Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Comprehensive Guide to Converting G5 Quad to Air Cooling
- Thread starter mac57mac57
- Start date
- Sort by reaction score
What difference is there between the dual pump and the single pump CPU cards? I thought only the cooler was different.
I can confirm that the actual PCB is the same between them. The only difference is the height of the integrated heat pipe / cooler assembly.
I have actually swapped out that assembly from Dual Pump cards with the larger assembly from the Single Pump cards and the result functions perfectly. I visually examined the CPU card PCBs before doing this - they ARE the same card.
I have actually swapped out that assembly from Dual Pump cards with the larger assembly from the Single Pump cards and the result functions perfectly. I visually examined the CPU card PCBs before doing this - they ARE the same card.
hi everybody! Many thanks for the comprehensive guide to the OP; this is a new concept about this subject that had been recurring many times on this forum. Reading so many interesting news about air coouling of the Quad. I approached to this (for me) very difficult modding operation myselt, in the past nobody had ever attempted to customize a PC/intel/AMD heatsink as the OP did...before everybody went the "original DC 2,3" modding way, and I followed it too. Having at my disposal only two 2,3 GHz G5 CPUs, i purchased two DC 2,3 heatsinks and modifying them with the help of a workshop I fityted the processor cards on them so as to obtain a 'FrankenQuad' 4x2,3 GHz. In the beginning all was well, temps were down an I was a happy camper with my reborn machine, but things started to crumble when suddenly I got the dreaded 'corrupted screen' @mac57mac57 is writing about in the previous posts. Currently ALL of my two CPUs are afftected by this issue, I tried to swap them and running only one or the other in the top CPU A slot....no way. I carefully verified with the help of a powerful lens if there were dents or cracks on the dies that could explain this behavior, but I saw nothing particular. I thounght that in the end maybe the 20 years old components and the die itself can have degraded with heat over time, or that the solder joints (the infamous lead free solder) can have given way somewhere. Has somebody - apart the OP- found this problem somewhere? Is there any solution?? (here in Europe these cards are a true rarity, let alone in Italy) . Thanks
Yes, exactly. heatsinks have been CNC'd and processors were fitted on them. Everything was perfectly running, very good temps also. Until one day, I fired it up and found this kind of desktop corruption you find in the enclosed photo. It's not the GPU, I replaced it with two different Geforce 6600 from other (functioning) G5s and the issue was there just the same.
Attachments
Wait, from this picture, I can only see a single 2.5GHz CPU. Are both of them installed?
I would not use a modded OS, as long as the root cause is found at least.
So first let's check if the CPU cards are socketed, and disconnect any optional add on cards. Connect only the disk that you use for the OS. Reset PRAM and NVRAM. Then run thermal calibration and the extended test in Apple Hardware Test.
Since you tried known good GPUs, I have a feeling one of your RAM sticks have failed. I have also encountered graphical glitches when using a cheap SSD with a SATA addon card in my G4, so it may even be that your hard drive is going bad and the OS is corrupted.
So if you go through all the steps and AHT doesn't report RAM issues, it may find the hard drive is failing.
If not, I'd reinstall the official Leopard and then run some tests on that, to stress the system and see how it is under load.
When does this glitch appear by the way? What do you have to do to trigger it?
Note:
My G4 MDD is picky regarding RAM, so when I was debugging it, the tests would in some cases finish and report the RAM issues only when I ran the extended test. In some cases AHT would not even load but freeze, or freeze while running the memory tests. So I recommend you take out all modules and use only one pair that you know are working and compatible, especially if you're using different modules. This way the tests will complete faster too, and you'll have time to check the rest afterwards. I have 8x2GB ECC DDR2 sticks in my Quad and the extended test took around 10 hours to finish.
Last edited:
I can confirm that the G5 (Quad) can be funny about RAM. Mine came with 4GB of Non-ECC DDR2 and I wanted to see if I could upgrade it to 8GB with some of the DDR2 memory I still had around. I did eventually find a pair of 2GB modules that worked, however a bunch of others I tried would either prevent the G5 from booting, or (and I have literally never seen this before) just simply not show up at all. To make things even weirder, with the "new" RAM in certain slots, the G5 would only see some of the known working modules, presumably the ones closer to the CPU than the non-working ones.
