It occurred to me that, as framework continues to make newer mainboards for their laptops, we’ll see a surge of older mainboards that no longer have a home, but are still relatively functional and efficient machines.
Given that these are laptop CPUs, they’re reasonably low-power and high-density. Overall, I’d say they’re well-positioned to act as server boards.
A framework motherboard, oriented vertically in a blade-style configuration, taking about 10mm of horizontal space, and 140mm, would fit comfortably in a 4U rack, 32 motherboards wide. With the right backplane supplying power, ethernet, and whatever other peripherals you might want, you could easily imagine the following form factors:
4U 32x compute cluster, all networked together for easy PXI boot and remote tasking
- Note that, presuming 40W per mobo (no screen, 28W CPU TDP, and some breathing room), this fits well within the 1500W possible from a single 120V outlet. Throw a redundant 1300W supply in the back and you’d have a pretty sweet compute rig.
It’d come to a sweet 128-core, 256-thread, 2TB RAM cluster. That’d cost you $20k or so with new parts, but if you figure you’re getting the boards 50% off (or even free as part of enterprise upgrades) a $10k server with those specs is pretty hard to beat. A comparable epyc 7713 has an MSRP of $11k on its own, without a board or RAM. The TDP wouldn’t quite compete with your epycs and xeons, but it’s not so far off as to be unreasonable.
1U+ storage cluster, with the backplane sporting a RAID controller and/or NVMe switch, and a single framework motherboard, with a collection of ports for external storage
- It’d be really cool if there was a way to make this able to seamlessly “fail over” to a second CPU with the same drives, but I don’t think that’s particularly feasible with modern hardware.
The chassis could also easily break out the two front expansion slots for whatever purposes you might want, like video out, removable storage, etc.
I’m imagining a setup where swapping out a single failing motherboard is a hot-swap operation, similar to a hard drive swap in a rackmount system. Pop open a small panel, power off the failing motherboard, pull it out, and pop in a new one, all without disturbing any other connected systems. Turn it on, it pulls a network config, and you’re running again.