Open Compute servers are widely adopted with Intel Xeon 26xx v3 and v4 family. Intel has launched a huge amount of variation of these chips which supports two kind of memory DIMM (RDIMM and LRDIMM) with various power consumption and maximum operating temperatures.
Whatever your options will be, keep in mind that in modern architectures, resiliency is software defined not hardware and that your application shall be designed to sustain any potential hardware failure.
Finding the right tradeoff for it could seem challenging. Hopefully some core parameters might be helpfull to select the right system when thinking about pricing and power balance, keep in mind that low core count cpus have less cache, requires less power and decrease service disturbance in case of a single server failure as the number of potential computing ressource lost is smaller.
Currently a system configuration with a dual Xeon 2620v4 (total of 16 cores), 64GB of main memory, and two SSD’s with 256GB capacity associated with a dual 10Gbps network adapter is the best tradeoff. The system cost less than 3000$ with premium components, can host up to 40 VMs depending on application profile and is also a perfect baremetal solution. Getting into other configuration will means specific configurations with higher cost, and perhaps some flexibility for your application that you can’t avoid.