We’ve no clue what kind of torture it was put through in those long years.

Tested it again yesterday & last night using settings provided here by @LightBulbFun AND the following from OWC / after the 3.1b5 firmware update & including the NVRAM set ups:

For 1.4 GHz…
“SW3: 1 UP, the rest down
SW1: 2, 3, 4 UP, the rest down”

“Once you have established that as stable try this:
1.53 GHz
SW3: 2 and 3 UP, the rest off
SW1: 1, 2, 3, 4, 5 UP, and 6 down”.


All attempts resulted in black screen, no-chime result and the heatsink never warmed up beyond room temp… as if no power ever reached the 7447A chip. Checked seat & re-seated and that provided same black screen, no-chime and no boot. This may be the reason that this first came to me loose in a box with its’ fans scavenged… it’s toast! But of course I’ll dig it out again in a year or so and try, try again.

*Funny (not funny)… I’ve a similar Sonnet that installed easily and continues to perform w/o fail.
 
All attempts resulted in black screen, no-chime result and the heatsink never warmed up beyond room temp… as if no power ever reached the 7447A chip. Checked seat & re-seated and that provided same black screen, no-chime and no boot. This may be the reason that this first came to me loose in a box with its’ fans scavenged… it’s toast! But of course I’ll dig it out again in a year or so and try, try again.
hmm that does not sound good, if you have a Digital multimeter see if you can probe across one of the yellow capacitors close to the CPU, as those will be across vCore and let you measure how much voltage the CPU is getting if anything at all

(make sure you do still have some sort of heat sinking on the CPU however when you do so!)

I wonder if it has power issues, if it might be worth Supplying 12V to the J1 location which will bypass the motherboard's CPU voltage supply and supply the CPU with 12V directly (which is then regulated down by the VRM on the CPU card itself))

this was commonly done with CPUs for G4 cubes to by pass the Cubes weak vRM system, and curiously has been done to my card so I wonder if mine came from a cube at some point? either way if you go to my thread linked above you can see in the pictures on Page 1 post 1 what im on about with regards to J1 :)
 
  • Like
Reactions: Wild Hare
Thanks @LightBulbFun - I did wonder about those wires connected to the J1 location in that pic and intended to ask if they were possibly for power supply to heatsink fan(s). Also in that same pic, it appears that your SW1 dip switches… 4 and 5 are ON and all the others set to OFF?

I may try the “alternative power in” approach via J1 connections - if there’s no chance of any (further) damage to this processor (that already will not function).

Also I wonder about additional grounding to the board via the typically used mount post that is not used for these types of processor upgrades. (As if that might make any difference.) Picture below is of a similar, working Sonnet upgrade in use here.

Purple.png
 
I would be careful about hooking 12v to that CPU. These boards are normally powered with 5v through the 300-pin socket and if you connect it to 12v you might end up with more fried stuff than just the CPU.

In your picture, "This post." is the 12v power feed for Quicksilver CPUs. Trying to connect that to ground might also do some interesting things...
 
Have also downloaded (again) NewerTech Firmware Update 3.1b5 & 2.0F1.1 Enablers - yet still question as to whether either is necessary, as some of these were supposedly P&P.

just reading back, you say Downloaded, but did you actually install the 3.1b5 Firmware patch?

is very much necessary, none of the 7457/7447/7447A/7448 upgrades where Ever plug and play, all of them needed prior firmware patching before they would boot

without the firmware patch you would just get fan spin and a black screen no chime

so as I said above make sure you have the firmware patch installed first...

the Sonnet firmware patcher will also work if your looking for something a bit more user friendly
 
Yes. Firmware patch was installed. Thanks.:)

Traded it away to someone much more competent with board-level repairs. It was damaged.
It now has a new 7447 processor & other “bits”. (Running @ 2.1 GHz - when last reported.)

NewerTech.png
 
“Once you have established that as stable try this:
1.53 GHz
SW3: 2 and 3 UP, the rest off
SW1: 1, 2, 3, 4, 5 UP, and 6 down”.
So, as i seem to have reasonable temperatures in the Cube, now, with the base-fan powered with 12 instead of former 5 volts, i'm not really done with that overclocking topic.

So, question is, do you know which voltage the 2 and 3 on, rest off, setting of SW3 referres to. Or is that just the output of some trial and error or the setting, your card, clocked to 2.0GHz (?) came with?

Printed on DIE specification of the CPUs of my card is 1.333. So 2.0 seems a bit out of reach. But 1.8 should be possible...with the right voltage-setting.
 
When you say "CPUs" it sounds like you have a dual version of this card? These are really nice cards because they allow you to easily adjust the speed and voltage of each CPU independently. i.e. the settings do not need to match between the two.

When you say "SW3" it sounds like you have a card that was originally sold for a tower. I'm curious then how you are powering it and how you are cooling it. The Cube VRM is a weak point in terms of upgrades, and it can fail dramatically.

