hpcs-16-mic-v2

Speedup of deep neural network learning on the MIC-architecture
git clone https://git.igankevich.com/hpcs-16-mic-v2.git
Log | Files | Refs

results.tex (2205B)


      1 \section{RESULTS}
      2 It is worth mentioning, that implicit working model proved to be easier to program, compared to the explicit model and enabled to reach an acceptable time of learning. It was accelerated by 13.5 compared to the sequential version.
      3 
      4 %new paragraph
      5 Table~\ref{tab:workload-2} describes results of experiments. Test number column indicates a test case from the following list:
      6 \begin{enumerate}
      7     \item Initial non-optimized version that was run on one-core processor. 
      8     \item Optimized version that was run on multi-core x86 processor (OpenMP directives and vectorization).
      9     \item Optimized version that was run on MIC architecture (OpenMP directives, vectorization and explicit data transfer).
     10     \item Optimized version that was run on MIC architecture (OpenMP directives, vectorization and implicit data transfer).
     11 \end{enumerate}
     12 Other columns in order from left to right: architecture, the number of threads, time of execution, speed up (compared with initial non-optimized version) and learning accuracy for each test.
     13 
     14 %new table
     15 \begin{table}
     16     \centering
     17     \caption{Comparing working time and precision of learning.}
     18     \begin{tabular}{llrrrr}
     19         \toprule
     20         Test number & Arch. & Threads  & Time, s & Speed up & Accuracy\\
     21         \midrule
     22             1   & x86   & 1      & 7952 & 1.0  & 19.19 \\
     23             2   & x86   & 48     & 542  & 14.7 & 18.99 \\
     24             3   & MIC   & 240    & 6889 & 1.2  & 20.05 \\
     25             4   & MIC   & 240    & 589  & 13.5 & 20.05 \\
     26         \bottomrule
     27     \end{tabular}
     28     \label{tab:workload-2}
     29 \end{table}
     30 
     31 %old table
     32 %\begin{table}[h]
     33 %    \centering
     34 %    \caption{Comparing working time and precision of learning.}
     35 %    \begin{tabular}{llrrrr}
     36 %        \toprule
     37 %        Arch. & Version  & Threads  & Time, s & Speed up & Accuracy\\
     38 %        \midrule
     39 %        x86 & Initial      & 1      & 7952 & 1.0  & 19.19 \\
     40 %        x86 & Parallel     & 48     & 542  & 14.7 & 18.99 \\
     41 %        MIC & Offload (explicit model)     & 240    & 6889 & 1.2  & 20.05 \\
     42 %        MIC & Cilk (implicit model)        & 240    & 589  & 13.5 & 20.05\\
     43 %        \bottomrule
     44 %    \end{tabular}
     45 %    \label{tab:workload-2}
     46 %\end{table}