macuserman
Well-known member
Whaaaaat! This is amazing, how did you keep this under wraps! What speed were you going for with this board?
PPC750GX vs. PPC750GL?
- Thread starter trag
- Start date
Siliconinsider
Well-known member
I honestly though no one would try because the board is exceedingly rare, and also it's a very difficult project. It took me two full month to sand down the PCB and retrace it all, and make verifications. Getting the 8 layer PCB manufactured also took a long time.
Ultimately there was the problem of sourcing the ZIF socket, and I had the idea of using a blank PCB with simple drill holes in it, and purchase some Mill Max pins that are then driven one by one in the PCB holes and surface mounted to the board.
Also, yes, the 750GX and 750GX are identical CPUs, the 750GL being the low power version. The same goes with the 750FX and 750FL, which are also drop in compatible with the 750GX. Note that the 750GL only comes in 800 and 933MHz versions, so they have always been overclocked on the PowerLogix boards. I have added some hardware DIP switches to set the PLL1, but this is probably not very useful as the PLL2 is driven by software. The PowerLogix software does allow to change the clock multiplier on the fly, and after the extension is loaded during boot.
We obviously have the same source, REI is a common source for legacy items that need to be supplied long term, often for military application. They use original masks from the original vendors to manufacture second source parts under licence when the original vendor decide to make the part obsolete. So while these look weird, they are new. The BGA balls are also shallower than the original IBM parts, I expect because they are RoHS solder balls, and the IBM ones are the traditional high melt 90Pb/10Sn balls.
Now the bad news is that the VRM of my PowerMac G3 beige did release the magic smoke when I tried it! I have traced and triple checked everything and could not find anything wrong, but it is possible that when inserted in the socket, there could be a short between either VCCcore, VCCio, or GND. Need to check if it is the high side or low side MOSFET on the VRM that blew up to give me a clue, and hopefully have a simple fix for the PCB (drilling a via on the offending connection).
Well maybe the morale of the story is that I should spend more time here!!! I hope I did not ruin any hopes or expectations, I'll gladly share what I have learned so far, and hopefully will try to resume the project as I had some major deadline to beat lately.
Ultimately there was the problem of sourcing the ZIF socket, and I had the idea of using a blank PCB with simple drill holes in it, and purchase some Mill Max pins that are then driven one by one in the PCB holes and surface mounted to the board.
Also, yes, the 750GX and 750GX are identical CPUs, the 750GL being the low power version. The same goes with the 750FX and 750FL, which are also drop in compatible with the 750GX. Note that the 750GL only comes in 800 and 933MHz versions, so they have always been overclocked on the PowerLogix boards. I have added some hardware DIP switches to set the PLL1, but this is probably not very useful as the PLL2 is driven by software. The PowerLogix software does allow to change the clock multiplier on the fly, and after the extension is loaded during boot.
We obviously have the same source, REI is a common source for legacy items that need to be supplied long term, often for military application. They use original masks from the original vendors to manufacture second source parts under licence when the original vendor decide to make the part obsolete. So while these look weird, they are new. The BGA balls are also shallower than the original IBM parts, I expect because they are RoHS solder balls, and the IBM ones are the traditional high melt 90Pb/10Sn balls.
Now the bad news is that the VRM of my PowerMac G3 beige did release the magic smoke when I tried it! I have traced and triple checked everything and could not find anything wrong, but it is possible that when inserted in the socket, there could be a short between either VCCcore, VCCio, or GND. Need to check if it is the high side or low side MOSFET on the VRM that blew up to give me a clue, and hopefully have a simple fix for the PCB (drilling a via on the offending connection).
Well maybe the morale of the story is that I should spend more time here!!! I hope I did not ruin any hopes or expectations, I'll gladly share what I have learned so far, and hopefully will try to resume the project as I had some major deadline to beat lately.
Siliconinsider
Well-known member
Oh no way! I'll check that!
All the vias are tented under the BGA and under the PGA, otherwise.
Edit: I checked that little solder blob and it did not make connection to anything, took it off.
All the vias are tented under the BGA and under the PGA, otherwise.
Edit: I checked that little solder blob and it did not make connection to anything, took it off.
