what i am almost sure is that there will be some chinese cheap adapter to convert the mpx connector on whatever we need or to 8pin pcie power.
Cheap Chinese Adapter for what? The notion that there simply just one big proprietary MPX connector is hugely disconnected from reality. There is
a connector in the MPX bays that is so far proprietary, but it is not the only connector.
"
Each MPX bay provides:
x16 gen 3 bandwidth for graphics
x8 gen 3 bandwidth for Thunderbolt
...
Or two full-length, double-wide x16 gen 3 slots (MPX bay 2)
Up to 300W auxiliary power via two 8-pin connectors
..."
https://www.apple.com/mac-pro/specs/
that is
each bay; there are
two of them. So there are four 8 pin power connectors in the new Mac Pro. Funky MPX converters for what? Part of the higher base system price is having power delivered by two different paths. Optionally, putting the power on a 8 pin sources is already paid for when buy the system.
There is also a another 6-pin that isn't in the MPX bays that Apple's specs don't mention. ( How that might be hooked to an optional/alternative power feed is hopefully documented at some point. I suspect some of that 6-pin power is coming from the supposedly "missing" 200W that in the gap between the MPX card's 500W and the 300W of the two 8-pins. So probably isn't really missing. At least for MPX bay 2 (slot position 3-4). Nominally, there should be at least one MPX module to feed the default Thunderbolt port's subsystem so not being able to all of them with zero compromise shouldn't be a surprise. The system is skewed to provide for the ports available by default. MPX Bay 1 'hogging' the extra 200W incrementally tilts the playing field, but it is not a system wide skew. ).
The access to the 8-pins pragmatically disappears when you plug in a MPX module. It appears to be rather symmetrical if you try to use the slot 1 or slot 3 for standard PCI-e add in cards. Same thing that custom slot pragmatically disappears. ( since probably sharing a source feed that probably a good thing to keep folks from overloading the power supply).
Apple has design that is a more than reasonable alternative resource sharing system here. Is it primarily biased to the largest number of "max Power" add in cards possible? No. has the previous Mac Pro designs ever been biased toward being king of the "Max Power" add-in-card capacity market? No. This really is not a change in direction at al.
unless there are some actual harware limitation to this with a sort of t2 chip
it remain unclear for me if the actual mpx connector next to the standard pcie slot is also a pcie slot but in a proprietary connector.
The T2 has little to nothing to do with the MPX Bays at all. The T2 is highly likely probably PCI-e lane provisioned off the PCH (I/O chipset ) just like all the previous Macs that have contained a T2. It is not directly hooked the CPU or its PCI-e lane providing at all. Hence nothing to do with the MPX Bays.
The T2 is largely being used in this thread as a bogey man FUD tactic to distract from the facts in this thread than anything relevant; noise as air cover for misdirection.
The 'MPX connector' is probably far more relevantly to this threads initial discussion hooked to the "other" x8 PCI-e slot in the MBX bay. Pretty good chance those are both downstream of a PCI-e switch. Something like
CPU x16 ---> PCI-e switch -----|
| ---> x8 1a MPX 'connector'
| ---> x8 slot 2
| ---> x8 3a MPX 'connector'
| ---> x8 Slot 4
and perhaps also
CPU x16 ---> PCI-e switch -----|
| ---> x16 slot 5
| ---> x8 slot 6
| ---> x8 slot 7
| ---> x4 Slot 8
For the "alt x8" slot in the MPX bay somewhat similar with the power pins being covered/uncovered with the alternative use of the MPX Bay.. For the entry half height 580X module those the x8 bandwidth wouldn't get split because the card doesn't having any Thunderbolt Ports to provision ( that is non dilution by card edge design not so much by physical blocking) .
The sub connection that the MPX "connector" probably has power , x8 PCI-e , and 4 DisplayPort v1.4 streams out , and some possibly some thermal feedback control signal. The Power and PCI-e bandwidth are shared though; that socket isn't the only path for those.
[ A single MPX Vega Ii would leave one of the TB controllers un-provisioned for display out, but have display out controllers on the card. ]
The MPX module versus pure history PCI-e card playing field is tilted slightly in the MPX module directiion if trying to put four of the biggets GPU possible into the system. But for 1-3 GPU where all of them are not "maximum power consumption on the market" there is a very large number of configurations that fit with nothing more than just buying/using the appropriate power cables.
P.S. The sub part of the market that wants to go chance the "max power" cards connected to the system can buy PCI-e expansion box. Hook it to Bay 2' x16 and attach an enclosure with another 1.4KW power supply and slap the 4 "Max Power" cards in there. Same track was taken with some cMP deployment configurations where really just feeding a computation 'farm' of GPGPUs with modest incremental data swapping.
