Agreed. I think FW have made a mistake here and not read the room well at all.
EDIT: While the above comment was my first reaction to the news. Apparently “Filip” was told a while ago of FW’s plans. So, that kind of invalidates my first reaction.
2 Likes
I’d like to just shut down any negativity.
I was informed by Framework about them working on an official oculink solution back in December after I published my first working solution. I just had no idea how far along it was but it seems like it was in its final phase now that they published it, so I wouldn’t put any direct blame on Framework.
It seems like their solution doesn’t include any kind of redriver or retimer, but I assume they did verify it, although I wonder if it will actually work at full bandwidth with all setups when you connect it to the regular x16 slot dock.
20 Likes
Going to try and bring things back from the edge…
I am eager to try out Filip’s adapter. I also think I’m going to try to resurrect my custom eGPU with the measurements that folks have provided for the 8i board. The walls are black metal pegboard (250mm x 350mm). They were $12 USD each when I bought them.
I do need to find a better flexATX PSU. The 500W Apevia just couldn’t handle the spikes from the RX 9070 XT.
5 Likes
You guys getting mad that Framework decided to make one of the most requested addons for the 16 is pretty wild to be honest. Just be happy about it, sheesh. I just wish people would stop trying their best to be negative.
I am so glad FW decided to make one. This means it will be supported well. While community projects are great, they are rarely supported as well as one from the actual manufacturer. The person who made it often decides to stop and if you missed out you lost your chance. There have been many examples already of this happening. An occulink adapter is one of those things that will benefit tremendously by having first party support. The fact they picked it up is awesome.
8 Likes
Exactly my thoughts when they informed me. It’s a shame there is no bifurcation support, but that should be covered by my adapter for people who want to repurpose the laptop for use in a server environment where they want more PCIe slots.
I still do plan to add a redriver to my board for the learning experience and better signal integrity, so whoever might want one of those in the future, I will definitely do a run of it as well if all goes well.
10 Likes
Just to confirm. Was your laptop plugged into a charger when you did this? There’s a BIOS setting where it drops down to PCIe 3.0 when its on battery thats enabled by default.
1 Like
I have a use case for this so I will definitely follow your efforts as well.
I do also have an x16 slot adapter on the list of things I want to do where the port is on the edge of the board instead of right in the middle of it for better enclosure designs. It should also support auto on/off if all goes well. Although thats still far down the line.
1 Like
Personally felt like a rug pull, where you did all this work only for it to be taken off with them. Knowing now that you were aware of this totally mutes any negativity I have towards it and in the end, yours provoked more options than theirs due to the bifurcation. Crazy it took 3 years and multiple attempts to get here and they hopped on board. I’m surprised it took until now.
I wonder how much it’s going to be because it appears to be a full kit, rather than individual parts.
They have economy of scale on their side, but I doubt it will be anywhere under 100-150 range since it includes the adapter itself, expansion bay adapter and enclosure, the x16 adapter and the cable.
1 Like
The flex psu/GPU bit of it makes me wonder who would actually use that though, and what the use case was that made them go that route. I am just failing to understand the use case of having the GPU module on its own as an EGPU, and it only having the 1 USB C port.
Hopefully we get a Q&A regarding the oculink dev kit
Which 8x board is everyone using, I’m looking at either the NFHK 8X Oculink SFF-8612 8611 to PCIE PCI-Express 16x Adapter with ATX 24pin Power Port for Mainboard Graphics Card : Amazon.com.au: Computers & also the cablecc Oculink 8X SFF-8612 8611 to PCI-Express PCIE 16x Adapter with ATX 24pin Power Port for Mainboard Graphics Card : Amazon.com.au: Computers (which look identical - just rebadged / different manufacturer?) - Are these basically the only options? Any positives / negatives apart from price difference on these?
I’ll initially be using a Radeon RX 6800 XT, of which I’ll be upgrading to what ever the best card I can get, 9070 XT or higher for when Squadron 42 releases.
Thank you for the full information that I was personally waiting to find out. Pretty amazing that they listened.
Now let’s see the price. 
