How are multiple CPUs utilized when doing inference with CNN?

Question

How current frameworks like Tensorflow or Onnx etc. are able to parallelize inference workload on CPU only? I understand multi-core CPU might be conceptual different as GPU