I acquired my Mac Pro 2006 1,1 (A1186) in November 2024 (IRRC). Despite being nearly twenty years old, it appeared to be in remarkably good condition and was fully functional when I received it. The only obvious issues were some minor cosmetic damage to the case, likely caused by the machine's considerable weight (over 25kg), and a faulty optical drive door that would no longer retract properly.

The system came equipped with an impressive 32 GB of RAM and all four hard drive bays populated, providing a total of 4TB of storage. As I had no particular use for the original hard drives, I sold them and cloned the main system disk to an SSD instead (you'll need a suitable 2.5" to 3.5" drive adapter for this).

My intention was to repurpose the machine as a Linux workstation and run Einstein@Home on its dual Xeon processors. Unfortunately, the project quickly took an unexpected turn.

I was unable to boot a Linux installer from USB, so I installed an SSD containing an existing Ubuntu installation from another computer. Ubuntu booted successfully, but the original Nvidia GeForce 7300 GT displayed severe graphical corruption, either due to driver issues or poor compatibility with the desktop environment. While experimenting with alternative Linux distributions and desktop environments, the system suddenly crashed.

After the crash, the Mac Pro would no longer start. Pressing the power button produced no startup chime and the machine failed to boot. Prior to this, I had occasionally needed to reseat the RAM riser cards to restore normal operation, so memory-related issues were among my first suspicions.

My initial troubleshooting steps included checking the motherboard diagnostic LEDs. When the power button was pressed, only the Standby (Trickle Power) and Power Good LEDs illuminated; none of the CPU- or temperature-related indicators were active. I also reset the CMOS and replaced the PRAM battery, but none of this made any difference.

At this point it was clear that further investigation would require dismantling the machine. What initially appeared to be a straightforward fault would eventually lead to a much more extensive repair than I had anticipated.

Dismantling:

[Disclaimer: The information presented on this page is based on my experience,
I am not responsible for what you do to your system.]

Before proceeding, it is worth noting that there are at least two internal variants of the 2006 Mac Pro. Mine is the version shown on the left below, identifiable by the arrangement of the RAM riser slots.

To gain full access to the lower are of the system that house the RAM and CPUs, the RAM cards can be pulled out, then the five screws shown in the image below must be removed to unfasten the housing itself. Four of these become accessible once the RAM riser cards have been removed, and removing the graphics card beforehand makes the process considerably easier.

The two lower screws caused me particular difficulty. Research suggested that this is a fairly common problem, and in my case both screws appeared to have been installed with thread-locking compound. The heads stripped almost immediately when I attempted to remove them.

To overcome this, I carefully cut slots into the screw heads using a rotary tool, taking care to vacuum away any metal particles as I worked. Once modified, the screws could be removed with a flat-bladed screwdriver. The standoffs beneath the screws may come out as well. This part of the job required patience, but eventually the screws were freed without causing any damage.

With the screws removed, the RAM housing becomes loose and the internal shrouds can be separated. The front intake fan section is particularly awkward, but can be carefully worked free once its retaining screw has been removed. The fan assembly connects directly to the motherboard via an integrated multi-pin connector, so it effectively disconnects as the shroud is removed.

At this stage, it is possible to access and clean the front fan and the leading edges of the CPU heatsinks without fully dismantling the system. Given the amount of dust present in this area, this alone is highly recommended for any second-hand unit.

Shroud and front intake fan removed to reveal dust.

To continue further, the CPU heatsinks must be removed. This requires a long driver (approximately 20 cm) to reach the eight mounting screws. Although a 3 mm hex driver is specified, I found that a T15 Torx driver was more practical and widely available.

Each heatsink includes a temperature sensor attached near its base, which must be disconnected from the motherboard before removal. It is also important to note the orientation of each heatsink, as the rubber pads differ between positions.

Once removed, I proceeded with CPU replacement and reinstalled the cleaned heatsinks. On first power-on, the system chimed and appeared to function normally.

However, this success was short-lived.

After shutting the machine down and reconnecting the SSD, it failed to boot again. The fans immediately ramped to maximum speed and there was no startup chime. At this point, the fault remained unresolved, and both CPU and power delivery issues were still under consideration.

Given the uncertainty, I began investigating the mainboard as a possible cause. A replacement board was sourced as a diagnostic step, since sourcing another power supply at reasonable cost was not practical at that stage.

