As shown, this is actually three drives showing one namespace each, which is either the QLC part or the Optane part, due either to reversed PCIe lanes in this slot or to the H10 drive being a different hardware revision.
You should get 2 separate PCIe devices with H10
For an Optane H drive to work as expected (Optane cache to QLC storage) you need an Intel 11th gen to 13th gen platform and the right version of the RST driver. A very simple procedure…
A programmable PCIe switch would actually be the solution to access both sub-devices.
But we actually do not have a description of the hardware, so I’m not sure whether that is the issue. Are the H20 attached to these ASM2812?
This little ASM2812 6 port / 12 lanes looks somewhat exotic… Is this some kind of flash-only NAS (Asustor Flashtor or the like), serving multiple M.2 from the limited lanes of an Intel N100?
The ASM2812 is being used on this old supermicro server < i mean old> A bios update won’t fix this. I was going to replace with a new motherboard until ddr4/5 ram prices went thru the roof. This is a bandaid until I can get something with a gen2 xeon that knows what optane memory is. the motherboard only has x8 slots so I found this controller for $80 on ebay and it works for what i was doing. I did just get a dell 7920 for my kid, though i may use it for a new truenas box since it has the right cpu, still need memory though..
That’s for it to show up as a single logical volume to a guest OS (hence the need for the RST driver) - we actually want two distinct volumes here as ZFS would do a way better job of caching.
If I recall correctly the first-gen H10 cards would work on select Coffee Lake systems as well. Bifurcation and lane reversal was the problem as you’ve noted.
A fully programmable one with fully configurable bifurcation yes.
Problem is that the switch in question needs to bifurcate the physical M.2 socket down to x2x2. Most of these devices only go down to single x4 in each physical slot - similar to how a motherboard x16 PCIe slot will sometimes go down to x8x8, x8x4x4 or x4x4x4x4 - but you probably won’t see x2x2x2x2x2x2x2x2.
Fair point. But we still do not really know what hardware is used here (beyond that it’s a really old SM) and we do not know what OP intended to do (“cache”? what’s that intended to mean?).
And whatever the purpose is, the short answer has to be that it’s not doable: Optane H is a very fine unworkable engineering mess.
my box has had a variety of PCIE based nvme cards some switched and some not, my findings is based on what version of the linux kernel an OS has, how your BIOS handles bifurcation and signit issues from switch chips and what devices you combine behind the switch OR straight bifurcation it is a crap shoot as to whether either approach works witn nvmes / or real optanes. And god help you if you are using MCIO to connect any of this….
