An adapter that splits a single PCIE slot(x16) to hold 4 x M.2 NVMe SSDs(x4 each) would be a great way to persist a Redis instance that is not just serving as a cache.
If the same can be done with Optane SSDs, the lower latency will at higher queue depth will certainly help.
This is not a low profile card, and wastes quite a lot of space. It should take two cards on each side of the board, with the connectors facing orthogonal to those of the x16 slot.
You can't put something as tall as a M.2 connector on the back side of an expansion slot without violating the form factor guidelines and enchroaching on the space of the next slot over. The only compliant way to put M.2 drives on the back is to use an offset edge connector so the main board is a bit lower than it usually would be. Amfeltec has some boards like this, but I think they have a patent on their offset connector. http://amfeltec.com/products/pci-express-gen-3-carrier-board...
Oh. Well, there are cases where it would fit without such an exotic connector, but those are non-compliant.
I assume you don't need licenses for just making a dumb PCIe card, if you don't name it with trademarks? Or are there patents you need to license to sell PCIe-compatible, non-electronic cards?
Why would that help -- genuinely curious? Shouldn't a redis server already not be bound by the secondary storage I/O speed? I thought it was a main memory system with asynchronous commits to disk?
If the same can be done with Optane SSDs, the lower latency will at higher queue depth will certainly help.