Figure 6a.
DOT was also tested by running the same calculation described in
Figure 5 on an Intel Paragon XP/S, also at the San Diego Supercomputer Center. This system contains 400 GP nodes, each of which performs at 34 Mflops on the Linpack 1000x1000 benchmark and 10 Mflops on the Linpack 100x100 benchmark. Only 256 of the 400 GP nodes have 32MB of memory, which is preferred for running DOT. The whole calculation takes only 8.7 minutes when 256 processors are used.
A log - log plot of the Paragon performance data is presented in
Figure 6b.
