OpenMetal

IaaS vendor selling bare-metal servers, hosted private cloud (OpenStack) and GPU servers, and the publisher of openmetal-tokens-per-second. Its blog runs technical explainers next to hardware pricing and consultation offers, which is what puts anything from it at T3 here: the measurements are first-party and the seller ran them.

Only the company's own site was checked at ingest; no Wikipedia or Wikidata entry is claimed, because none was verified.

The angle worth noting for this spoke: OpenMetal writes for buyers weighing CPU inference and private hardware against rented GPUs, so its content covers the CPU end of serving that GPU-centric sources here ignore — llama-cpp aside, this is the spoke's only material on Xeon-class throughput. The purchasing side of that argument belongs to ../cloud-wiki.

Related

openmetal-tokens-per-second · tokens-per-second · llm-inference