I've seen chips marked 1250 run at 2GHz. Ability to overclock seems to be mostly related to the mask revision and luck. I have experience with these cards and can suggest some settings. What are the current speeds/voltages and temperatures? PowerFractal and dnetc are the best programs I've seen to do load testing (watch temperatures if you try them in an overclocked/upgraded Cube!).
 
When you say "CPUs" it sounds like you have a dual version of this card? These are really nice cards because they allow you to easily adjust the speed and voltage of each CPU independently.
Yes, it's a Dual. But "sadly" no 7448, but 7447a, rated @1333. I think originally sold as 1.5GHz.
i.e. the settings do not need to match between the two.
I think, i'd prefer to run them at the same clock, even having seperate dipswitches.
When you say "SW3" it sounds like you have a card that was originally sold for a tower. I'm curious then how you are powering it and how you are cooling it.
No, even having the voltage switches, it should be a Cube-version, as it came with the flat heatplate and - more important - with the VRM-bypass allready attached. Cooling is with a 15mm thin 80mm fan, (now) running at 12v.

The Cube VRM is a weak point in terms of upgrades, and it can fail dramatically.
I have a Giga-C. But the Cube does not boot without the bypass beeing plugged in anyway.
I've seen chips marked 1250 run at 2GHz. Ability to overclock seems to be mostly related to the mask revision and luck. I have experience with these cards and can suggest some settings. What are the current speeds/voltages and temperatures?
Can not view or take pictures at the moment, as the part is build in. But speed is 1.6. 1.7 booted fine but was very unstable under load with the stock voltage. So, without increasing this, it will most likely not go any faster.

Sadly cannot readout the temps, because the CPU-manager soft only runs up to 10.4.8 while the Cube runs Leo.
PowerFractal and dnetc are the best programs I've seen to do load testing (watch temperatures if you try them in an overclocked/upgraded Cube!).
Do they work with Leo? Never heard of anything, that could read out these card's temp-sensors but PowerLogix own soft.

Yes, suggestions for the voltage/SW3 settings would be very nice! 😀
 
  • Like
Reactions: LightBulbFun
If you've compared single and dual versions of these cards you may have noticed that besides the CPU chip there are a bunch of corresponding large power supply components (either present or absent). These are part of a buck converter that drops the voltage of the input power (5, 12, or 28v) to the low voltage used by the CPU chip: Vcore (~1.3v). Each CPU chip has its own power supply. It makes sense that the Cube will not boot without power (VRM bypass) connected.


The Vcore on these is configured with a voltage divider between the Vcore and the feedback to the logic controlling the buck converter. A voltage divider is a simple ratio of resistor values, and changing the value of either resistor (within reason) can adjust the resulting Vcore.


One simple way to change the value of a resistor in a circuit is to add another resistor in parallel. With surface mount components, a new resistor can be easily soldered directy on top of an existing resistor. This will result in a lower resistance. If you do this to the voltage divider you can either raise or lower the Vcore, depending on which side of the divider you make the change.


To make it easier to change the Vcore on these boards, the later versions have the SW3 switch block added. This basically selectively connects resistors in parallel with the existing voltage divider to *raise* the Vcore. The amount of change and resulting value is all determined by the values of the various resistors installed. I've seen different values used, particularly between 7448 and 7447 versions, and between Cube and tower versions. You could look up the datasheet on the controller chip for the buck converter, measure the resistor values, and calculate the resulting Vcore. And/or you could use a volt meter to measure the operating voltage on an installed and running board.

Now as I mentioned, the SW3 block connects (or adds) resistors in parallel. So with all of these switches OFF, the card is set at the lowest possible voltage (without soldering compoonents). Turning switches ON will raise the voltage, from whatever the "base" Vcore is.

The first two switch blocks set the speed for each CPU: SW1 controls the speed setting for CPU0 and SW2 does the same for CPU1. SW3 has 6 switches and sets (or can increase) the voltage for both CPUs. Switches 1-3 are for CPU0 and 4-6 are for CPU1. Note that this is "flipped" geometrically because looking at the board, the 3 switches on the left are for the CPU on the right. Now if you look at the other side of the board, just behind that switch block, you'll (probably) see 6 resistors in a symmetrical configuration. This means that on SW3, switch number 1 does the same thing for CPU0 that switch number 6 does for CPU1. Looking at these 6 resistors again, you'll (probably) see that the outside resistors have the highest values, which means they'll have a smaller effect when added in parallel. The 3 resistors for each CPU have different values and can be added (i.e. cumulative) in different combinations to get various increases in Vcore.

Maybe you don't care about any of this and just want to "make it go," but with a little understanding of what is going on you should have an easier time than randomly flipping switches.

TL;DR:

Looking down at SW3 with position 1 on the left, all OFF is the lowest voltage and all ON is the highest. The two outside switches (1 and 6) make the smallest changes and the two inside make the largest. So the sequence of symmetrically increasing voltage for both CPUs would look like:

Code:
vvvvvv
^vvvv^
v^vv^v
^^vv^^
vv^^vv
^v^^v^
etc...

