Power Mac G5: Processor Mounting Hardware - Precise Models

[Nullcaller]

Why?​

Lately I've been dwelling on the idea of using a CPU from the set of Late 2005 PMG5 components I bought (CPU, logic board and some other bits and bobs) to turn a Late 2005 G5 that I have into a weird 2.0 GHz Quad configuration. As far as I can tell, all of the required components for the second CPU seem to be there on the 2.0 GHz board that I have, and the few empty pads that seemingly don't have the components populated are actually also not populated on the real Quads.

To this end, I have successfully desoldered a CPU socket from a board of an earlier model that I just had lying around. This was just a test using a board that I can screw up and not be too sad about. The pads did not get damaged, and the socket got desoldered cleanly, so, you know, no harm done, in the end. I'm still waiting on 0.65 mm solder balls (my uneducated guess as to what size I need for this socket) to arrive so that I can test my solder-the-thing-back-on skills and then go for the real thing.

Spoiler: Socket Desoldering Instructions (if you're interested)

If anyone's gonna attempt something similar, I needed to raise the temperature to 350 degrees Celcius on my hot air station. I heated it at 250 degrees for 3-5 minutes, then at 300 deegrees for 3-5 minutes and at 350 for 3-5 minutes, and only then was I finally able to desolder it. Be careful when pulling it up and away from the board, as one end of the socket might not be hot enough to be desoldered while the other one is, so pull away really slowly and stop if you feel or see any resistance. There should not be any resistance whatsoever. If there is, that means you're pulling pads from the board.

I'll upload a video of the procedure when I have time for it.

But while I'm waiting on the solder balls, I thought I'd try and address another issue with this mod, which is that the heatsinks from the 2.0 and 2.3 GHz models cover the area designated for the second processor.

Upd. Jul 20: I found the part numbers for the plug and the socket and I'm looking into ordering a socket instead of desoldering one.

What?​

To start somewhere, I took some measurements of the processor mounting system of the Late 2005 PMG5, and made OpenSCAD models of:

1. the giant 7 mm thick plate to which everything mounts (without the heatpipe support piece in the area where the hotplate sits, because it can't be reached without disassembling the actual heatsink) - heatsink_plate_wo_hp_holder.scad,

2. the VRM heatsink (without the indentation where the heatpipe should sit) - vrm_heatsink_wo_hp_cutout.scad,

3. a WIP modded version of the 7 mm plate, which would allow me to mount two processors (I will need to acquire appropriate PC coolers and modify the plate further to add mountpoints for them) - heatsink_plate_mod.scad.

4. what should be a model of the processor daughter card that's precise to 0.02 mm (though it may not actually be that precise) - processor_board.scad

Couple of things about the models. They should be quite precise. All the measurements were done with callipers that should allow 0.01 mm precision measurements, and pretty much all of the variables that have their values set numerically in .scad files contain actual real measurements with no math done to them. The math is done automagically using other derived variables. All that said, however, (1) I'm a big goof and could've made a few mistakes here or there and (2) the 7 mm plate is cast and it is therefore very difficult to measure it correctly, so no guarantees. I did, however, print out a few 0.2 mm thick versions of the models on my 3d printer to compare with the real hardware, and I can attest that all the holes and openings align with the real ones quite well.

I have also made measurements of all the fasteners (except the spring, for which I don't have the equipment to measure the Young's modulus) with what is probably 0.02 mm precision and will be updating the archives with their models soon™. There's a dimension 'drawings' (more like frantic notes) file attached to the post if you want the dimensions now.

Just to be clear, all model files (.scad and .stl) published here are made by me and are licensed to anybody who wants them under CC-BY 4.0, unless the forum rules or EULA state otherwise, in which case they take precedence.

What's Next?​

Currently, the preliminary plan for making the 'permanent solution' is 3d-printing a plate, then using it to cast a part out of metal. Or perhaps paying somebody to print the part out of high-performance plastic, like PEEK. The idea is that the result should be able to fit all Quads regardless of whether or not I succeed with my weird custom 2.0 GHz thing that I'm doing. The plan for the VRM heatsink is to add the heatpipe cutouts, convert the model to a CNC-mill-friendly format and have it made in a professional workshop.

If anyone reading happens to have a pair of callipers and a Quad G5 CPU daughtercard just, you know, lying around detached from the LCS, I would be very grateful if you took some measurements of the heatpipe cutouts and the little extra copper bits right next to them, along with some measurements of the two screw holes on the side of the VRM heatsink which are different on the Quad version of the VRM heatsink.

