Resolved Power consumption difference between CPU A and CPU B - PowerMac G5

[mikiotty]

Hi everyone!
I just acquired an Early 2005 Dual 2GHz PowerMac G5. The Mac runs great, however using iStat menus I noticed that CPU A always draws about 10 more Amps than CPU B. Obviously, power and thermals reflect the greater current, resulting in 10W greater power draw and 10C higher temperatures on CPU A, with the fans spinning more frequently on the upper processor.
CPU usage looks about the same between the two. What could cause this difference? Thank you!

 

[B S Magnet]

Hi everyone!
I just acquired an Early 2005 Dual 2GHz PowerMac G5. The Mac runs great, however using iStat menus I noticed that CPU A always draws about 10 more Amps than CPU B. Obviously, power and thermals reflect the greater current, resulting in 10W greater power draw and 10C higher temperatures on CPU A, with the fans spinning more frequently on the upper processor.
CPU usage looks about the same between the two. What could cause this difference? Thank you!

Cooling efficiency.

Specifically, a combination of ancient, dried OEM thermal paste and fine dust on fans and grilles are possible factors for uneven power consumption and cooling.

Regardless, having just acquired a G5, a thorough internal disassembly and clean-out, followed by adding in a good thermal paste (some folks like Arctic Silver, my go-to is Noctua NT-H2). Full disassembly includes taking out the heatsink/CPU assemblies and also the motherboard, in order to access and clean out the ducting and also the thermal paste on the memory controller’s heatsink (visible and accessible only after the motherboard is taken out). I did a partial clean-out of a 2004 dual G5 in 2019, leaving the logic board in place, and then doing it all over again in late 2020, this time taking everything out, after dealing with overheating and thermal-based system abort/reboot problems on the U3/memory controller, located on the backside of the motherboard. Taking that out revealed how much filth had accumulated after 16 years (with at least three or four years of sitting in storage before I got it).

If you do follow this path of overhaul/clean-out, we have a couple of threads on here with member experiences discussing their taking apart, cleaning, and fresh-pasting their G5s. This, of course, atop the iFixit guides.

But for a system which is now physically 17 years old, it’s a good bet it has never gotten a thorough clean-out, and CPU A is now running hotter because it is cooling less efficiently than CPU B — possibly due to blockage in the heat sink fins/grille.

Important: after doing all this, you will need to run the thermal calibration utility on the Apple Hardware Test disc for your model. There are a couple of places where to acquire that .dmg (which will need to be burned to a DVD), and we should be able to help guide you on where to find those.

 

[mikiotty]

Cooling efficiency.

Specifically, a combination of ancient, dried OEM thermal paste and fine dust on fans and grilles are possible factors for uneven power consumption and cooling.

Regardless, having just acquired a G5, a thorough internal disassembly and clean-out, followed by adding in a good thermal paste (some folks like Arctic Silver, my go-to is Noctua NT-H2). Full disassembly includes taking out the heatsink/CPU assemblies and also the motherboard, in order to access and clean out the ducting and also the thermal paste on the memory controller’s heatsink (visible and accessible only after the motherboard is taken out).

If you do follow this path, we have a couple of threads discussing member experiences with taking apart, cleaning, and fresh-pasting their G5s. This, of course, atop the iFixit guides.

But for a system which is now physically 17 years old, it’s a good bet it has never gotten a thorough clean-out, and CPU A is now running hotter because it is cooling less efficiently than CPU B — possibly due to blockage in the heat sink fins/grille.

I will do that this weekend, and I thought about that, but wouldn’t a “hotter” cpu be set to lower the power draw?

 

[B S Magnet]

I will do that this weekend, and I thought about that, but wouldn’t a “hotter” cpu be set to lower the power draw?

Not necessarily, no.

