How long did pretraining take?

I could not find any mention of training time in the tech report. Could you tell us? Something like "On `X` B200 GPUs, training took `Y` GPU hours for Nemotron 3 Super 120B-A12B with `Z` precision.", or "Pretraining throughput for one B200 GPU was `X` tokens per second.` (if you want to keep networking overhead a secret).