Model Version History​

1.0 - Initial release with 2.0 GHz heatsink plate (without heatpipe backing section), 2.0 GHz VRM heatsink (without heatpipe cutouts), WIP modded 2.0 GHz heatsink plate.
1.1 - Added precise processor board model (including processor package and die, socket and VRM heatsink mount post), added chamfer generation code to lib.scad.

 

[Nullcaller]

By the way, here's some information on how the G5 heatsinks are put together, while I struggle with the socket.

They seem to be using what I'm assuming is the standard Tin-Bismuth[-Silver] low temperature solder to affix the heatpipes to the cold plate. Here's a photo of the Late 2005 G5 heatsink with that shiny metal stuff:

The older model heatsinks I have also seem to be using the same stuff:

The cooling fins (on all heatsinks, including the VRM heatsinks), however, seem to be using thermal epoxy of some sorts. It's that dark-grey stuff on the heat pipes:

I've tried to remove some of it with a sandpaper, and sure enough, I see none of that metallic sheen:

The fact that the thermal epoxy isn't applied along the entire circumference of the heatpipes, but instead in relatively thin lines going along the length of the heatpipe seems to suggest that the heatsinks mostly rely on press fit for thermal conductance, rather than on the thermal epoxy. The thermal epoxy is probably there mostly to hold the heatsinks together, and not to act as a thermal interface material. The full disassembly of the processor mounting hardware seems to involve removing the cooling fins both on the later and the older models, which would involve somehow undoing both the epoxy and the press fit.

 

[Bug-Creator]

As far as I can tell, all of the required components for the second CPU seem to be there on the 2.0 GHz board that I have, and the few empty pads that seemingly don't have the components populated are actually also not populated on the real Quads.

Even if this is true, and if there is no lock against it in FirmWare, it still seems like an a## backwards approach.

If you have access to a real Quad, use that board, slap in 2 2.0GHz CPUs (this works the other way round) and maybe downgrade the PSU as a next step.

You should now have a 2.0GHz Quad with a little bit less power used.

 

[Nullcaller]

Even if this is true, and if there is no lock against it in FirmWare, it still seems like an a## backwards approach.

Depends on what it is you think I'm trying to achieve. I highly doubt I'll be able to come across a Quad where I'm at, and my goal with this project is not to get a cool weird Quad anyway. It is cool, and it is weird, but what's more important is that I want to understand how these machines work and improve my skills to a point where I'm comfortable working on them. If there's a firmware restrisction, so be it, I'll try and learn how to bypass it. If the hardware just plain doesn't support it, hey, I learnt something. I don't want to have a Quad. I want to make it.

 

[Nullcaller]

A little bit of an update with my socket shenanigans.

On the scale from a working Quad to the Hindenburg disaster, today has been off the charts. That's because the Hindenburg disaster, at the very least, resulted in a cool band name. But "Led-free Socket Balls" doesn't have quite the same ring to it as "Led Zeppelin", I'm afraid.

The socket is significantly harder to remove on the Late 2005 boards compared to the earlier boards. When I desoldered the socket from an earlier model board, I didn't remove any heatsinks and it was completely fine. 350 degrees, boom, desoldered. This resulted in me misjudging the situation with the Late 2005 board and thinking all it needed was maybe a bit more heat. But turning the heat up to 400 degrees just resulted in socket damage, at which point it was better to just remove it and replace with a new one later down the line, since I did recently find that you can just order them from Mouser. In the end, I needed to raise the temperature all the way to 480 degrees (the maximum value for my hot air station) to remove the socket at all, slowly and painfully. Usually, you know, the solder balls melt and let go of a surface-mount component. But not this time. This time, the socket itself deformed and let go of the solder balls instead. I later tried melting the solder balls, and 400 degrees from my hot air station wouldn't do it. It was completely fine when I added leaded solder, though, of bloody course.

Morale of the story, if you're gonna do anything with a Late 2005 board, make sure you remove all the heatsinks (I didn't even know they were removable, and to be fair to my past self, they're about as easy to remove cleanly as the socket) and probably use a preheater placed underneath the area you want to work on, set to something like 100-200 degrees.

One good thing is that the board seems completely unphased by all of this. "What's this, 480 degrees? It tickles, I guess" There's no damage, no heat marks, no nothing. I did pull two pads when removing the socket, but I honestly expected much worse with this socket fiasco. Those are rookie numbers, and ultimately just a good pretext for me to practice fixing pulled pads. I'll fix the pads, order some sockets, and put this little disaster behind me, hopefully.

Meanwhile, imma go back to modelling fasteners and stuff. Way easier. Way less stressful. Way more rewarding.