Most modern servers support all three ways of reporting out-of-band health. Here's the tradeoff Linivo customers actually run into when choosing.
Every BMC vendor ships a hardware MIB, and it's the same polling model you already use for switches and routers, so it fits your existing sensor deployment with no new protocol to open. It's also the least expressive: most vendor MIBs report generic "sensor 3 = 41" style readings rather than named, typed fields.
Redfish returns structured JSON over HTTPS with named fields (PSU redundancy state, individual drive wear level) instead of numbered sensors. It's the best option when it's available and your sensor can reach the BMC over HTTPS, which is usually true since it's the same port SNMP's TLS variant would need anyway.
Some smaller vendors' BMCs, or older hardware, only expose IPMI over LAN, no SNMP agent, no Redfish. It works, and it's well-supported, but the sensor data returned is closer to SNMP's granularity than Redfish's. Use it as the fallback when the other two aren't available, not as a first choice.