Anyway, in case of sudden inexplicable problems with a G5 that worked before, I'd always check the RAM first as well.
Anyway, in case of sudden inexplicable problems with a G5 that worked before, I'd always check the RAM first as well.
@silentdragon95, I'd like to expand upon the "what to do when my formerly working G5 no longer does?" question.
I agree that RAM is often a culprit, but after much research, prompted by my own G5 Dual, which years ago just stopped working, and so stubbornly that I thought it was permanently bricked. I tried everything. I reseated the RAM, I reset the SMU, I unplugged it and left it sit overnight.... Nothing seemed to work. The machine stayed stubbornly quiet, refusing to spring to life.
My continued Google search for a solution eventually turned up one unexpected result, which turned out to not only work, but has solved similar problems for me many times since. Simply stated, reseat the video card. This little bit of "magic" revived my G5 Dual immediately. Expanding this thought, I would suggest reseating all PCIe cards in the slots. Perhaps there is something funky about the G5's PCIe bus that can occasionally get it into an inoperative state? I don't know, but I do know that this has solved the "non-responsive" problem for me many times over the years.
So, yes, reseat the RAM, reset the SMU, but if none of that works, reseat the video card, and perhaps all PCIe cards present.
Note that there is a little plastic tab at the front of the video card slot that catches when the card is inserted and holds it in place at the front end. The slot screws hold the other end in place. You will need to essentially break that little tab off to reseat the card. It feels wrong, but do it!
I agree that RAM is often a culprit, but after much research, prompted by my own G5 Dual, which years ago just stopped working, and so stubbornly that I thought it was permanently bricked. I tried everything. I reseated the RAM, I reset the SMU, I unplugged it and left it sit overnight.... Nothing seemed to work. The machine stayed stubbornly quiet, refusing to spring to life.
My continued Google search for a solution eventually turned up one unexpected result, which turned out to not only work, but has solved similar problems for me many times since. Simply stated, reseat the video card. This little bit of "magic" revived my G5 Dual immediately. Expanding this thought, I would suggest reseating all PCIe cards in the slots. Perhaps there is something funky about the G5's PCIe bus that can occasionally get it into an inoperative state? I don't know, but I do know that this has solved the "non-responsive" problem for me many times over the years.
So, yes, reseat the RAM, reset the SMU, but if none of that works, reseat the video card, and perhaps all PCIe cards present.
Note that there is a little plastic tab at the front of the video card slot that catches when the card is inserted and holds it in place at the front end. The slot screws hold the other end in place. You will need to essentially break that little tab off to reseat the card. It feels wrong, but do it!
Last edited:
Well, @haramia, that has been my consistent experience on every late 2005 G5 I have worked on.
Yes, if you are really dextrous, you might be able to push that tab out of the way while simultaneously pulling the card out of the slot, but I couldn't do it. In the end, I just snapped off the little plastic tab and all was well.
Yes, if you are really dextrous, you might be able to push that tab out of the way while simultaneously pulling the card out of the slot, but I couldn't do it. In the end, I just snapped off the little plastic tab and all was well.
Last edited:
I have not followed this topic closely enough but I thought the whole idea of using the HS from the 2.3 and 2.0 DC was that they DIDNT need modding? Somewhere on this forum I have seen people saying some HS require no mods so if it ISNT the DC HS then which model HS is it? I just bought a G5 Quad (not got it yet) and want to keep it long term
My interest in the G5 comes and goes in cycles. You know, I find some interesting info, try to do something, then I get stuck at some or other point, and I get distracted either by life or by something else. I guess it's an ADHD thing or something.
So anyway, I've been on a new cycle as of late, and so I've been perusing the various documents that the Internet has to offer on the subject. You know, looking at leaked IBM documentation, collecting datasheet revisions, as you do. "Ooh, this one says it's v2.2, I don't have that one yet!.." And this time around I've come across a number of very interesting documents that I haven't seen before, which I think are of interest for the purposes of this particular forum topic.
