Discontinuous Galerkin methods Lecture 1 - Brown University
Discontinuous Galerkin methods Lecture 1 - Brown University
Discontinuous Galerkin methods Lecture 1 - Brown University
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
1 PFlop/s<br />
1 TFlop/s<br />
Moore’s law and the performance gap<br />
1 GFlop/s<br />
1 MFlop/s<br />
1 KFlop/s<br />
Moore’s Moore s Law<br />
Scalar<br />
IBM 7090<br />
UNIVAC 1<br />
EDSAC 1<br />
Super Scalar<br />
CDC 7600<br />
CDC 6600<br />
Vector<br />
IBM 360/195<br />
Cray 1<br />
Super Scalar/Vector/Parallel<br />
TMC CM-5 Cray T3D<br />
Cray 2<br />
Cray X-MP<br />
Parallel<br />
TMC CM-2<br />
ASCI Red<br />
Earth<br />
Simulator<br />
ASCI White<br />
Pacific<br />
1941 1 (Floating Point operations / second, Flop/s)<br />
1945 100<br />
1949 1,000 (1 KiloFlop/s, KFlop/s)<br />
1951 10,000<br />
1961 100,000<br />
1964 1,000,000 (1 MegaFlop/s, MFlop/s)<br />
1968 10,000,000<br />
1975 100,000,000<br />
1987 1,000,000,000 (1 GigaFlop/s, GFlop/s)<br />
1992 10,000,000,000<br />
1993 100,000,000,000<br />
1997 1,000,000,000,000 (1 TeraFlop/s, TFlop/s)<br />
2000 10,000,000,000,000<br />
2003 35,000,000,000,000 (35 TFlop/s)<br />
1950 1960 1970 1980 1990 2000 2010<br />
Current predictions in required<br />
performance needs show a large<br />
H. Meuer, H. Simon, E. Strohmaier, & JD<br />
and growing gap, known as the<br />
- Listing of the 500 most powerful<br />
Computers in the World<br />
- Yardstick: Rmax from LINPACK MPP<br />
performance gap<br />
Ax=b, dense problem<br />
ate<br />
TPP performance<br />
!"#$%&'#$()&#*)#+$,-./01&*2$,"344#*2#<br />
3<br />
Raw performance largely<br />
driven by innovation in<br />
hardware has so far followed<br />
Moore’s law: ‘Bang-per-buck’<br />
doubles every 24 months<br />
38+<br />