With RAM prices and availability becoming a problem, why not make SBCs without onboard RAM and use a standard laptop SO-DIMM slot instead?
This could have several advantages:
Any thoughts?
I get where you are going. It sounds like a great idea at first, but it just creates more headaches down the road. The main pain point is that these SoC vendors (SpacemiT/ESWIN/…) don’t own the memory controller/phy IP, and their testing capability is very limited, unlike Intel/AMD. On top of that the board vendor (MilkV) adds another layer of "overhead” (DIMM slots will for sure negatively affect signal integrity). So the DDR training code has to make a big safety margin to accommodate for any DIMMs you may insert. Sometimes even that can fail, because the SoC/board vendor may never have tested the DIMM you chose. Allegedly Titan board is massively delayed due to DIMM slot layout issues impacting stability. It’s less of a problem in the PC/server world because those companies have huge amount of test resources, and DIMM vendors usually send out samples for motherboard companies to test out, so both sides validate against each other way ahead of going to market. I’d rather these SoC boards come with soldered DDR chips, and let the SoC/Board vendor test out everything and make sure it works, so I never need to worry about those issues. The DDR price will for sure knock out low-end SBCs, but I guess we’ll just have to live with that until the AI bubble bursts.
I agree with your technical reasoning, but I think an SBC with a SO-DIMM slot would likely sell more boards, especially at a time when demand seems fairly low.
The Titan is the only board that has really caught my interest lately, and that’s largely because I can buy it at a reasonable price and source my own memory. I would very much like to own a Jupiter 2, but I have a hard time justifying the cost. The biggest factor is the RAM pricing, which remains astronomically high and adds a substantial premium to the overall price of the system.