When it arrived, the condition of the packaging did little to inspire confidence. The board had been shipped with minimal protection, and could be felt through the outer wrapping.

Some people seem to lack any understanding about how computer parts such as mainboards, graphics cards, or RAM should be handled, or how best to package them. Typically such things are static-sensitive, but not even putting an item like this in a box... come on.

Despite my doubts, I proceeded with installation. The Mac Pro logic board replacement is a relatively involved process due to the number of tightly packed connectors and limited clearance within the chassis, but the swap was completed successfully.

The original board was tested once more before removal and briefly appeared to function, further adding to the uncertainty. Based on this, I suspected an intermittent fault possibly related to power delivery.

The replacement board initially failed to produce a startup chime, reinforcing the suspicion that the underlying issue was not resolved. However, when tested again the following day, the system did start successfully.

 At this point the behaviour became inconsistent. The machine would occasionally boot and operate normally, but stability was not reliable. After further testing with the Linux installation and workload stress, the system eventually failed again in a similar manner.

This pattern strongly suggested an upstream power instability rather than a single permanently failed component. By this stage, the power supply had become the primary suspect once again.

Delving into the Power Supply:

With the processor theory and suspected mainboard issues becoming increasingly unlikely, my attention returned to the power supply.

Finding a replacement unit proved difficult. Apple produced multiple variants for the Mac Pro 1,1, including differences in connector layout and internal design. Some versions appear to use a single high-capacity 12 V rail, while others, like mine, use multiple 12 V rails (in this case six). The multi-rail design is less appealing for potential upgrades, as load distribution is not clearly documented and there is a greater risk of unintentionally overloading individual rails.

As is often the case with multi-rail PSUs, it is not obvious how the internal components are mapped. In this unit, six rails appear to be present but only four output connectors are exposed, with no clear labelling of their distribution.

Before proceeding, a word of caution: power supplies can contain potentially dangerous voltages even after they have been disconnected from the mains. Do not attempt to dismantle or repair one unless you understand the associated risks and know how to work safely.

The Mac Pro PSU is secured within the chassis and connected via four cable assemblies. Once these are disconnected, the unit can be removed and split into two sections.

From my understanding, the right half deals with the AC input and the left half provides the DC outputs.

My previous experience repairing older electronics often involved replacing electrolytic capacitors. While not a precise diagnostic method, degraded capacitors are a common failure point in ageing power supplies, particularly on output rails where thermal and electrical stress is highest. Dust accumulation can further accelerate this ageing process by reducing cooling efficiency and increasing operating temperatures.

In this case, I identified a cluster of six capacitors on the 12 V output side that appeared to be the most likely starting point. None showed obvious signs of failure, but visual inspection alone is not always reliable.

My usual method for replacing through-hole capacitors is slightly unorthodox. Rather than desoldering the entire component in one operation, I first remove the capacitor body, leaving the legs in place. These can then be desoldered individually after applying flux from the reverse side of the board. I find this approach easier when dealing with large, tightly mounted components.

On this occasion, the repair initially went poorly. As each leg was removed, solder reflowed into the holes, making it difficult to seat the replacement components cleanly. After several attempts and increasing frustration, I set the project aside.

Several months later, after gaining more experience with similar repairs, I returned to the PSU.

This time I changed approach. I shortened the replacement capacitor leads to approximately 5 mm and applied a generous amount of thick flux. Each capacitor was held in position while heat was applied from the underside of the board, gently working the joints until the softened solder allowed the leads to pass through and the components to seat correctly.

Once all six capacitors were replaced, I reflowed each joint with a small amount of fresh solder, trimmed the excess leads, and cleaned the board thoroughly with isopropyl alcohol.

Given the difficulty of the original repair, I had little confidence it would succeed. I expected either immediate failure or no response at all from the system.

Instead, the Mac Pro powered on.

A moment later it chimed.

Shortly afterwards, it booted into the operating system.

Most importantly, it remained stable under operation.

At that point, it was clear that neither the processors nor the mainboard were the root cause. The replacement CPUs remained installed, but the fault had almost certainly originated in the power supply. Whether the capacitors I replaced were definitively faulty is difficult to prove without further testing, but the repair restored the machine from a non-booting state to a fully functional system.

Installing Linux:

The next step was getting Linux running reliably on the Mac Pro 1,1.

This model presents a well-known challenge: although it uses 64-bit Xeon processors, it is constrained by a 32-bit EFI firmware implementation. As a result, many modern Linux installers will not boot directly from USB in the usual way.