CPU A is the primary CPU on a dual-CPU G5 system. When using diagnostic/developer tools like Apple’s CHUD, one can disable the second CPU (CPU B) for testing in single-CPU mode, but not CPU A. Dual CPU G5s are, from my own usage experience, not optimized to load-balance across the two CPUs — which would be more a function of the software’s proficiency (i.e., OS X) to load-balance on the fly and not a flaw of the hardware or firmware itself. Again, this is an informed guess. An engineer would know much more and could probably clear up or even refute parts of this guess.

On my system, CPU A typically runs, on average, 3–4°C warmer than CPU B. I’ve never managed to have them reach parity for very long (and usually, that parity only happens in the few minutes after a cold boot when the system has been powered off for a while).

Also, following several years of use, the G5 motherboard/U3 is relying on thermal calibration values from probably when the system was built at the factory. Those values are probably no longer accurate, and the system is not accommodating for slow, gradual changes to environmental factors and external cooling efficiency (like how well the thermal paste moves heat to the copper heatsink pipe or how well that pipe is being cooled by airflow through the aluminium fins) over the span of that time.

 

[mikiotty]

Not necessarily, no.

CPU A is the primary CPU on a dual-CPU G5 system. When using diagnostic/developer tools like Apple’s CHUD, one can disable the second CPU (CPU B) for testing in single-CPU mode, but not CPU A. Dual CPU G5s are, from my own usage experience, not optimized to load-balance across the two CPUs — which would be more a function of the software’s proficiency (i.e., OS X) to load-balance on the fly and not a flaw of the hardware or firmware itself. Again, this is an informed guess. An engineer would know much more and could probably clear up or even refute parts of this guess.

On my system, CPU A typically runs, on average, 3–4°C warmer than CPU B. I’ve never managed to have them reach parity for very long (and usually, that parity only happens in the few minutes after a cold boot when the system has been powered off for a while).

Also, following several years of use, the G5 motherboard/U3 is relying on thermal calibration values from probably when the system was built at the factory. Those values are probably no longer accurate, and the system is not accommodating for slow, gradual changes to environmental factors and external cooling efficiency (like how well the thermal paste moves heat to the copper heatsink pipe or how well that pipe is being cooled by airflow through the aluminium fins) over the span of that time.

You are just confirming all my thoughts on this. I will try and blow the Mac with compressed air tomorrow, because I don’t have the time to disassemble it completely, and run the thermal calibration. I have all the diagnostic tools from my previous experience with my old Late 2004 G5 that died several years ago, and now lives in the form of a file server with Big Sur on it (if you know, you know ).
These are the temps and power consumption ad idle after 10 minutes from startup. After running Geekbench (which gave a respectable score of 1905), temps went down to 60C for CPU A and 45C for CPU B.

I am a bit worried to disassemble this machine, particularly on the CPU connectors (my previous one had one of them with bent pins). I will try to bring the temperatures as close as ideal (which they actually are, the hottest CPU never exceeded 76C on full load) without disassembling it. If it comes to that, I will surely renew the thermal paste and put some NT-H2 on the CPUs.
BTW, which OS are you running on your G5?

 

[B S Magnet]

You are just confirming all my thoughts on this. I will try and blow the Mac with compressed air tomorrow, because I don’t have the time to disassemble it completely, and run the thermal calibration. I have all the diagnostic tools from my previous experience with my old Late 2004 G5 that died several years ago, and now lives in the form of a file server with Big Sur on it (if you know, you know ).
These are the temps and power consumption ad idle after 10 minutes from startup. After running Geekbench (which gave a respectable score of 1905), temps went down to 60C for CPU A and 45C for CPU B.

View attachment 1974707
I am a bit worried to disassemble this machine, particularly on the CPU connectors (my previous one had one of them with bent pins). I will try to bring the temperatures as close as ideal (which they actually are, the hottest CPU never exceeded 76C on full load) without disassembling it. If it comes to that, I will surely renew the thermal paste and put some NT-H2 on the CPUs.
BTW, which OS are you running on your G5?

