(And although I couldn't find any reference that showed that this device has a PLX or other PCIe switch - it really needs a switch to function. And a switch can dynamically allocate bandwidth according to demand - each of the four slots can get the 500 MB/s if the other three are idle. "Splitters" are for 12 volt DC power cords - a serial packet-based network like PCIe needs more intelligent switches.)
It has one, likely Pericom as these are cheaper for multinode things that do not base on 'splitting' but rather on 'sharing'.
Same stuff you get for miners on ebay nowadays, they have as i noted issues in Macs often.
Correct me if I am wrong, but a single 4-lane PCIe slot on a 2010-2012 Mac Pro has a throughput of roughly 2,000 MB/s, not 500, right? (500 MB/sec per lane, x 4.)
Correct. The SLOT has x4, however the Amfeltec GPU switch uses only an x1 uplink.
My splitter uses x4 and splits to 4x dedicated x1.
The Squid board uses x16 and splits to 4x x4 but also has the ability to use x4 uplink and still share x4 with both ports.
We discussed all these things two years ago but after that it became clear that trying to modify the cMP to make it stay relevant was expensive, unstable and bug filled. It's cheaper and safer to make a Hackintosh or use Thunderbolt devices connected to a MBP.
Depends i guess? If you have fun with it, why not. This is a expansion kit on top of a 5,1 - It uses the x16 slot to power 2 x4 SSDs and 2 GPUs at x4 each via a Squid board, and splits slot 4 into 4x x1 for USB and other cards.
This totals to:
- 1 x1 1.1 expanded from Airport
- 1 x4 2.0 full height in tower usable (slot 4)
- 1 x16 full height in tower usable (slot 1)
- 2 M.2 x4 spots external (Squid connected to slot 2)
- 2 x4 slots external (Squid)
- 4 x1 slots external (Splitter connected to slot 3)
https://prnt.li/f/82298dc9c4fc1dcc8fa5c0e688c17ecb-mu7dohd0ei.jpg