I attempted to work around this by trying a range of Linux distributions, including both newer and older releases, but none of them successfully booted from USB on this machine.

In the end, I took a more indirect approach. I installed an SSD in another Intel-based system and used that machine to perform the initial Linux installation. I chose an older Ubuntu release that was compatible with the Mac Pro’s graphics hardware, as later versions and some alternative distributions proved unreliable when moved across.

Once installed, the SSD was transferred back into the Mac Pro. Although this approach felt somewhat like a workaround, it proved to be the most practical way of getting a working system in place.

With Linux finally running, I turned my attention to installing BOINC and setting up Einstein@Home. The system initially struggled with outdated software repositories, but once those issues were resolved I was able to install BOINC and begin testing workloads on the original dual-core Xeon processors.

At this stage my plan was to measure system stability and power consumption before moving on to the quad-core Xeon upgrade I had already purchased. The goal was to compare performance and efficiency between the original and upgraded CPU configurations under sustained computational load.

I installed Psensor, a simple Linux hardware monitoring tool used to track system temperatures in real time. This allowed me to monitor CPU and system thermals under load while running BOINC workloads, giving a clearer picture of how the ageing cooling system and Xeon processors were coping during sustained computation.

Swapping the Xeons:

Swapping the Xeons marked the point where I could finally push the system a bit further.

The original processors in my system were a pair of dual-core 2.0GHz Xeons, and after researching compatible LGA771 processors, I purchased a pair of quad-core Xeon E5320 CPUs. They were inexpensive, readily available, and offered twice the core count of the originals, albeit at a slightly lower clock speed of 1.86GHz.

With the heatsinks already familiar to me from the earlier dismantling work, the installation process was relatively straightforward. With the plastic heatsink cover removed, and the heatsinnk towers themselves lifted out, the old processors were carefully replaced with the quad-core Xeons, and fresh thermal compound was applied before reassembly. Everything was reinstalled with particular attention to the heatsink screwes, being sure to not over tighten them.

Once reassembled, the system powered on without hesitation.

This time, however, I could immediately push it further.

With all eight cores now available from the upgraded CPUs, I loaded the system with Einstein@Home workloads through BOINC and monitored it under sustained load. Psensor was used throughout to track CPU temperatures and confirm that the cooling system was still keeping everything within safe limits, as I expected it would, given that systems could have been purchased from Apple with this setup.

Despite the earlier instability and failed boot attempts, the system proved to be completely stable under load. Both the operating system and BOINC ran without interruption.

At full load, the system drew approximately 345W from the wall, a useful benchmark for understanding its power consumption under sustained computational work.

What had begun as a machine that refused to boot had now become a stable 8-core workstation running continuous scientific workloads.

Upgrading the GPU:

With the system now stable under sustained load, my attention turned to another potential upgrade; the graphics card.

The original Nvidia GeForce 7300 GT is extremely limited by modern standards, and I had considered replacing it with something more capable. However, this immediately raised questions around power delivery. Many modern GPUs require additional 6-pin or 8-pin PCIe power connectors, and it was not initially clear how safely this could be achieved within the constraints of the Mac Pro 1,1 power design.

The system does provide additional power outputs beyond the standard drive connectors, but Apple’s internal distribution of power is not as straightforward as a modern PC power supply. While the PSU is often described as having multiple 12V rails, the way these are routed through the logic board makes it difficult to determine how load is actually distributed between components such as the CPUs, storage devices, and expansion hardware.

I briefly considered using the available Molex power connectors from the optical drive bay or tapping into the SATA power lines for the hard drive backplane. However, these are also integrated into the system’s internal power distribution path, and I was wary of placing additional load on circuits that were already supporting the dual Xeon processors and other components.

On closer inspection, I discovered that Apple had in fact provided a more appropriate solution: two dedicated mini 6-pin power connectors intended specifically for graphics cards. These offer a far cleaner and more direct way of powering a GPU upgrade without modifying the internal power distribution.

With this in mind, I decided to proceed with a modest and realistic upgrade target. After checking compatibility and expected power draw, I settled on aiming for a mid-range card such as an Nvidia GTX 1060, which represents a significant improvement over the original hardware without pushing the system into unrealistic power requirements.

I have since ordered the necessary mini 6-pin adapter cable and will revisit the GPU upgrade once I'm ready to try it out.