I run 10.5.8. It gets very light use because the memory controller is slowly failing (it overheats, even after the work I did on it). Unfortunately, I’m going to have to get another board or another G5 locally if I want to remedy this:

Bear in mind the system right now is running as a file server, an ARD server (serving up what I see in ARD from another Mac, making this screen cap), and has a dormant browser minimized. It’s doing next to nothing, and this is actually cooler by about 6–8°C than before disassembling it in late 2020. It is very well ventilated: the tower is not on the floor, but up at around 1.3m above the ground, on my desk, with well more than a metre of clearance on both ends. But take note of my memory controller heatsink: this is where my board’s chronic issues live. I have it only running four identical sticks of 1GB, having taken out four other sticks which come from different manufacturers, as having them all in kicks up that temperature even higher (and causing the spontaneous system reboots).

When disassembling it, the CPU pins on the daughtercards are not what I tend to worry about. The tricky part is removing the thick cable between the PSU and the front board (i.e., with the power light, extra jacks, and power button). It’s surprisingly easy to mess up this connector if one is not careful.

 

[mikiotty]

I run 10.5.8. It gets very light use because the memory controller is slowly failing (it overheats, even after the work I did on it). Unfortunately, I’m going to have to get another board or another G5 locally if I want to remedy this:

View attachment 1974717

Bear in mind the system right now is running as a file server, an ARD server (serving up what I see in ARD from another Mac, making this screen cap), and has a dormant browser minimized. It’s doing next to nothing, and this is actually cooler by about 6–8°C than before disassembling it in late 2020. It is very well ventilated: the tower is not on the floor, but up at around 1.3m above the ground, on my desk, with well more than a metre of clearance on both ends. But take note of my memory controller heatsink: this is where my board’s chronic issues live. I have it only running four identical sticks of 1GB, having taken out four other sticks which come from different manufacturers, as having them all in kicks up that temperature even higher (and causing the spontaneous system reboots).

When disassembling it, the CPU pins on the daughtercards are not what I tend to worry about. The tricky part is removing the thick cable between the PSU and the front board (i.e., with the power light, extra jacks, and power button). It’s surprisingly easy to mess up this connector if one is not careful.

So looking at your screenshot I am quite happy with my temperatures, considering that this Mac has never been disassembled (still has the original security pin on the processor cover). My memory controller never exceeded 50C, and I have all my slots populated with Kingston 1GB identical sticks. Now I know why my previous G5 failed, since it was doing 80C on the memory controller with only 2 sticks in it ?
I will have to source the long hex screwdriver to unscrew the CPUs if I decide to disassemble the Mac, I don’t have my old one anymore.

 

[mikiotty]

@B S Magnet I have another question: what would you consider "normal" temperatures for a G5?

 

[ArkSingularity]

I will do that this weekend, and I thought about that, but wouldn’t a “hotter” cpu be set to lower the power draw?

CPUs actually become less power efficient as they warm up, and will require more power to do the same tasks than they would at lower temperatures. This is due to increasing internal resistance as the circuitry heats up, and of course exacerbates heat problems and causes the CPU to run even warmer if the cooling system can't keep up.

In general, the processor won't take steps to reduce its clock speed and power consumption until it gets near thermal junction (the maximum rated safe temperature for the CPU before it starts to fry itself). I'm not sure how the G5's behave, but the Intel's generally aren't set to throttle much until they get to about 100C.

 

[mikiotty]

CPUs actually become less power efficient as they warm up, and will require more power to do the same tasks than they would at lower temperatures. This is due to increasing internal resistance as the circuitry heats up, and of course exacerbates heat problems and causes the CPU to run even warmer if the cooling system can't keep up.

In general, the processor won't take steps to reduce its clock speed and power consumption until it gets near thermal junction (the maximum rated safe temperature for the CPU before it starts to fry itself). I'm not sure how the G5's behave, but the Intel's generally aren't set to throttle much until they get to about 100C.

That could explain it. Thank you!

