OcuLink Expansion Bay Module

I think that this one might be better for the oculink expansions that @Seneral and @Josh_Cook have been developing. I believe that, for the one I linked, the PCIe drivers on the framework will directly connect to the gpu, so it should be whatever both devices support. From what I understand, the expansions just expose PCIe lanes through oculink. Please get a second opinion before buying, I am definitely not an expert. Also, everyone please correct me if i’m wrong.

1 Like

@Philam

Correct me if I’m wrong, but it doesn’t seem @Philam wants the SSD for the SSDs sake (at least not externally), but as an option for OcuLink? Either way, that second link seems incapable of PCIe Gen4 speeds, with two ports (and no actual switch especially) I suspect the unused SATA connector would create a TON of reflections, essentially destroying signal integrity. At least from my limited understanding.

On a more general note, maybe @Josh_Cook can confirm, but he has hinted before that he’s also working on a complementary dock. Whether the benefits are mainly price or guaranteed Gen4 compability, or whether it will exist at all, I don’t know.

As for the first link, it MIGHT be able to run Gen4, but advertising it as such is stupid since it has no redrivers and many connectors. So depending on what M.2 slot you put it in, you might just be able to get Gen4 (which might explain if you heard that it does support it), but it will by far not be guaranteed.

For example, my planned test board will try the same - no redriver for now, just fingers crossed and hope the driver in the framework is good enough to work on Gen4 speeds with all those connections (really uncertain, especially since it’s designed for laptops and small distances, not like a desktop that expects long traces already).

Trying to use that board in a potential dual SSD expansion bay would further those issues even more, since you’d be adding another connector into the mix (and I doubt the dual SSD board would add a redriver just in case you add a long cable instead of an SSD). The built-in SSD slot might be worth a try if you’re really desperate. I’d just wait for @Josh_Cook’s solution

1 Like

Thanks for the info. I recently read a post on EGPU.io where someone is using oculink with the Chinese domestic market Lenovo 14+ and even there he runs into some signal issues… So I agree, my “plan” is to wait for the community solution here.

And yes, that is correct, I ONLY care about oculink ;).

I’ve had a few shower thoughts this morning.

The one dock and similar supporting thunderbolt and oculink must just be muxing the pcie input with a priority for one or the other. That should also mean it would be possible to make a thunderbolt to oculink adapter (at similar reduced performance).

One thing that frustrates me today is switching between my work setup and home setup as my work Mac only has thunderbolt obviously so I have a convoluted kvm setup.

In theory I would think it could be possible to build a oculink switch. Using a thunderbolt to oculink adapter from the Mac and oculink from framework a single egpu could be shared. It would probably need a bunch of redrivers and a delay or something to reset the GPU side before attaching to the other computer.

Then just a push of a button on my desk all my displays would switch computers and offload some graphics processing from the power limited work Mac.

Plug m.2 to occulink adabter in a tb to m.2 enclosure, done. Probably even relatively compact.

The PCIE mux is probably the more difficult part there, they exist but I don’t know of any off the shelve solutions that aren’t part of a main-board.

Also occulink is still not hot-pluggable on regular hardware so the computer with the direct connection would probably need a reboot or be powered off at the time of switch over.

I’m not well-versed in the consumer side of oculink, or even on the server-grade stuff I’m a bit more familiar with, but it’s interesting to note that slimSAS or oculink is falling in favor of MCIO aka Mini CoolEdge IO connectors these days, interesting little discussion here: Wendell from Level1Techs on Gamers Nexus

1 Like

The beauty of doing it in an expansion bay is that you can just use a different one if suddenly a vibrant oekosystem of mcio (or whatever pcie over cable connector shows up next) devices shows up.

Hell maybe someone actually ends up defining an actual standard for external direct pcie (not tunneled like tb/usb4), maybe even with hotplug at some point.

3 Likes

First shipments are soon on their way to us

Any updates regarding the module @Josh_Cook ?

