Table 6 Calculation energy 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

 

T (s step−1 atom−1)

P (W)

η = T × P (J step−1 atom−1)

NVNMD (this work)

2.0 × 10−7

108

2.1 × 10−5

MLMD

1 CPU + 1 GPU26,31

5.6 × 10−5

280

1.6 × 10−2

Summit supercomputer31

2.7 × 10−10

7.6×106

2.1 × 10−3

CMD

≈10−7

≈102

≈10−5

  1. T is the calculation time efficiency. P is the power consumption. The NVNMD (this work) uses three CPU cores (Intel i7-10700K) and one FPGA (Xilinx xcvu9p). The MLMD uses CPU (IBM POWER9) and GPU (NVIDIA Tesla V100), as introduced in Ref. 26,31. The CMD results are taken from Table 5.
Back to article page