The first one is called 'IBM PowerPC 970MP RISC Microprocessor and CPC945 Bridge and Memory Controller Design Guide'. This is a document that is interesting in and of itself, as it describes in detail the overall high-level system architecture, which I've been slowly trying to piece together from various datasheets, manuals and schematics. But for the purposes of this topic, I think section 3, called 'Thermal Design Guidelines', is particularly interesting.
The second one is called "Development of BGA Solution for the IBM PowerPC 970 Module in Apple's Power Mac G5". I think it's particularly interesting in this context because it is provides you with a ballpark estimation of the numbers a cooling system should aim for, and allows you to understand and take note of some of the design considerations IBM engineers had when designing the original cooling systems.
Lastly, there is the third one, which is the 'PowerPC970MP Thermal Considerations' application note, which doesn't exactly say anything new in particular, but I think is interesting nonetheless. I also think you'll likely agree with me if you just look through it without focusing on specifics, and it also seemed weird not to include it here.
All of this is to say, from reading this documents, it really feels like IBM engineers were terrified these chips are gonna cook themselves, so they did what engineers do in such circumstances, and decided to double and triple and quadruple check everything. Thermal diodes, mounting force, mounting force planarity, all of that for the chip itself and for the CBGA package, the effect of thermal cycles, the unformity of mounting force spread on the solder balls... It feels like they've tried to do everything in their power and outside of it to ensure that the cooling solution is functionally perfect. If they could hire a dude that stands on the factory line and tastes a little bit of underfill from each processor board, going "ooh, that one's good" and "no, absolutely not, this one goes to the scrap bin", they would.
Now, whether or not the fear the G5 chips instilled in IBM engineers was justified, I don't quite know. But I also can't help but feel some secondhand fear from reading all of those documents, and that's why I felt it important to post these documents here. 'Cause, you know. We all need some more fear in our lives (/j)
So anyway, I've been on a new cycle as of late, and so I've been perusing the various documents that the Internet has to offer on the subject. You know, looking at leaked IBM documentation, collecting datasheet revisions, as you do. "Ooh, this one says it's v2.2, I don't have that one yet!.." And this time around I've come across a number of very interesting documents that I haven't seen before, which I think are of interest for the purposes of this particular forum topic.
The first one is called 'IBM PowerPC 970MP RISC Microprocessor and CPC945 Bridge and Memory Controller Design Guide'. This is a document that is interesting in and of itself, as it describes in detail the overall high-level system architecture, which I've been slowly trying to piece together from various datasheets, manuals and schematics. But for the purposes of this topic, I think section 3, called 'Thermal Design Guidelines', is particularly interesting.
The second one is called "Development of BGA Solution for the IBM PowerPC 970 Module in Apple's Power Mac G5". I think it's particularly interesting in this context because it is provides you with a ballpark estimation of the numbers a cooling system should aim for, and allows you to understand and take note of some of the design considerations IBM engineers had when designing the original cooling systems.
Lastly, there is the third one, which is the 'PowerPC970MP Thermal Considerations' application note, which doesn't exactly say anything new in particular, but I think is interesting nonetheless. I also think you'll likely agree with me if you just look through it without focusing on specifics, and it also seemed weird not to include it here.
All of this is to say, from reading this documents, it really feels like IBM engineers were terrified these chips are gonna cook themselves, so they did what engineers do in such circumstances, and decided to double and triple and quadruple check everything. Thermal diodes, mounting force, mounting force planarity, all of that for the chip itself and for the CBGA package, the effect of thermal cycles, the unformity of mounting force spread on the solder balls... It feels like they've tried to do everything in their power and outside of it to ensure that the cooling solution is functionally perfect. If they could hire a dude that stands on the factory line and tastes a little bit of underfill from each processor board, going "ooh, that one's good" and "no, absolutely not, this one goes to the scrap bin", they would.