1 Like
It’s good they let you know. You still have your design, you could hit alternate options, like MCIO or a dual x4 card very easily with just a few changes. None of the work is lost, having all of that work done already leaves the future open for other projects based on it. It leaves people a roadmap to connecting to the x8 PCIe bus which is full of possibilities for built-in devices!
1 Like
The Problem ist that the external GPU (RX 6800) is not getting recognised at PCIe 4.0 (My old GTX 1060 does get recognised automaticly). However if i force it to PCIe 3.0 with that Battery Saver Setting it works just fine. Probably a Cable Issue.
I’m actually glad to see this - not for the FW16, but because it will let me reuse the FW16 GPU I currently have when I switch over to FIlip’s board (at least, whenever FW releases the devkit for purchase). I might use the old GPU in my homelab or alongside a miniPC for a Steam Machine type of setup. Until now, there was no way to repurpose old FW16 GPUs (other than reselling them).
I wanted to wait a bit after their announcement to see the discussion here before weighing in. Frankly, I think this is a win for the community, as mentioned above - this means there’s an official option. Filip’s board will still provide bifurcation support for those who want that (and I almost certainly will when I upgrade my mainboard someday - I might put Filip’s board with my current mainboard into my aforementioned homelab as some kind of mini server.)
The price is likely to be a little steep, as this is a Dev Kit, and it’d be perfectly reasonable for them to have a limited number produced. Oculink support is really really desired by a section of the community, but we’re probably a smaller group than you’d think. I’m okay with this approach - with any luck, this might also influence/improve future availability of more FW16 expansion modules, as more dev teams could experiment with other hardware to see what’s possible.
I’m still going to keep an eye out for Filip’s board with a redriver/etc, too, because that might be really nice to have down the road.
2 Likes
The Cablecc seems to be a resale of the same board - even has the NFHK logo on it in the product photo. Possible you may get a v1.0 or v2.0 board, we’re not sure yet what/if there are really any noticeable differences in their functionality. Also, the 3D model I’m designing as an enclosure for GPUs with this eGPU will work with either version.
3 Likes
Just wanted to provide a progress update - had some time to refine the enclosure for the NFHK eGPU some more today. I’m just nailing down a couple more details, then I can make adjustments for tolerances and publish an “alpha” version for folks to test (maybe without the “shell” enclosure at first). Had to make a “joining” piece for the sled, due to the fact I’m trying to enable folks with printer beds as small as 200x200mm to be able to print this.
I also have drafted an adapter for SFX power supplies:
And another for Flex ATX:
I could add in an option for TFX later, but given how uncommon that size is, I’ll wait and see if there’s any demand for it.
4 Likes
My OcuLINK board got here today, and seems to work fine, with one caveat - Whenever I actually have a GPU connected, my CPU gets locked to 600 MHz. Saw this was an issue for a few other things where the EC crashes, but here it seems to happen every time I boot with a GPU connected. As soon as I power off, unplug the GPU, and power back on, it works fine. Happens on both BIOS 3.04 and 3.05. I’m running the Ryzen AI 9 HX 370, attempting to add an AMD 6700 XT. Running CachyOS, where AMDGPU_TOP reports a 4.0x8 link, so it seems like everything is working except for the EC throttling the hell out of my CPU. Already submitted a support ticket, since it seems like plenty of you have it working just fine without this. Anyone had similar issues, or is my laptop a special snowflake here?
-Edit: I have now also tested with the other graphics card I have laying around, the very auspicious AMD R7 240. This one works without any CPU throttling, and connects at 3.0x8. So my guess at the moment is that it’s a firmware issue where the EC is crashing when it experiences certain loads during the boot sequence, but I don’t know enough to dive in and actually test that.
-Edit 2: Remembered I had a Vega 64 with some nasty overheating issues gathering dust in a closet. This one also shows no issues with CPU throttling. The 6700 XT works fine on a desktop system, so maybe something with ReBAR or the PCIe gen 4.0 link? I’ll see if I can find out more this weekend when I have some more time to fiddle with it.
Amazing! Would love to get the files to print it out and give it a test. Have a friend with a 3D Printer who has offered to print me some things. Have just ordered the cablecc board and now to wait patiently for the Framework Oculink Dev Kit to be released.
1 Like