Calculation time for 100 time-steps for 30,011 particles are plotted against machine type. We used one SUN Ultra SPCARC-II 400 MHz, one Fujitsu VPP processor with 9.6 Gflops, and SUN CPU with 1 to 3 MDGRAPE-2 boards. The lower figure is the same as the upper one except that 'Sun' is excluded. Direct summation (no cutoff method) is used for all cases. We used AMBER 5.0 optimized for 'Sun' and 'VPP' for non MDGRAPE-2 calculation. And we used modified version of AMBER 5.0 for MDGRAPE-2 calculation. The system has a small protein and water molecules. Please see accuracy page for detail of the MD system. The speed does not change so much as the number of atoms for a protein changes. Sun with 3 MDGRAPE-2 boards are 88 times faster than Sun without MDGRAPE.

Calculation time for 100 time-steps for 30,011 particles are plotted against cutoff type. First three data are the same as the previous figure. Last three data are calculation time by Sun machine with cutoff method of 8, 10 and 12 angstrom. The calculation time with MDGRAPE-2 boards is almost the same as that with cutoff method, which is often inaccurate(Please see accuracy page).

Calculation time for 100 time-steps with SUN machine connected with 1 to 3 MDGRAPE-2 boards are plotted against the number of particles. Direct summation, dipole tree code and quadrupole tree code were used. Theta is an opening angle which specify the accuracy of the tree code(Please see accuracy page).

Potential energy is compared between non-cutoff energy calculation and cutoff (8A, 10A, 12A). Energies with short cutoff method tend to differ from exact energy with non cutoff method.

The conditions of the following three simulations are the same as this one.

Energies are compared between host calculation and MDGRAPE-2.In both calculations,interactions have no cutoff.Two energies match very well.

Energies are compared between host calculation and tree code with MDGRAPE-2. In host calculation, interactions have no cutoff. In MDGRAPE-2, dipole tree code was used with a opening angle parameter of 0.3 and 0.5. Three energies match well, but a little difference exists when the opening angle is 0.5.

Energies are compared between host calculation and tree code with MDGRAPE-2. In host calculation, interactions have no cutoff. In MDGRAPE-2, quadrupole tree code was used with a opening angle parameter of 0.3 and 0.5. Three energies match very well.