[geeks] Mac Pro 1,1 CPU Upgrade - failure?

[Mark Benson]

On 11/12/2016 16:48, Andrew Jones wrote:

>> This says the thermal junction between the CPU heat spreader and the
>> heat-sink is bad. I took it apart, and found CPU-A heatsink wasn't
>> totally touching the plate, or at least had made an incomplete junction
>> (would explain a lot) but CPU-B looked fine. I redid both the CPU
>> heatsink junctions, using bit more arctic silver in case my skimping had
>> stopped it working. Reassembled it and... now it's even worse. The CPUs
>> are at 57C and the machine crashes within minute of booting.
>>

New rub:

I torqued down the CPU heatsink bolts 'very tight', but it doesn't seem 
to have made much odds.

I can use Linux happily, but burnP6 is taking the CPUs cores from 38C to 
58C in less than a few seconds. If I dick with the CPU fan rates (rev em 
up to 2000RPM) the CPU cores seem to sit at 58-60C and are stable in 
Ubuntu, and collapse back to 37 or so after I kill the processes.

OS X last longer if I let the machine stand for a while, than if I 
restart immediately. This is thus a heat issue, I'm pretty sure. Feels 
to me like something vital is running a thread on OS X on one particular 
CPU core that's then overheating and blinking out. IF I, again, ram the 
fan speed up, it lasts longer, but still flakes out eventually. I manage 
to get the Mac Pro 2,1 Firmware update installed on the EFI ROM and also 
flashed the SMC ROM to a matching 2,1 revision version that's better 
able to handle 8-core setups. Still no luck, though. Even the OS X 10.6 
Installer DVD locked up during the initial stages of the install.

> You want to use as little thermal paste as possible.  The coat should be
> so thin it's not visible to the naked eye.
>
> Thermal paste has a better thermal conductivity than air, but it's much
> worse than metal-to-metal contact.  The goal is just to exclude any air
> from defects on the surface.

Right, will yank them off again next time I've got time and clean off 
the excess. I am using Arctic Silver (which is metal loaded) but I don't 
suppose it's THAT much better. Given the toque I applied it

> P.S. don't take the degrees C values too literally.  I don't think those
> thermistors are calibrated against anything.  They're there to detect
> catastrophic rises in temperature to avoid toasting the chip.

Noted, although the rate at which things go up and down is concerning.