LightBulbFun
Well-known member
Oh wow thats amazing! hopefully you can get the remaining niggles sorted before not too long and get it booting
its actually a really good thing you added those PLL dip-switches
because the achilles heel of the PowerLogix 750FX and 750GX CPUs is their base multiplier is only set to about 500Mhz and to clock up to full speed 1Ghz, you need to use as you say their extension and software, to set the software controlled but this software does not work with the last versions of OS X Tiger
and your also SOL if your running other odd ball setups like OS X Server 1.x or Linux etc
so im quite pleased to see hardware control over the multipliers
one thing I have always wondered is how PowerLogix got round the Unknown CPU block in BootROM of the PowerMac G3 BW, theres no BootROM patches needed and the CPU cards have no complex logic on them either
so the PowerMac G3 should not boot without patches because the 750GX's PVR is completely different to that of 750
but when you boot one of these you will see in OpenFirmware it reports itself as a regular 750 right down to the PVR value which I find most curious, I almost wonder if the 750GX itself has some sort of "legacy" mode where it pretends to be a 750?
Trash80toHP_Mini
NIGHT STALKER
ISTR the discussion predates the sinkhole by a year or two. IIRC it was a lengthy discussion in a thread about a Blackbird CPU Card G3/PBX sandwich?
Trash80toHP_Mini
NIGHT STALKER
PPC740L G3 CPU Daughterboard For Blackbird Powerbooks
Just as a point of order. That's a 750CXe, not a 750CX. The 750CXe was a slightly updated version of the 750CX and was produced in the year following the 750CX, which was 2001, with the 750CX reaching production levels in 2000. If one wants to make a point regarding the 750CX, it's not really...
68kmla.org
Siliconinsider
Well-known member
Well it was a good plan to add the DIP switches then! Did not thought of the impossibility to use software PLL settings in specific cases.
macuserman
Well-known member
My burnt 800mhz zif card arrived today and one of the pins on the zif pin array is snapped off.... I wonder if that was the cause of the initial failure or happened in shipping it wasn't packed terribly well, but I don't see the pin the bag. Not sure if it is repairable or not anymore, I was thinking chip swapping would be the way to go on it, but not sure if the zif pin array could be swapped from a donor card.
Siliconinsider
Well-known member
You could try fixing the pin, it would be difficult to swap the entire socket without doing reflow soldering with solder paste, but that's also possible.
macuserman
Well-known member
I sent you a message, be happy for you to use this as a debug board to compare with your design and possibly even repair.
This was posted elsewhere, but for reference:
www.sterpin.net/uk/pismologixuk.htm
In case the link goes dead, here are a couple of the pictures:
Recreating the 7400 to 750FX adapter board might be more versatile than recreating just the ZIF version. Maybe the cheap 750GX chips could be installed in a variety of places...
www.sterpin.net/uk/pismologixuk.htm
In case the link goes dead, here are a couple of the pictures:
Recreating the 7400 to 750FX adapter board might be more versatile than recreating just the ZIF version. Maybe the cheap 750GX chips could be installed in a variety of places...
Siliconinsider
Well-known member
That's a sweet little adapter!
Phipli
Well-known member
Hey Siliconinsider, incredible work.
Just double checking which VRM type you had. I'm not 100% sure what the failure mode is, but some Beige G3 VRMs used to stop regulating properly and toast things when you installed an upgraded CPU.
Siliconinsider
Well-known member
Oh that's interesting, yes I don't have the Royal Tech VRM in my machine, still have to move forward with tracing the problem in my design...Hey Siliconinsider, incredible work.
Just double checking which VRM type you had. I'm not 100% sure what the failure mode is, but some Beige G3 VRMs used to stop regulating properly and toast things when you installed an upgraded CPU.
theres actually a whole bunch of 750GL's on ebay right now if anyone wants some to experiment with
They're actually still available. Has anyone done anything with them?
If someone has one running, and could compare/benchmark it against some other chips like the 750F, 750L, 7400, etc... it would help to tell whether it's worth the trouble.
Daniël
Well-known member
I've been eyeing them, but the shipping options to Europe costing either $76, or $161, has so far managed to keep me from buying
To begin with: this post was made possible by joevt, who figured out the software. Thank you joevt!
So that ebay seller looks like a good source for 750G chips. The listing has been active for years and it shows hundreds have been sold. What are people doing with them?
I was curious about the 750G, as these chips are actually available and with the built-in 1MB of L2 cache it's basically the top of the line for the G3 series in the same way that the 7448 is for the G4. So I placed an order and stuck one on the CPU board out of an AGP.
