OK, so this is a crazy, off the wall idea... but what if it was possible to design and implement a replacement PCB that would fit the standard CPU tray and heat sink of a classic Mac Pro 5,1? If you don't mind me summarizing 557 pages and over a million page views from the "Waiting for Mac Pro 7,1" thread, the consensus is "Give us an updated Mac Pro like our beloved 5,1, with a modern supported processor and PCIe slots and Thunderbolt and NVME storage." What is the minimum that needs to change in a 5,1 for that to happen: the CPU tray, right?
It seems logical that the main guts of the Intel chip set (CPU, north bridge, south bridge, RAM) probably live on that CPU tray board; are the signals that go off-board to the back plane pretty much standard interface signals like PCIe and SATA and USB? Because if that is true, would upgrading the chip set on the CPU tray instantly upgrade those standard buses, too? PCI 2.0 becomes 3.0? SATA 2 becomes SATA 3? USB 2.0 becomes USB 3? Thunderbolt would have to come off the board via connectors, perhaps to a slot opening. I was thinking the low risk approach would be to implement the Xeon-W series of the iMac Pro in a single processor configuration, which would leave room for NVME storage on the CPU tray board. Basically ride the coat-tails of the iMac Pro as far as MacOS compatibility.
Here are some of the possible pitfalls I can think of...
?? Maybe the existing chip set is split across the interface (south bridge is on back plane?) - would be difficult to mate that up to a modern chip set if so.
?? The existing CPU tray may have proprietary glue logic chips on it.
?? Apple stores low level driver code in a proprietary flash, I've heard; is that on the CPU tray board (probably)?
?? The iMac Pro uses Apple-specific SKU models of the Xeon-W's; would the more generalized Xeon-W processors available to the public work? The hackintosh community probably know the answer to this.
?? The signals that go across the connector are mostly standard, but can't run at higher speed due to circuit length / signal integrity limitations?
?? The system management controller circuitry is incompatible with a later chip set, or can't be easily reverse engineered to connect to later chip sets?
?? Who would have the necessary engineering tools for simulation and layout of cutting edge circuitry?
?? Who in the community has anything close to the level of engineering knowledge for such a project? Anyone know any retired Apple engineers with a fondness for classic Mac Pro's?
?? Would macOS expect to find a T2 chip in any design that uses the same processor as an iMac Pro?
?? The process is possibly do-able, but would take so long when done by volunteers that the result would be obsolete by the time it was done.
Process
The whole process of creating such a board would take super-specialized knowledge: identify all the chips on the current CPU tray, come up with a physical model for the mounting points of the board and the heat sink, reverse engineer the current 5,1 circuitry (compare to vintage reference designs from Intel?), figure out all the signals on the connector, come up with a schematic for the later chip set that sends equivalent signals out on the connector (convince Intel to release reference design after signing a zillion NDAs?), lay out the new design, test and measure signal margins, iterate to fix bugs, set up a supply chain, arrange for contract manufacturing of what is probably a small run of a few thousand boards.
Would it make economic sense? How would you structure it to compensate those who would put in hundreds of hours of engineering time? Create a non-profit or a foundation to manage it? Release it all as open source to the community?
OK, shoot me down now. As the great analog electrical engineer Robert Pease used to say, "Show me where it says I (we) can't do this."
It seems logical that the main guts of the Intel chip set (CPU, north bridge, south bridge, RAM) probably live on that CPU tray board; are the signals that go off-board to the back plane pretty much standard interface signals like PCIe and SATA and USB? Because if that is true, would upgrading the chip set on the CPU tray instantly upgrade those standard buses, too? PCI 2.0 becomes 3.0? SATA 2 becomes SATA 3? USB 2.0 becomes USB 3? Thunderbolt would have to come off the board via connectors, perhaps to a slot opening. I was thinking the low risk approach would be to implement the Xeon-W series of the iMac Pro in a single processor configuration, which would leave room for NVME storage on the CPU tray board. Basically ride the coat-tails of the iMac Pro as far as MacOS compatibility.
Here are some of the possible pitfalls I can think of...
?? Maybe the existing chip set is split across the interface (south bridge is on back plane?) - would be difficult to mate that up to a modern chip set if so.
?? The existing CPU tray may have proprietary glue logic chips on it.
?? Apple stores low level driver code in a proprietary flash, I've heard; is that on the CPU tray board (probably)?
?? The iMac Pro uses Apple-specific SKU models of the Xeon-W's; would the more generalized Xeon-W processors available to the public work? The hackintosh community probably know the answer to this.
?? The signals that go across the connector are mostly standard, but can't run at higher speed due to circuit length / signal integrity limitations?
?? The system management controller circuitry is incompatible with a later chip set, or can't be easily reverse engineered to connect to later chip sets?
?? Who would have the necessary engineering tools for simulation and layout of cutting edge circuitry?
?? Who in the community has anything close to the level of engineering knowledge for such a project? Anyone know any retired Apple engineers with a fondness for classic Mac Pro's?
?? Would macOS expect to find a T2 chip in any design that uses the same processor as an iMac Pro?
?? The process is possibly do-able, but would take so long when done by volunteers that the result would be obsolete by the time it was done.
Process
The whole process of creating such a board would take super-specialized knowledge: identify all the chips on the current CPU tray, come up with a physical model for the mounting points of the board and the heat sink, reverse engineer the current 5,1 circuitry (compare to vintage reference designs from Intel?), figure out all the signals on the connector, come up with a schematic for the later chip set that sends equivalent signals out on the connector (convince Intel to release reference design after signing a zillion NDAs?), lay out the new design, test and measure signal margins, iterate to fix bugs, set up a supply chain, arrange for contract manufacturing of what is probably a small run of a few thousand boards.
Would it make economic sense? How would you structure it to compensate those who would put in hundreds of hours of engineering time? Create a non-profit or a foundation to manage it? Release it all as open source to the community?
OK, shoot me down now. As the great analog electrical engineer Robert Pease used to say, "Show me where it says I (we) can't do this."