A lot of people have been assuming that. That it’s like a docking port, that just secures with a latch. Available information shows it uses screws and may require some disassembly to even access the screws. See Framework’s github.
Links and pictures
Yep, will need to take out the keyboard and remove two screws to remove it.
If you look in the promo video, the keyboard is completely removed, then the bay changed, then the keyboard put on again. I thought that was weird before I looked at the sheets properly but it makes sense. Still doesn’t look like swapping will take too long but the durability may be an issue.
Here are the relevant data sheets:
Manufacturer: https://neoconix.com/wp-content/uploads/AS-000031-XBM-X-Beam-Product-Spec-Rev-D.pdf (this might not be the exact product Framework uses, but it’s the closest I could find)
The manufacturer says that it’s rated for 50 cycles (I assume this is for the release model of the connector) and disassembly definitely requires removing the keyboard and removing at least 2 screws, so it’s probably not something you want to do often. Yes, it looks like it’s more for upgrading or occasional swaps and less for “lets plug the GPU in when I get home”.
Yeah it’s closer to a an improved mxm format (or hell m.2 device) than an expansion card.
If you are thinking you can swap expansion bays like expansion cards, you are gonna have a bad time.
It’s the FX Beam connector.
Overview pdf
The FX beam connector data sheet doesn’t give a durability rating and says instead “Integrates X-Beam Connector”, so I think it is okay to look at the X beam data sheet. Or do you have another source of information?
I missed that. The lack of the 3 plastic pins on the X beam made me think there could be other differences.
Board to board connectors do usually have a relatively low number of specified mating cycles, even regular PCIe connectors do only have 50 (for example this), most m.2 connectors 25, fine pitch mezzanine connectors sometimes even only 10 (this one for example), the display connector on the framework-mainboard has 30 if I remember it correctly.
For comparison USB-C connectors usually have a specified durability of 10000 mating cycles.
So does this also “destroy” the dream of an oculink EGPU that is plugged in daily? I guess it depends on what some people will 3D print and cook up, right?
I assume it should be possible to permanently attach some kind of oculink connector and then simply plug in and remove the cable.
Not necessarily. In this case the expansion bay would be an adapter between the fragile X-Beam connector and the Oculink connector. So only the durability of the oculink connector matters. However from the data sheets I found, uculink connectors and cables aren’t rated for much more connection cycles. I found numbers around 250. Which means it’ll probably only last around a year comfortably if you really unplug it daily (there is obviously some safety margin in the rating). (Edit: ignore that, see next post, I’m just bad at searching)
The first result I got for OCuLink connectors is amphenol listing a rating of up to 10,000 mating cycles. https://www.amphenol-cs.com/product-series/oculink.html
Thanks, then the dream is real. If mating cycles of oculink are similar to thunderbolt 3/4, then it is high time we overcome thunderbolt speeds that are simply inappropriate.
So far the EGPU environment is a space worthy of supporting more, people are overclocking GPUs to get most of inappropriate bandwith, others are sawing holes in the bottoms of their laptops to get an oculink connection working…
My hopes are now with one laptop and one company to clean this up.
Asus has its own proprietary solution (for notebook EGPUs…), and I hope Framework will open such speeds to the wider public.
I was definitely planning to hot swap gpu and battery expansions. This is a bummer.
Actually, I wouldn’t get hopes up, at least for daily plug / unplug durability. I looked through some of their datasheets in hopes of finding harder numbers. Found some pretty low minimum numbers.
Their general datasheet for that G14 series of connectors only says “Durability: 50 mating cycles min.”
Another datasheet linked to on the page for a 8x oculink connector lists “Durability- Internal : Number of cycles: 50 minimum”
Thanks to MJ1 for this. This would change the entire picture for me and explain the added value of thunderbolt. I know that oculink comes from server environment and was back in the day meant as competition for thunderbolt. Alright, but if mating cycles are so poor in comparison, then the use of thunderbolt really simply makes sense…
I was trying to get my head around why noone pursued a technology like oculink more in the past, given that it is in terms of speed all that would be needed to achieve close to desktop performance.
It seems the picture is now coming together. Thunderbolt is simply incredibly versatile and durable - so then I would take the point of using it…
In any case, you are looking at a post of one of the guys who would simply replace the plug every half year ;).
PS: So, I keep my hopes up. Hoping for a combination of ModDIY and maybe ETSY or 3D printing and I will use oculink and a Chinese EGPU and replace plug if needed every half year - Yes, that’s the plan for December 2023. Let’s see if all this goes up in flames closer to release…
Where the hell did you get the idea that it was meant as competition to thunderbolt? They are pretty much entirely different things.
Occulinc was intended for wired pcie inside servers, think flash based storage backplanes (like sata/sas cables on steroids) or banks of accelerators and stuff like that.
It’s use as an egpu connector is more of a recent accident but given it is actually meant for pcie that’s an improvement over previous solutions that abused stuff like hdmi and displayport connectors for external pcie.
It kinda was it was (and still is) just incredibly niche. External direct pcie interfaces have been done in pretty much any way you can imagine.
Thunderbolt is meant for consumers, kinda like usb on steroids which comes with it’s pros and cons (pros: compatibility, hotplug, can output all kinds of things and is meant to actually be plugged in and out on a regular basis, cons: cost and overhead)
Given it may just be a passive adapter that could be quite viable, there is also the possibility of that connector living a lot longer depending on how rough you are with it.
@Adrian_Joachim Dude, please work on your language. To answer your question where I heard about it being a competitor, I read an article on EGPU.IO website yesterday in which this was mentioned, citation here: “OCuLink is a non-proprietary PCIe interface developed by PCI-SIG, the organization for standardization of the PCI interface and was designed to compete with Thunderbolt in 2012.” Link here: OSMETA GK01 OCuLink eGPU Review and Installation Guide | eGPU.io
Rest assured that I don’t just write stuff that I dream up in fora here. But please tone down the language, there are some people here that are just enthusiast consumers and don’t work in tech (in case you do). This forum should be a place where people can learn.
If you disagree with the statement, I get it. But that’s that.
@Philam Yeah if egpu.io who tend to make actually pretty good content make those claims I can’t really fault you for believing that. The statement still does not make sense but that’s definitely not on you. Kinda expected better from egpu.io though. Sorry if my frustration about the statement came over as frustration about you.
Saying occulink was designed as a competitor to thunderbolt is like saying a forklift was designed as a competitor suv. Yeah they both have wheels and can drive around but they have wildly different goals and intended usecases.
In fairness egpus were more of an afterthought usecase for both interfaces, thunderbolt was meant for displays and fast storage and docks and stuff and occulink for storage and accelerators (which are sometimes gpus I guess but used very differently from what we’d think of an egpu). E-gpus were, at least until recently always 3rd party “hacks”(as in used outside the intended functionality), if there is pcie, someone will inevitably hook a gpu up to it, be it the wifi slot, a expresscard slot or a thunderbolt port and I do love that.
Anyways, with Thunderbolt 5 and USB 4.2, eGPUs should not need Oculink anymore, but until then (a year or two?) we are going to see Oculink in consumer hardware because it makes sense for eGPUs.