It's difficult to find good comparisons between the 750G and its predecessors. There are geekbench results posted for different computers like the iBook with the 750F, or 750C, or PowerMac or PowerBook with G3 upgrades (including the 750G). However as is typical with geekbench, the numbers are highly variable (i.e. not very useful as a benchmark). Although one thing is clear, a G3 of any variety is no match for later G4 chips. Because of this and being a 3.3v chip, I think it's better suited to the 603, 604, and G3 series of computers anyway. The 7400/7410 chips are the last 3.3v G4, so comparing to those makes more sense. I had previously tested a 750S with external L2 cache in an AGP, and it was significantly slower than a 7400 G4 at the same clock. Could the 750G do better?
The chip I tried did 1.1GHz with ease, and I compared it to a 650MHz 7410 with 2MB of L2 cache (a high end 3.3v G4). In most cases the 750G was clearly faster. In a few cases where altivec was used by the software, the G4 was faster. Also, the memory bandwidth on the 750G was *much* lower. This seems abnormal, but any benchmark results I've seen for any G3 under OS X were similar. Even a G4 chip in 60x bus mode has much higher memory speed. This could be some limitation of OS X? Fortunately the nice built-in L2 cache helps a lot. With it disabled, the chip is between 2x and 8x slower. On something with a slower bus, say 50Mhz, the cache would be even more important.
Hopefully this result will encourage other experimenters out there. There are still some mysteries about the aftermarket upgrades that used these chips. If anyone has a 750G (or even a 750F) and would be willing to share data, it would be nice collect some information about these CPUs in this thread. For example, what happened with this?:
So that ebay seller looks like a good source for 750G chips. The listing has been active for years and it shows hundreds have been sold. What are people doing with them?
I was curious about the 750G, as these chips are actually available and with the built-in 1MB of L2 cache it's basically the top of the line for the G3 series in the same way that the 7448 is for the G4. So I placed an order and stuck one on the CPU board out of an AGP.
It's difficult to find good comparisons between the 750G and its predecessors. There are geekbench results posted for different computers like the iBook with the 750F, or 750C, or PowerMac or PowerBook with G3 upgrades (including the 750G). However as is typical with geekbench, the numbers are highly variable (i.e. not very useful as a benchmark). Although one thing is clear, a G3 of any variety is no match for later G4 chips. Because of this and being a 3.3v chip, I think it's better suited to the 603, 604, and G3 series of computers anyway. The 7400/7410 chips are the last 3.3v G4, so comparing to those makes more sense. I had previously tested a 750S with external L2 cache in an AGP, and it was significantly slower than a 7400 G4 at the same clock. Could the 750G do better?
The chip I tried did 1.1GHz with ease, and I compared it to a 650MHz 7410 with 2MB of L2 cache (a high end 3.3v G4). In most cases the 750G was clearly faster. In a few cases where altivec was used by the software, the G4 was faster. Also, the memory bandwidth on the 750G was *much* lower. This seems abnormal, but any benchmark results I've seen for any G3 under OS X were similar. Even a G4 chip in 60x bus mode has much higher memory speed. This could be some limitation of OS X? Fortunately the nice built-in L2 cache helps a lot. With it disabled, the chip is between 2x and 8x slower. On something with a slower bus, say 50Mhz, the cache would be even more important.
Hopefully this result will encourage other experimenters out there. There are still some mysteries about the aftermarket upgrades that used these chips. If anyone has a 750G (or even a 750F) and would be willing to share data, it would be nice collect some information about these CPUs in this thread. For example, what happened with this?:
I would be happy to plop a 750GL on it and see what happens
LightBulbFun
Well-known member
thats very cool to see, even just from a novilty perspective, just to push the G3 to its maximum limits, it would be fun to see one in an MDD, because they have a 167Mhz bus and all that
what software things did you have to figure out in the end, was it BootROM patches or OS X side software patches? I notice your still using 10.4.8 for some reason rather then 10.4.9 or 10.4.11
one thing I have always wondered about the PowerLogix 750GX ZIF's is I have always wondered how do they work? because the PowerMac G3 blue and white, has a CPU block in its latest bootROM, if the PVR is not recognised, the machine halts
but somehow the powerlogix ZIF is able to masquerade as a normal 750 G3, and its PVR is reported back as that of a 750, so you dont have to patch the BootROM on the G3 blue and white
but theres no logic on the ZIF's its all just passives, so I have always wondered how that works! does the 750GX itself have some sort of "pretend to be an older chip if you pull down this line" mode?
I have to admit, I totally forgot! and the iBook in question is still sitting disassembled in a box awaiting some new RAM as the onboard appears to be bad
I am still interested in trying it but will have to see if i can find some GL's closer to home or get someone to proxy them as the shipping is a little expensive to teh UK from that ebay listing
Similar threads
