The Kust code has been compiled and run on a multicore system with intel Care b $13900$K $(24$ cores with $32$ threado$)$ and $64$ OByles RAM. By setting the Rayon ThreadfoolBulder, the number of threads was set to $1$ and $32,$ respectively, and the clapsed time is shown in Table $1.$ Calculate the speed up and predict the fraction of the$(4$ marks$)$serial portion for NThreads $=32$ using Amdahl's lawTable $4.$ Elapsed Parallel Processing TimeElapsed Processing Time $($in milliseconds$)8/0\mathrm{.}034\mathrm{.}8$Calculate the upper bound limit of achievable speedup using Amdahl's law.$(2$ marks$)$Considering the importance of matrix multiplication in deep learning, explain how could parallel computation impact the performance of training and running a deep