Other "Ambitious" (read: likely dumb) idea for bifurcating riser cable

[Valantar]

I've been a bit annoyed by the proliferation of dual NVMe Intel ITX motherboards and the lack of the same on the AMD side (logical due to the PCIe layout, but still annoying). So I got an idea, which is more than likely both unrealistic and uninformed (I like to use the word "optimistic"), but I'd still like to toss it out here as it seems people on this forum have both incredible technical knowledge and surprising resources to bring niche devices into existence: since (AFAIK) both X3/470 and B3/450 support bifurcation, could a riser cable be made using a thin flex ribbon cable (think: the cables used inside of smartphones) that wraps under/around the last 8 lanes of the PCIe connector of a GPU while it sits in the motherboard slot, thereby allowing for either two GPUs (don't know where the 2nd would go, unless you go single-slot with water cooling) or an NVMe SSD, but not requiring a custom case (at least for the latter)?

I suppose the questions here are: how thin can a sufficiently shielded flex cable be made (I'm guessing it'd need at least two layers due to the pin density of the PCIe connector), and are PCIe slots wide/"roomy" enough that you could fit it between the card connector and the slot pins/sides without breaking the slot? Does lane bifurcation require additional communication with the motherboard, beyond simply attaching a new device onto the now-separate lanes? Would, say, the PRSNT2# pin of the GPU in the slot (B81) need to be rerouted to the PRSNT2# pin for x8 cards at B48, or can the GPU handle that itself (I kind of assume the latter, given that GPUs work in x8 (electrical) slots). Is there anything else I'm missing? Given that the latter 8 lanes only have one presence detection pin, am I correct in assuming it wouldn't be possible to further split these lanes into two SSDs?

If I'm not explaining this harebrained idea quite clearly enough, I made a quick sketch:

 

[Duality92]

I understand what you're saying, but a PCIE device has pins on both side of the PCB, which makes things considerably harder since you'd have to use two cables like you're referring to. I think it might work, if done with two cables, but it'd be a one time insertion as you'd probably deform pins in the PCIE socket too far that they wouldn't do proper contact with a regular PCIE device PCB after.

After that, we don't even know if it would work.

 

[aquelito]

I'm not sure you could make a proper riser without a PLX chip and other components.
Which means a PCB

Your idea is pretty interesting but it would be much easier to just modify a case. Or design your own.

EDIT : OK. You thought of this like an extension with a PCB at its end.

 

[cleveland]

If I'm not explaining this harebrained idea quite clearly enough, I made a quick sketch:

From left to right, the first "x8" portion" of your schematics is not electrically identical to the second "x8" portion.

 

[Valantar]

I understand what you're saying, but a PCIE device has pins on both side of the PCB, which makes things considerably harder since you'd have to use two cables like you're referring to. I think it might work, if done with two cables, but it'd be a one time insertion as you'd probably deform pins in the PCIE socket too far that they wouldn't do proper contact with a regular PCIE device PCB after.

After that, we don't even know if it would work.

Nope, you wouldn't need two cables. That's why the flex cable folds around the GPU's ... foot? connector? thingy? in a "U" shape. It can then have contacts on both sides quite easily, as it would be the same side (outer/bottom) of the flex cable (and flex cables with surface contacts or multi-layer flex cables are definitely not a new invention). Still, a picture says more than a thousand 68 words:

But you're entirely right, we have no idea if this would work. Which is why I'm trying to figure out if some of the brilliant and resourceful people on this forum might have some feedback, or want to do something with this idea. I know I have neither the resources nor knowledge to do so.

I'm not sure you could make a proper riser without a PLX chip and other components.
Which means a PCB

Your idea is pretty interesting but it would be much easier to just modify a case. Or design your own.

EDIT : OK. You thought of this like an extension with a PCB at its end.

Oh, of course, definitely with a PCB of some sort on the end. It just depends on what it would connect to. For m.2, the logical thing would be a tiny board with an m.2 slot and the smallest possible power connector (berg connector? ugh! but might be necessary). For a second GPU, it'd need an x16 or at least x8 slot, obviously, and it would need a PCIe 6-pin power connector powering the slot.

But a PLX chip shouldn't be necessary for use with motherboards that support PCIe bifurcation. I don't know (at all!) how this works, but at least with all the commercially available PCIe x16-to-any-number-of-m.2 risers, they don't have PLX chips on them (though they seem to have stuff to help with power delivery). That is how I've understood this, at least. Intel MSDT doesn't support bifurcation, so for them, these cards are limited to X299, but they should work on AMD boards. (Please correct me if I'm wrong here.)

Also, designing my own case won't help me add more m.2 to a board that only has one slot Besides, if this is possible, with the tiny size of m.2 drives (and a long enough flex cable) this could work in pretty much any case.

From left to right, the first "x8" portion" of your schematics is not electrically identical to the second "x8" portion.

Wikipedia: PCIe Pinout. Besides the lack of even rudimentary googling, how much weight are you putting in a 1-minute pencil sketch from someone who freely admits he doesn't have the slightest clue what he's doing? Still, even I could figure out that there are 49 pins per side for the first x8 and 33 for the second x8.

Also: "schematics"? That thing there?

 

[Thehack]

This won't work because the tolerance for the board and the connector won't fit anything worthwhile to transfer data over.

It'll be simpler to to just create a daughter board (PCB riser) and have a second PCIe connector split off of that.

 

[Valantar]

This won't work because the tolerance for the board and the connector won't fit anything worthwhile to transfer data over.

It'll be simpler to to just create a daughter board (PCB riser) and have a second PCIe connector split off of that.

Simpler? Well... that won't let you use a standard case, which was the whole thought behind this idea. I wouldn't call a custom case or modding new PCIe slots into your case 'simple'. Besides, risers like that aren't exactly a new idea?

But you're sure of that? Even if the end of the flex is kept thin enough to fit (i.e. as thin as possible while retaining the requires number of lanes), but the rest (from as close to the slot as possible) is made thicker to add thicker traces and better shielding? Of course I don't know if variable-layered flex cables are a thing, but it doesn't sound impossible. On the other hand, keeping the wrap-around part separate from the riser cable itself would also be possible, given sufficient soldering skills.

 

[pavel]

Cross-talk issue will be real. There is a good chance that lines that were galvanically insulated will still be picking up some signal through kapton/FPCB