Explanation request for where you obtained architectural numeric specs for A100, TPUv3, and MI210.

[topcuburak]

Dear developers,

I was trying to enlarge new GPUs and TPUs for the LLMCompas hardware models. While some numerical values are obvious, such as the number of SMs, operating frequencies, and so on, some, such as overhead and the I/O model, are not that certain.

Is it possible for you to share such a detailed manual describing how to reconfigure the hardware_model for newer architectural specs?

Thank you,
Burak.