Again, this assumes your board is assembled like others I've seen.



When overclocking these boards, particularly in a Cube, temperature is important. Fortunately there is a built in temperature sensor that can be read by software (CPU Director). Rather than turn everything up to eleven and hitting go, I would suggest incremental changes while monitoring temperatures and stability.

For example, if the card is currently at dual 1.4GHz, I would try bumping one CPU to 1.5GHz. If it's not stable, then I'd set it back and try increassing the other CPU. If it were stable and cool enough, then I would bump up the second CPU to 1.5GHz, and so on... If I reach a point where one of the CPUs is not stable, but temperatures are still low, then I would try bumping up the corresponding voltage for that CPU. Changing the voltage usually means disassembly of the heatsink, so I try to "map" out the stable speeds for both CPUs before changing voltage(s). The speeds and voltages do not need to match between CPUs. The CPU chips probably came from the same "batch" but will not necessarily overclock the same.
 
Now as I mentioned, the SW3 block connects (or adds) resistors in parallel. So with all of these switches OFF, the card is set at the lowest possible voltage (without soldering compoonents). Turning switches ON will raise the voltage, from whatever the "base" Vcore is.
Thanks for this extended explanation of things, which will help me alot! Sure it is way more useful to understand (a bit of) what's happening than just randomly trying!

So one of the first questions is answerd. The one, if this switchblock regulates the core-voltage of individual CPUs or of both at once. But in general it shows the way to go.

I think, i will transfer the card to a tower first, as it's a lot less hassle to mount/unmount it like in the Cube. Second, i will create a 10.4.8 boot-volume for easy access to temperature readings via CPU Manager. Thirde, i will run the card like it is now trough some benchmarks an look what temeraures i get with that. And after this, i start to moderately increase the voltages and the clocks along with it, keeping an eye on the temperatures, given it stays stable at all.

So let's see, if i can get a Dual 1.8 at least.
 
So let's see, if i can get a Dual 1.8 at least.
Long story short: I could not! ☹️

Yesterday evening it tried some, but i must confess, i lost faith to the project very early and, in the end, am happy to be back where i started now.

I swapped the card to the Sawtooth, cause the Cube would have to be completely dissasembled and reassemled for every change of settings to be made. Not even the multiplyer switches are accessible with the card build in. But there was the first problem: The card does not work without the VRM-bypass-feed...which the tower does not provide. So i could kind of solve this with a dead Cube-mainboard i had around of which at least the 28(?)volt throughput for the ADC power worked. With that attached, i ran the card first like it was, to see, if it works at all. Worked fine, so i was ready for the next step.

I removed that Cube-specific heat plate to check the present settings...to see, that SW3 had switch 2 and 5 set to "on". Which adds the second resistor to each CPU's core voltage feed, as i learnd. So i also switched on 1 and 6 to increase voltage as moderate as possible. And also set the multiplyer form 16x to 18x (which might have been a bit to much in one step). With these settings the Mac panicked right in the middle of the OS loading. Ok, too high clock or to low voltage! But then came the worse: Even with lower multiplyer-settings, even lower than i started with, output was always the same: No bong and kernel-panic halfway up the boot. Not even also setting the voltage down again could cure this. Later came to my mind, that i at least should have tried a PRAM-reset, as this was neccessary just to bring the bong back even with another CPU. But at this moment i was very afraid, my nice dual CPU-card was toast. And the certainty, that it would turn out plain impossible to find another one of these, namely the special Cube-version, was not made to increase my mood at this point.

So i decided to cancel the project. If only there was something to cancel! I set the card back to it's initial settings, reassembled the Cube and desperately hoped for everything to work like before. Lukily it did! Booted without any glitches and Geekbench finished multiply times also with no issues and same results as before. Not many times i have been happy like this to see a Mac running. I'm very glad now, to (still) have a Dual-1.6GHz-G4 Cube. And i will not put it into any overclocking-adventures again.

...even with Geekbench 2.2.7 only scoring in the 1190ies, which is not much more, than the guys with the 2GHz singles get. And even if i'm missing some of the card's real potential...

😕
 
Last edited:
Thank you for reporting your experience. It could be that the thermal paste between the CPU dies and Cube heat plate was disturbed when you adjusted the voltage, which lead to higher temperatures? Did the dies have good contact with the plate? Fresh thermal paste? Did the plate get pretty hot when you had it running in the tower? Or did you use some other heatsink setup?

My Cube has a heat plate with cutouts near the switch blocks (similar to the tower heatsinks), so it's possible to adjust the speed without removing the CPU. fyi, there is a 17x setting (in between 16 and 18).

cutouts.png


Also, if you happened to take any good pictures of the back side of the CPU card, I could probably work out what the voltages would be from the labels on the resistors.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.