GPU throughput data of different models and different numbers

[kkzzzz0922]

I read the paper and specific code, want to know how the GPU throughput data of different models and different numbers are obtained. Is it obtained through actual measurements or calculation formulas?
Look forward to your reply

[gudiandian]

They are obtained through actual measurements.

[kkzzzz0922]

Thank you for your reply, but I would also like to ask about the source of T4's throughput data. How did you get it? I tried to run adaptDL, but there were too many errors. And no more specific throughput data was found in the paper.

[gudiandian]

The T4 throughputs were parsed from the data provided by the original adaptDL repo. The errors might be caused by wrong arguments parsed to the scheduler. Please provide more details (your commands and the errors) if you need more help.