It exists! Custom OcuLink adapter for the Dual M.2 expansion bay module

I shared a post already in the OCuLink eGPU works with the Dual M.2 expansion bay module thread, but wanted to create a new thread specific to the custom OCuLink adapter.

The original thread has multiple posts that mostly detail the history of how this custom adapter came to be.

I have had the adapters for about a week and have tested a couple of them, with testing on a third one underway.

If you want to get one of these, there are a few things you need to know. I will quote some text from my post in the other thread (with some minor edits).

First, I want to make sure some things are as clear as I can make them. I am an individual. I have family commitments and a full-time job. There are some things about the process that you might not like, but I just don’t have the time or resources to go out of my way to do anything “above and beyond”. I need to prioritize the simple stuff first, such as limiting shipping to the US and Canada early on. I am exploring that a little, but again, I am just one person. Please be patient

Second, I cannot fully test every piece, nor do I have a wide range of docks/equipment to test on. I will be testing a few pieces on a single Framework 16 (Ryzen 7 7840HS motherboard) with the dual M.2 expansion bay. Testing is happening with a custom eGPU enclosure (no redriver and not working at this point), an Aoostar AG02 800W, and an RX 9070 XT. I do plan on trying some more testing with the custom enclosure, but it is lower priority. I will also be doing some basic testing with a PCIe to M.2 riser with an SSD, but that too is lower priority.

I cannot accept any liability if something goes horribly wrong. All I can say is that I installed 3 of the initial batch pieces that just didn’t work, 1 updated prototype revision that did work with the AG02, and 3 of the final version (that also work with the AG02). The single working prototype was in my laptop the longest (roughly 2 weeks). None caused any issues with my FW16. I shipped one of the original non-working pieces and the working prototype piece to OVER_CL0CK as well. He has confirmed the original design piece did not work and the revised version does work with his DEG1. It seems highly unlikely that something catastrophic would go wrong, BUT if you decide to acquire one of these, you are doing so at your own risk.

The price per piece will be $60 USD. Shipping in US and Canada will be $10. I do not have information about shipping to other countries yet. Depending on demand to those countries, it might actually be better if someone wants to buy multiple and redistribute.

If you are still interested in getting one of these, I have started a list.

This thread can also be used for people to share their experiences, thoughts, questions, etc about this custom adapter. I plan on posting more details about the testing too.

13 Likes

These are the testing details for the adapters.

I have an AOOSTAR GEM10 Mini PC with the AMD Ryzen 7 7840HS and 32GB LPDDR5(6400MHZ) RAM that has a built-in OCuLink port. This is what I have used to compare against my FW 16 with the Ryzen 7 7840HS and 64GB of DDR5-5600 memory. Even though there are differences, the results are (mostly) surprisingly close.

I have tried testing with a custom eGPU based on a cablecc board that does not have a redriver. I have tried a few different methods to get it working, but it just isn’t recognized.

My successful testing has been with an Aoostar AG02 800W using an RX 9070 XT. GPUz reports the connection at PCIe 4.0 x4.

The differences between the 3 tested pieces definitely fell within margin of error, so when I reference % differences, I took the median of the 3 pieces.

In Steel Nomad, Port Royal, and Time Spy Extreme, the variance was from 0.38% to 1.12%. Basically, margin of error. Sometimes this was in favour of the FW, sometimes the Mini PC.

The single largest difference in any of the testing was in the PCIe bandwidth test in 3D Mark. The Mini PC scored 6.55 GB/s while the FW 16 with the adapter scored 7.20 GB/s. The 7.20 GB/s result was consistent across the 3 tested pieces.

In the Cyberpunk 2077 benchmark, the FW 16 had 1.7% advantage average FPS, 3.4% advantage minimum FPS, and 1.8% advantage maximum FPS.

All of the above tests were in Windows.

I tested some runs in Ghost of Tsushima in Linux (I actually run Linux as my daily) using MangoHud. The difference between the Mini PC and the FW was within margin of error, especially considering there isn’t a built-in benchmark and you get variance between runs anyway. Ghost of Tsushima is the game I am currently playing through. I haven’t experienced any strange behaviour during testing/playing; no crashes or performance issues.

Looking forward to hearing the results from others!

4 Likes

Dude, that’s a sweet looking oculink adapter!

3 Likes

I’m in Europe, wondering if more people from that region are interested? Like collectively get it over and redistribute in Europe/EU?

2 Likes

I’m interessted.

2 Likes

I have the Expansion dev setup and few FPGA boards, so I can do some eye digrams over some different connections. This should give a idea of what Gen you can acheieve and if you need a retimer or something.