OcuLink Expansion Bay Module

Fully agree with having a more costly option with all bells and whistles. Eh… needless to say, that is essential for me. Don’t exactly want to fry my trusted GPU or brandnew FW (of which I will only have one ;)). So I am putting out my strong customer preference for high-end version which may be more expensive.

If I wasn’t clear enough, I would never put something in regular use without ESD protection, but a retimer is not for ESD protection - though it can do that in absence of other protections.
If my experiment fails, it will only be a PCIe Gen 3 board, but it will be a safe PCIe Gen 3 board.
It’s just in my current design there is no ESD protection, I was fine with that to verify the design itself works, since I was planning on populating it myself, and 0402 caps don’t lend themselves to that very well. I’ll reconsider.

Using a retimer is the sensible and safe option. Its not easy to calculate loss over all these non-standard connectors after all. But we’ll see. I wouldn’t be surprised if only the lower four lanes work fine at PCIe Gen 4, the upper ones that are squeezed on the corner might have too much crosstalk. If that’s so, I’d just redesign it in 6 layers, which I will likely do anyway since I do want the option for a mux.

3 Likes

For anyone curious about the difference between redrivers and retimers, Ti has a great video.
When to use a PCIe retimer vs. redriver | Video | TI.com.

4 Likes

Is it actually known that the Phoenix APUs can bifurcate their x8 link down to x4x4? Because some of you seem to be assuming just that.
All previous, single-Die Ryzen APUs could not do that.
The mobile versions all seem to have the the same x8+x4+x4+x4 layout with the last x4 being bifurcatable down to x1 for WiFi etc. Whereas the desktop versions are typically described as x16+x4+x4, with the x16 port only bifurcateble into the same x8x4x4 the mobile versions seem limited to…

Valid arguments, but I think it’s safe to assume they do support bifurcation natively and Framework explicitly supports it in the interface.
For once, they already have an existing, if flawed, dual SSD design without a separate bifurcation chip, and they also provide alternate pin assignments for two separate refclocks, which would only make sense if the 2 x4 lanes are bifurcateable natively.

2 Likes

Good point, I knew of the 2x NVMe example, I just did not think to check it for a PCIe-bridge…

But wait, if what you’re saying with the x8+x4+x4+x4 split is true, that means they chose to implement a bifurcation chip for the two onboard nvme drives using the x8 lanes, and leave the x4x4 for the expansion bay, despite it only being used for GPUs for now. Seems they really did plan ahead to leave the extra cost of the bifurcation chip on the motherboard to make expansion card designs easier on users. If that’s true, that is quite the foresight, and pretty cool

OK thats interesting indeed, and would talk for using 4x connectors, considering there is a dock that can still do 8x with 2 connectors…

Retimers are indeed good idea - I’m myself also not too familiar with PCIe, but looking the datasheets makes sense. They won’t help with ESD though, you’re going to still need low-capacitance TVS arrays in there. Bypass caps are not enough.

1 Like

What’s meant by it having Native” bifurcation, like does the Framework Laptop have a PLX/PEX MUX Chipset embedded into the motherboard?

Would add cost maybe but also add more functionality to third-party modules that could come to rely on it.

1 Like

Bifurcation is a feature of the CPU, since that’s where the PCIe root complex is located. No chip needed.

And as far as I know there is no such thing as a “bifurcation chip”, since bifurcation specifically refers to the ability to natively split ports of the CPU. The chip solution would use a switch.

Read about the USB 2.0 A “Hub” Ports a little way up. Possibility to have 2 internal + 2 external? For hiding mouse dongles and such while not using up an expansion card slot.

2 Likes

That could actually be pretty cool, a recessed usb 2.0 port on the oculink 8i port to keep those little dongles flush with the back plastics

@Seneral , I understand that you are the “developer” in this thread. do you have a timeframe or ideas of when and how to make the design or even boards available?

I’m not really, just the only one with (incomplete!) open source designs.
I’ve recently focused on my main project again so haven’t worked on this further, but I have fixed some gripes with KiCad in regards to custom rules (which is basically a requirement for 6-layer boards), though there’s still one more bug to fix before it’s usable without going crazy.
So as soon as I find some extra time somewhere I will attempt to recreate my current 4-layer design as a 6-layer board which will be a lot better and easier to work with going forward, including 8i port instead of dual 4i port, USB hub, and potentially PCIe redriver/retimers.
Ofc I invite anyone that wants to do an open source version to do just that. I can share my 6-layer template and fixed KiCad build for anyone willing.

FYI here’s the KiCad improvements

8 Likes

Sounds absolutely amazing - l truly hope someone in the community will build it.

2 Likes

Would something like this ever be available on the FW marketplace for the 16? I’d really love to have one and I dont have a 3D printer for the back plate.

I think the state of play is that framework staff said here it is something they will “probably” not build themselves, so for now I would bet we have to rely on the community. But framework is definitely aware and very supportive of developments like this, so let’s see.

PS: My money is on Josh and Seneral, fingers crossed.

2 Likes

Would you provide the model for the back of the expansion bay that includes the ports?

I haven’t made one yet, it’d need some guesswork too (last time I checked). Feel free to try.
As for my board, I’ve not yet decided on a final port layout. I won’t be able to work on it much for the next month either, sadly.

4 Likes

@Seneral I’m interested in doing a production run of this thing once its more polished. Would love to put work into this project in that way.
Cant help much with the design though and hope you find time for it :slight_smile:
I think it would be a test run to verify (5-10 pcs) and production runs (batches of 20-50?) following. First I have to find out about selling requirements, tax stuff and so on…
Btw, I’m based in Europe. Any pro tips on this?