Table 5 Calculation time efficiency comparison between the proposed NVNMD and the established MLMD/CMD.
From: Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture
Method | System | Computer architecture | Hardware resource | Ti (s step−1 atom−1) | μt (s step−1 atom−1) |
|---|---|---|---|---|---|
NVNMD (this work) | Benzene | NvN | FPGA | 2.0 × 10−7 | 2.0 × 10−7 |
Naphthalene | FPGA | 2.0 × 10−7 | |||
Aspirin | FPGA | 2.0 × 10−7 | |||
Antimony (bulk) | FPGA | 2.0 × 10−7 | |||
Germanium telluride (bulk) | FPGA | 2.0 × 10−7 | |||
Li-Ge-P-S (bulk) | FPGA | 2.0 × 10−7 | |||
MLMD | vN | GPU | 5.6 × 10−5 | 2.4 × 10−5 | |
SiO2123 | 80 CPU cores | 3.6 × 10−5 | |||
Cu (original DP)124 | GPU | 2.8 × 10−5 | |||
H2O (original DP)124 | GPU | 9.5 × 10−6 | |||
Al-Cu-Mg (original DP)124 | GPU | 8.7 × 10−5 | |||
Cu (compressed DP)124 | GPU | 2.8 × 10−6 | |||
H2O (compressed DP)124 | GPU | 2.6 × 10−6 | |||
Al-Cu-Mg (compressed DP)124 | GPU | 5.4 × 10−6 | |||
GeTe (compressed DP)56 | GPU | 3.7 × 10−6 | |||
Li-Ge-P-S (compressed DP)62 | GPU | 9.4 × 10−6 | |||
CMD | Lennard-Jones32 | vN | 12 CPU cores | 7.0 × 10−7 | 7.3 × 10−7 |
Chain32 | 3.2 × 10−7 | ||||
EAM32 | 1.9 × 10−6 | ||||
Chute32 | 1.7 × 10−7 | ||||
GROMACS with SIMD125 | 64 CPU cores | 7.6 × 10−8 | |||
GROMACS without SIMD125 | 1.3 × 10−6 |
