To post this with the current title of "Cost of running ChatGPT is $100K per day, or $3M per month" is absurd.
The assumptions in the thread are ok, but is it possible that the 15-20 words/s that they mention is because they are batching inference?
To give myself some context, for $3,000,000 that is 1,600 A100 GPUs across 200 servers on GCP, 24/7.