Now, whether or not the fear the G5 chips instilled in IBM engineers was justified, I don't quite know. But I also can't help but feel some secondhand fear from reading all of those documents, and that's why I felt it important to post these documents here. 'Cause, you know. We all need some more fear in our lives (/j)
Thanks for the screen shot. That is EXACTLY the type of screen distortion I was experiencing. It was definitely NOT the video card. I tried several, all with the same (or equally bad) results.
Lots of trial and error later, it was clear that it was one of the two CPU cards. I eventually isolated which one it was and ultimately was able to replace it. I now have a full Quad again, instead of the half Quad I WAS running when it had only one working CPU card.
So, in your case, solving this requires either figuring out what is wrong with the errant CPU card and fixing it (a task far beyond my expertise level!) or getting a "new" card to replace the bad one. This is the route I took - I managed to find yet another G5 Quad at a very reasonable price and purchased it, scavenging the CPU cards, RAM, etc. from it. All of the extracted goodness went into my main G5 Quad and it now runs beautilfully.
It has been many months since I did the air cooling upgrade on one Quad and an LCS rebuild on the other, and both run well now. I use my air cooled Quad pretty much every day now . All these months later, CPU temperatures remain in the high 30 C region, which is excellent. I have seen one of the CPUs spike up to the high 70s during build processes, but always only for a short duration and then right back down again. I would say the machine is thermally stable!
Good luck with yours!
Now that I actually think about it, this image corruption is kind of quite interesting. That is, it shouldn't be happening. The CPUs are only really connected to U3/U4 north bridges, via elastic interface, which is the bulk of the 300-pin connector. The corruption yall are experiencing really feels like something physical, like some of the IC pads not making a good enough connection, or some sort of internal board damage. If the system loads successfully and is stable except for the image corruption, I can't really think of any way the CPU cards might be damaged in to cause this.
pure guess - but may be cache coherence (or lack of it) can cause such graphic corruption? On x86 some vram can be mapped "write combined", not sure what G5 does for such cases ....
I think you're probably right! I've studied the screenshot on a larger monitor (I haven't been able to view it on anything but my phone until now), and I now see that the image corruption seems to have happened with the wallpaper, and not the other UI elements. See, I was thinking it'd been acting as a sort of an overlay on top of the screen. So if the system is mostly stable, it would make a ton of sense if errors only occur on a delayed write to RAM (EI) from the CPU. Like, data sits in a fast buffer, gets corrupted because the buffer isn't working correctly, then gets sent out on the EI bus. Bad news is, in that case, the CPU dies are probably toast.
As for how that might have happened, electromigration is one thing that comes to mind... Thing is, there are only two thermal diodes on the 970FX, and the 970MP probably didn't improve on that, much. So if your system shows you that the overall die is at 70 degrees, chances are, there are a few hotspots on the die that are at 100 degrees or more. Which is no problem, since semiconductor designs are usually evaluated to work at temperatures from -40 to 125 degrees. But that's why mounting pressure is important when it comes to 970MP and bare dies in general: these hotspots are tiny, you have to make sure the heatsink is thermally 'connected' to the areas where the hotspots are, otherwise they'll overheat and die faster than the thermal diode can say "wait, waah?"
As for how that might have happened, electromigration is one thing that comes to mind... Thing is, there are only two thermal diodes on the 970FX, and the 970MP probably didn't improve on that, much. So if your system shows you that the overall die is at 70 degrees, chances are, there are a few hotspots on the die that are at 100 degrees or more. Which is no problem, since semiconductor designs are usually evaluated to work at temperatures from -40 to 125 degrees. But that's why mounting pressure is important when it comes to 970MP and bare dies in general: these hotspots are tiny, you have to make sure the heatsink is thermally 'connected' to the areas where the hotspots are, otherwise they'll overheat and die faster than the thermal diode can say "wait, waah?"
I can confirm that swapping the errant CPU card fully resolved the problem. So, no matter what the root cause may be, it was on that card. Install fresh card, and the problem goes away. I have not experienced ANY screen artifacts since.