4 ms·
I actually had an ODroid H2 that failed maybe 6 months ago (not by any fault of its own really, it was installed in a "difficult" environment and probably got m
by jcrawfordor 4y ago
I actually had an ODroid H2 that failed maybe 6 months ago (not by any fault of its own really, it was installed in a "difficult" environment and probably got much hotter than was reasonable). I went shopping around for a newer replacement and was surprised to find that the market of very small, e.g. nano-ITX embedded motherboards has basically dried up and at that point it was very difficult to find something that was even comparable with the H2 at the same price point. Much of what I was seeing were $150 boards with very old Celeron chips that underperformed the H2 (due to being older designs mostly, but still being manufactured). I think ARM adoption is large enough that there's not much of a market for x86 embedded boards that isn't heavy equipment that's very price insensitive, and even then the fact that most of the options are built around 9 year old processors suggests that this market might be mostly legacy products.
You could probably save a bit of money, but I would guess not more than $40 or so, by buying a motherboard and socketed processor separately, a more conventional PC setup. But these usually aren't readily amenable to completely passive cooling which is an important criteria for me (and off the shelf passive coolers for even modest TDPs are expensive on their own), and it's also hard (maybe impossible?) to find the dual network interfaces in a setup that would come in below $165.
In my case I ended up going with a NUC and external RAID controller/drive enclosure but it was an appreciably more costly setup-admittedly for a much higher end processor, but I think I ended up at around $500 before buying drives for the external enclosure which of course ended up as the largest cost.
The application is a video surveillance NVR, I wanted x86 as the commercial software is Windows-based and not built for ARM. 2.5GbE network interfaces were not important (the bottleneck would be the archival spinning drives well before that), but dual network interfaces were as the existing setup was based on a dedicated surveillance switch running off the second interface of the NVR and I like that this is a simple setup with fewer configuration points than a VLAN hosted on the core switch (there's a reason most commercial appliance NVRs are set up this same way).