...
Unfortunately the IOMMU groups of the PCI-E slots aren't very nice:
- Slot closest to the CPU - it's in the group 1 (as seen below with the Quadro P400), so the same group as the CPU, so no go. Passing this through to an Arch Linux VM made the TX1320 M3 freeze and show a CPU error in IPMI > Component Status
- 2nd PCI-E slot counting from the CPU - it's in group 1 (so same as CPU, so also no go)
- 3rd PCI-E slot (PCI-E x1), this one has a separate IOMMU group (group 13)
- 4th PIC-E slot (PCI-E x1, black plastic), also has separate IOMMU group (group 14)
...
Just FYI, a little extra info, as I also still have that TX1320 M3 running:
The reason why the IOMMU group for Slot 1 and 2 are the same is pretty simple. Fujitsu used the 16 Lane PEG port for both of these slots. They configured the PEG as 2x8, making it possible to provide two physical slots. However, that is the reason, why it is the same IOMMU group.
Also might be interesting for some:
Slot 3 is a PCH port and can be further bifurcated. If you have full access to the BIOS flash descriptor (i.e. Programmer ready and direct flash access). This way I had an Intel Optane H20 drive in PCIe 2x2 config running in their.
Another thing to note:
Looking for an alternative for the original 4x SAS Backplane, I found, that the Fujitsu "A3C40173252" RX2540 8x SAS backplane fits perfetcly into the TX1320 cage. It needs a modded power cable, as the TX1320 M3 HDD PWR connector provides 4x GND, 2x 12V and 2x 5V, while the backplane wants 3x 12V and 3x GND. However, nothing, that a little cable crimping or soldering can't solve.
And concerning the HDD cages:
HP caddy "PN 371589-001" seems to fit and is available for cheap. However (!) one has to remove the clamping mechanism's lever to make it fit. The drives connect firmly enough without the lever, in my humble opinion, anyway.
Edit:
Another quick hint, for those who do not know, yet.
With the help of the Coffeetime utility, one can even run CoffeeLake QuadCores on the board. However, as for the PCIe bifurcation topic, this requires a hardware SPI programmer. Already had an i3-8100 and i3-9300 running on the board, without any issues. For the bigger Hexa- and Octa-Cores more modifications on the CPU are necessary, which is why I refrained from trying that.