Each time a new AI training benchmark is introduced, the fastest training time gets longer. Then, hardware improvements gradually bring the execution time down, only to get thwarted again by the next benchmark. Then the cycle repeats itself.
9 days ago