Damn, that’s steep.
Still not working for me though. I cleaned it up in my quote as well. US domain creates a redirection issue, and even the .com cleaned up results in “Sorry, the page you requested can not be found:(” . But yeah as long as nobody can find a reliable source (more reliable than even one aliexpress listing, please), I’ll start my next attempt with dual 4i
EDIT: Actually, I’d really like the expansion bay deep dive before I start working. Certain stuff is not 100% clear, e.g. the maximum outline, the location and dimension of the back slit we can use for ports, etc. I’ll probably do the most difficult parts now with the best assumptions
You’re right, from the US it works, even the .com link. Weird.
But yeah likely just one manufacturer selling off their remaining stock after being forced to buy 4200 for their own low-quantity product (they’re only listing 100). Wouldn’t rely on it (in addition to the high price)
And this is why I highly recommend having a good component supplier in China, allows you to get products that aren’t listed anywhere else. I paid 5 dollars each for the 4i or 8i. You really shouldn’t need many via’s, I’m also going to say Amphenol isn’t the only company with OcuLink connectors…
I wouldn’t suggest doing OcuLink as a first time design due to the PCIe signaling
Hey if you could forward some to use that’s be phenomenal. I could ask somebody that does have contacts but I’m not sure the lower accessibility (and maybe reliability) is worth it.
Yep definitely not by choice.
PCI-E lanes cross over due to orientation of the connectors, so that at least.
And the Framework Expansion Bay Interface is also pretty tight, just barely managed with some vias for only 4 lanes with the clearances cheap manufacturing offers (jlcpcb). Definitely need vias there, don’t even think you could do without vias-in-pads for the full 8 lanes AND proper grounding. But any insights are appreciated, since I’m going to do it anyway
They are inferior in bandwidth, not inferior in general, you pay for the performance by not having hotplug and power delivery and usb and netowrking and all that good stuff.
It’s definitely very worth it for some applications but it also has some huge downsides for others (people that like undocking without rebooting for example).
Damn, nice! Should’ve said you were working on it, I’m basically done with my dual 4i + DP module
I’ll likely test mine too since I managed to do it with the cheapest options on jlcpcb (except ENIG coating ofc), so it’s only a 40€ order with shipping
Nice, just in time! Ordered my boards 1hr before your update, was still able to cancel them to adjust.
Feedback:
One critical dimension is still missing, and that is the location off the locking screw for the cover door. It could be measured from the 3D orthographic view, but that’s not optimal. Only then can we be sure about the PCB-edge-to-cover-door offset by means of the cover door dimensions.
What part number are these M2 standoffs? In the 3D view, they look to have threads, I expect that is not the case and the shell will have the threads instead. Additionally, previously they were 5.4mm holes, now they are 6.2mm - the only M2 (threaded) PCB inserts we used before needed 3.6mm only. I also can’t get any M2 spacers/standoffs that are this wide (~7.2-7.4mm?) - biggest are 5mm. So I suspect these are a special model (maybe soldered on?)
What part number are the FXBeam interposer standoffs? Or, if they are custom, are they soldered or can we pick them off our default board?
Can you confirm the fan connector pinout and part number? In the SSD reference design, it is Pin 1 = GND (for JST-like connector A1001WR-S-04PNLNT1T00R), now it’s flipped, Pin 4 = GND and a totally different connector type (FPC). The part number specified is AFC58-S04FIA-R1, which we cannot find online. Could it perhaps be AFC05-S04FIA? That one looks very similar to some prototype board pictures and has a similar name. It looks to be a standard FPC connector, if it is, could you provide the dimensions of the fan FPCs so we can choose a suitable connector?
Next to each fan connector, there seem to be holes for plastic pins. The right one changed shape from last revision and is now oval and thus would not fit the same circular pin. Is this intentional? Both plastic pins in the 3D model look to be the same and circular
You’ve reduced the space around the FXBeam connector needed to route all the pins greatly with this revision, making it nearly impossible to work with cheap 4 layer boards. Can you confirm the board space really needs to be free? White is the original outline, turquise is the new one. In red is what I had hoped was available from all the images we saw (and the GPU does use that space).
The FXBeam pinout in the doc has some different names at least, and different control signals at worst. E.g. PF3/PG3 now has a I2C (_VS) pinout, whereas previously it was EXT_SSD1_RST#/EXT_SSD2_RST#. These sound like completely different signals to me. Are these alternate pinouts just like PWR_GOOD/DP_A_AUX for an additional PCIe REFCLK? if so, how are these switched/controlled?
Thank you for giving us more information already, before even the Expansion Bay deep dive, it’s greatly appreciated - let’s hope the rest is filled in in time for it. At least what’s there now is enough information to get the boards themselves mostly done (but sadly not fully)
We’re preparing updated CAD to address your questions @Seneral.
The standoffs are unthreaded and should be 3d-printable.
The FXbeam nuts are custom, and we’re working on a way to make them available to developers.
We have switched to an FPC. There should be multiple receptacles that are compatible. We’ll share a drawing of the FPC side to make it easier to find compatible ones.
The additional space is needed for mechanical clearance. It is probably the case that a 6 layer PCB is needed for designs that use PCIe.
Damn, that’s unfortunate. I’ll try to violate some guidelines, and EMI be damned, I’ll try to get it done in 4 layers. We’ll see
But thanks for the answers!
Possibly, yeah. But I do want to make the barrier of entry as low as possible, that’s just what I want for open source stuff, even if most will buy the boards off of someone.
Currently the PCB order is still 43€ with shipping for 5 PCBs + Stencil, the three ports (DP + 2x OCulink 4i) will be <12€ even from mouser + shipping, which is in the realm of possibility for an individual to do.
Big asterix is that this cheapo version PCBs is 100mm wide only - I had hoped to use the one central screw at the top, but since it’s confirmed now we can’t really use it, the PCB will only have half a screw securing it on the top if you want to save on the ~15€ the 100mm version saves. Not great with three big ports there.
Ofc the costs of the cable + PCIE breakout board are nearly 110€ so yeah, the add in card costs are neglible.
Still, principle. Lowering the barrier of entry.
PS: Seems I can cram it in just barely, but clearances are suboptimal… Let’s hope crosstalk will be alright. I may also sacrifice some ground plane instead