 

[B S Magnet]

@B S Magnet I have another question: what would you consider "normal" temperatures for a G5?

I’m not really an expert here beyond owning the presently problematic Power Mac G5 I’ve had for eight (mostly good) years, but as with any processor known to be generally hot-running, I’d like to see CPU temps during total idle (just Finder open, nothing else) someplace under 40°C, and I’d frankly be happy to never see the CPUs under load exceed 60°C.

As said, that’s an ideal place, not necessarily a realistic one — especially with this ageing hardware.

With PMG5s, whose (something like) eleven fans cannot generally (or easily) be controlled manually due to the way the design of the architecture maintains a lot of that self-regulating beyond the reach of a user (unlike, say, the ability to control internal fans on Intel Macs or on PowerBook G4s), this becomes an additional obstacle/frustration. (The one override anyone can make is to mask the silver sensor reflector area on the clear cover, to induce the pulse width fans to go into full-bore mode. It works fairly well, dropping overall CPU temps by at least ten degrees in my experience, but those fans are uncomfortably loud.)

Perhaps one day I’ll run across a cheap local G5 whose owner is wanting to get rid of it quickly. I can then use the one I have as a parts donor. My hope is I might able to come across a dual-core model (hardcore folks always want that quad-core, but I’d be plenty happy to come across, say, a dual-core 2.3GHz unit). I fathom a DC 2.3 model can be kept cooler by virtue of there being more physical space internally for the heat to dissipate more effectively.

That’s one other thing I hadn’t thought to mention previously: CPU A is likely hotter than CPU B, in addition to being the primary CPU, because physically speaking, it’s located directly over CPU B. CPU B’s heat rises. QED.

tl;dr: “As cool as possible, given your limitations/tolerances to noise.”

 

[mikiotty]

I’m not really an expert here beyond owning the presently problematic Power Mac G5 I’ve had for eight (mostly good) years, but as with any processor known to be generally hot-running, I’d like to see CPU temps during total idle (just Finder open, nothing else) someplace under 40°C, and I’d frankly be happy to never see the CPUs under load exceed 60°C.

As said, that’s an ideal place, not necessarily a realistic one — especially with this ageing hardware.

With PMG5s, whose (something like) eleven fans cannot generally (or easily) be controlled manually due to the way the design of the architecture maintains a lot of that self-regulating beyond the reach of a user (unlike, say, the ability to control internal fans on Intel Macs or on PowerBook G4s), this becomes an additional obstacle/frustration. (The one override anyone can make is to mask the silver sensor reflector area on the clear cover, to induce the pulse width fans to go into full-bore mode. It works fairly well, dropping overall CPU temps by at least ten degrees in my experience, but those fans are uncomfortably loud.)

Perhaps one day I’ll run across a cheap local G5 whose owner is wanting to get rid of it quickly. I can then use the one I have as a parts donor. My hope is I might able to come across a dual-core model (hardcore folks always want that quad-core, but I’d be plenty happy to come across, say, a dual-core 2.3GHz unit). I fathom a DC 2.3 model can be kept cooler by virtue of there being more physical space internally for the heat to dissipate more effectively.

That’s one other thing I hadn’t thought to mention previously: CPU A is likely hotter than CPU B, in addition to being the primary CPU, because physically speaking, it’s located directly over CPU B. CPU B’s heat rises. QED.

tl;dr: “As cool as possible, given your limitations/tolerances to noise.”

So today I finally managed to disassemble my G5 and clean it. I only repasted CPU A using Gelid GC-Pro (very similar to NT-H2) because I couldn't get to the bottom screw for CPU B due to my big screwdriver, but man did it make a difference! I dropped my average temperature by at least 10°C, topping up at 61-62°C (before it would easily get to 71°C).
Even after running thermal calibration though, the difference in power consumption is about the same. I'm starting to think that it's normal this way, and as soon as the computer works fine, I don't really care.
Thank you for all your help!