A startup company makes a specialized ML (machine learning) hardware which trains a model 20 times faster than the normal hardware.
PLEASE SHOW WORK
A startup company makes a specialized ML (machine learning) hardware which trains a model
20 times faster than the normal hardware.
a. If the system spends 70% of the time training the model and 30% running inference, what is the
speedup if the new ML hardware is used? Assume the old hardware is used for inferencing.
b. Assume the model only needs to be trained once and then used for inferencing for a while. What
percentage of training and inferencing result in an overall speedup of 4? Assuming the new
hardware is used only for training, while old hardware is used for inferencing.
c. Suppose you have measured the percentage of model training is 70%, and 80% of the training
operations could be performed in parallel. What is the speedup of using 4 units of new ML
hardware?
Trending now
This is a popular solution!
Step by step
Solved in 3 steps
