Changeset 977


Ignore:
Timestamp:
Mar 23, 2011, 1:16:48 PM (8 years ago)
Author:
lindanl
Message:

Add section 6 and some other minor changes

Location:
docs/PACT2011
Files:
4 added
4 edited

Legend:

Unmodified
Added
Removed
  • docs/PACT2011/04-methodology.tex

    r969 r977  
    7373Memory  & 2GB \\ \hline
    7474Hard disk & SCSI \\ \hline
     75Max TDP & 65W \\ \hline
    7576\end{tabular}
    7677\end{center}
     
    9394Memory  & 4GB \\ \hline
    9495Hard disk & SCSI 1TB \\ \hline
     96Max TDP & 73W \\ \hline
    9597
    9698\end{tabular}
     
    113115Memory  &  \\ \hline
    114116Hard disk &  \\ \hline
     117Max TDP & 95W \\ \hline
    115118
    116119\end{tabular}
  • docs/PACT2011/05-corei3.tex

    r969 r977  
    144144\includegraphics[width=85mm]{plots/corei3_energy.pdf}
    145145\end{center}
    146 \caption{Energy consumption on core i3 (J/GB)}
     146\caption{Energy consumption on core i3 (nJ/B)}
    147147\label{corei3_energy}
    148148\end{figure}
  • docs/PACT2011/06-scalability.tex

    r969 r977  
    11\section{Scalability}
    22\subsection{Performance}
    3 }
     3Figure \ref{TOT_Parabix2} shows the performance of Parabix2 on three different cores: core2, corei3 and sandybridge.
     4The processing time, which is evaluated as CPU cycles per thousand bytes,
     5is divided up by bitstream parsing and byte space postprocessing.
     6Bitstream parsing, mainly consists of SIMD instructions,
     7is able to achieve 17\% performance improvement moving from core2 to corei3;
     822\% performance improvement moving from corei3 to sandybridge,
     9which is relatively stable compared to postprocessing,
     10which gains 18\% to 31\% performance moving from core2 to corei3;
     110 to 17\% performance improvement moving from corei3 to sandybridge.
     12
     13As comparison, we also measured the performance of Expat on all the three cores.
     14The results are shown in Figure \ref{TOT_Expat}.
     15The performance improvement is less than 5\% by running Expat on corei3 instead of core2
     16and it is less than 10\% by running on sandybridge instead of corei3.
     17
     18Parabix2 scales much better than Expat and is able to achieve an overall performance improvement
     19up to 26\% simply by running the same code on a newer core.
     20Further improvement on sandybridge with AVX will be discussed in the next section.
     21
     22\begin{figure}
     23\begin{center}
     24\includegraphics[width=85mm, trim= 0 200 0 200]{plots/TOT_Parabix2.pdf}
     25\end{center}
     26\caption{Total CPU cycles /KB of Parabix2}
     27\label{TOT_Parabix2}
     28\end{figure}
     29
     30\begin{figure}
     31\begin{center}
     32\includegraphics[width=85mm]{plots/TOT_Expat.pdf}
     33\end{center}
     34\caption{Total CPU cycles /KB of Expat}
     35\label{TOT_Expat}
     36\end{figure}
     37
     38
    439\subsection{Power and Energy}
     40
     41The newer processors are not only designed to have better performance but also more energy-efficient.
     42Figure \ref{power_Parabix2} shows the average power when running Parabix2 on core2, corei3 and sandybridge with different input files.
     43On core2, the average power is about 32 watts. Corei3 saves 30\% of the power compared with core2.
     44Sandybridge saves 25\% of the power compared with corei3 and consumes only 15 watts.
     45The energy consumption is further improved by a shorter processing time as we moved to the newer cores.
     46Core2 consumes 3 to 4 times energy than sandybridge.
     47
     48\begin{figure}
     49\begin{center}
     50\includegraphics[width=85mm]{plots/power_Parabix2.pdf}
     51\end{center}
     52\caption{Average Power of Parabix2 (watts)}
     53\label{power_Parabix2}
     54\end{figure}
     55
     56\begin{figure}
     57\begin{center}
     58\includegraphics[width=85mm]{plots/energy_Parabix2.pdf}
     59\end{center}
     60\caption{Energy consumption of Parabix2 (nJ/B)}
     61\label{energy_Parabix2}
     62\end{figure}
Note: See TracChangeset for help on using the changeset viewer.