1 Like

Haha, I JUST wanted to write exactly the same post… :wink:

1 Like

MCIO looks interesting. The connectors appear to at least be available in quantities fewer than 2000.

Haven’t looked to find the signaling spec. I’m assuming hotplug is still non-existent.

1 Like

If anyone is actually still working on the module (I heavily doubt it now), I hope they are collaborating with Framework to put it on the market place - this would be really amazing.

2 Likes

I got the impression @Josh_Cook has intentions to sell his version, so I suspect he’s either still working on it or is already done (since testing boards have been made long ago), but I would also appreciate a status update so people can plan ahead.

3 Likes

I put my entire money on the community here ;). If noone in the community is able to do it, I am out of options except booting from the 2230 ssd and using an oculink adapter while keeping the lid open… That’ it. But I wanted to go for a revolutionary, community based approach here. I am all in. Let’s see, maybe Framework will jump in.

PS: the alternative would be to wait for the new oculink TGX Egpu together with the Lenovo notebook, but I am sticking with FW16 no matter what I think. Batch 5, so still have some time ;).

1 Like

I could resume my PCB efforts, but I’m not PCB designer by trade, but a software engineer.
I’ll at most be able to get a simple design that might only be PCIe Gen 3 - and since I’m busy with another project, I haven’t found a reason to continue with Josh’s design presumably coming with retimers (which are most likely necessary for Gen 4).
There’s always merit in an open source design, and even a simpler design, so I intend to eventually continue it - might have to expedite that if there’s no word from Josh.

2 Likes

Thank you, that would be amazing. But in any case, I for example would certainly need Gen 4 for higher speeds than USB4 which the laptop already has.

However, I consider myself probably the last person in the community to try to build any “pressure” for anyone in the community on this. Where I am coming from is simply that I think it would be awesome if we had a community solution - I would love it and agree it has always merit.

So I went “all in” on this, this thread here is the reason why I am on this forum, and the perspective of having oculink is a bit the prerequisite for buying the FW16 for me. One Dock V2 is on the way from China, FW16 is ordered in batch 5, ssd and Ram are already here ;). So it’s a bit the last piece in the puzzle.

So thanks a lot for your efforts and let’s see if we get an update from anyone else or Framework as well. I believe in the community and the company, so my bet is that either one will deliver. I also think quite a lot of people are interested in this, and it would be an oversight by Framework to not help out here eventually if there is no community solution.

8x pcie3 is already about double of best case usb4

I would take anything ;). But with One Dock v2 at least, I think I can only rely on 4x, so would need Gen4.

Or maybe it would depend on the cable? This would be really awesome if it could work with 8x Gen3.

Yeah you aren’t turning 8x pcie3 into 4x pcie4 cheaply, those darn pcie switches are still stupid expensive and at higher pcie revisions continue to get more expensive.

Yes, this is apparently the “flaw” of one dock v2, but I ended up choosing it for easy portability, and as I ordered a long time ago and wanted to commit to it.

The creator of one dock may also be able to help out, I already asked once, and will leave it for now and wait for the community here first ;).

PS: I had a quick chat with the creator of one dock v2, it has redriver chips for enhanced signal integrity, so according to the creator of one dock it should work well with the 2280 ssd connection for stable oculink. Maybe you could look for a specific oculink cable option you prefer, latest developments can be found here: https://egpu.io/forums/custom-egpu-chassis/oculink-what-it-is-why-you-need-it-and-where-to-get-it/paged/22/

So this is just for information in case anyone would like to use oculink already now. This is my plan, boot from the 2230 ssd, and use the 2280 spot for oculink connector. FW16 is exceptionally well suited for this, no need to remove or cut into any backplate. Just open it, and game on ;). Should be possible in 2-3 minutes as setup.

So I will be using oculink connection once I get it, and of course keep hopes that community or Framework can add any other solution, if possible at any time…