Thank you all to those whom posted their SMC dumps.
I've confirmed the following facts:
1) Every single MacPro5,1 ever sold on the planet is affected by this bug. There is no difference between the systems, nor have there been any minor hardware modifications to the logic board or CPU daughter card. This is a fundamental issue caused by the SMC firmware only.
2) Apple gives special treatment in the SMC firmware to the AMD 4870 and NVIDIA 880GT, possibly the 5870 as well (but not the 5770). It essentially "assumes" these cards have their own fans and ignores their power consumption through the PCI-e slots.
3) For all other PCI-e cards, Apple uses the power consumption of each PCI-e slot (plus the auxiliary cables for slot 1) to calculate the relevant fan speeds. The thermal readings from the system ARE taken into consideration but this DOES NOT determine the calculated min/max speed of the fan (only what speed the fan runs at between these values).
The reason why some people appear to be affected by this more then others appears to have to do with how much power the system is consuming (and whether or not you've got the Apple RAID card installed, or dual CPUs, how much RAM you've got, how many disk drives and what those drives are, etc etc). I can find absolutely no evidence anywhere that this is a hardware problem caused by physical differences between the computers.
TLDR; Apple's SMC firmware is faulty. Their assumptions about the fan speeds are incorrect because the eVGA 680 GTX (and others) initialize into a low power state right off the bat, which throws off the initial SMC fan speed calculations.
There appear to be two solutions to this problem:
1) Get Apple to fix their SMC firmware (how, I'm not entirely sure- they could easily assume that ALL cards with the AMD or NVIDIA PCI-e VID are graphics cards, and therefore have their own fans- or they could fix whatever it is that actually determines the fan speed incorrectly)
2) Get eVGA to release a firmware that puts the card into a high power consumption mode until the host drivers initialize
#1 would obviously be preferential, but I believe that if the GPU were drawing more (or close to max) current during bootup, the SMC would correctly calculate the fan speeds based on the maximum current drawable by the GPU, and this would likely solve the problem. Of course this would mean that the GPU fan would need to throttle up to 100% during boot up, but since the 5870 Radeon (from Apple) does the same thing I doubt anyone would care.
Either way, it looks like we're hosed for now. I'd try and fix the issue myself, but the MacPro5,1 and 4,1 both use a split primary/secondary SMC setup (one chip on the logic board, one on the CPU card). To my knowledge, Apple has never released a flashing application capable of programming both SMC chips as the only EFI utility we have (SMCFlasher.efi, also known as "smcutil") was designed for the single SMC systems (namely the MacPro1,1, 2,1, and 3,1).
-SC
I've confirmed the following facts:
1) Every single MacPro5,1 ever sold on the planet is affected by this bug. There is no difference between the systems, nor have there been any minor hardware modifications to the logic board or CPU daughter card. This is a fundamental issue caused by the SMC firmware only.
2) Apple gives special treatment in the SMC firmware to the AMD 4870 and NVIDIA 880GT, possibly the 5870 as well (but not the 5770). It essentially "assumes" these cards have their own fans and ignores their power consumption through the PCI-e slots.
3) For all other PCI-e cards, Apple uses the power consumption of each PCI-e slot (plus the auxiliary cables for slot 1) to calculate the relevant fan speeds. The thermal readings from the system ARE taken into consideration but this DOES NOT determine the calculated min/max speed of the fan (only what speed the fan runs at between these values).
The reason why some people appear to be affected by this more then others appears to have to do with how much power the system is consuming (and whether or not you've got the Apple RAID card installed, or dual CPUs, how much RAM you've got, how many disk drives and what those drives are, etc etc). I can find absolutely no evidence anywhere that this is a hardware problem caused by physical differences between the computers.
TLDR; Apple's SMC firmware is faulty. Their assumptions about the fan speeds are incorrect because the eVGA 680 GTX (and others) initialize into a low power state right off the bat, which throws off the initial SMC fan speed calculations.
There appear to be two solutions to this problem:
1) Get Apple to fix their SMC firmware (how, I'm not entirely sure- they could easily assume that ALL cards with the AMD or NVIDIA PCI-e VID are graphics cards, and therefore have their own fans- or they could fix whatever it is that actually determines the fan speed incorrectly)
2) Get eVGA to release a firmware that puts the card into a high power consumption mode until the host drivers initialize
#1 would obviously be preferential, but I believe that if the GPU were drawing more (or close to max) current during bootup, the SMC would correctly calculate the fan speeds based on the maximum current drawable by the GPU, and this would likely solve the problem. Of course this would mean that the GPU fan would need to throttle up to 100% during boot up, but since the 5870 Radeon (from Apple) does the same thing I doubt anyone would care.
Either way, it looks like we're hosed for now. I'd try and fix the issue myself, but the MacPro5,1 and 4,1 both use a split primary/secondary SMC setup (one chip on the logic board, one on the CPU card). To my knowledge, Apple has never released a flashing application capable of programming both SMC chips as the only EFI utility we have (SMCFlasher.efi, also known as "smcutil") was designed for the single SMC systems (namely the MacPro1,1, 2,1, and 3,1).
-SC
