Skip to main content

Table 5 Docking calculation times and acceleration rates for three proteins of different sizes.

From: Protein-protein docking on hardware accelerators: comparison of GPU and MIC architectures

  

Small

Medium

Large

Receptor (#residues)

GRB2 C-ter

SH3 domain (61)

CCS metallochaperone (249)

Nitrogenase Mo-Fe protein (2026)

Ligand (#residues)

Vav N-ter

SH3 domain (70)

SOD1 superoxide dismutase (153)

Nitrogenase Fe protein (578)

PDB ID

[PDB:1GCQ]

[PDB:1JK9]

[PDB:1N2C]

FFT size

80 × 80 × 80

128 × 128 × 128

216 × 216 × 216

Docking time [second] (vs. 1CPU)

1CPU

38.3

(1.0×)

186.4

(1.0×)

1105.6

(1.0×)

 

8CPUs

8.4

(4.6×)

38.5

(4.8×)

177.5

(6.2×)

 

GPU

5.8

(6.6×)

10.8

(17.3×)

62.2

(17.8×)

 

MICoffload

58.7

(0.7×)

77.0

(2.4×)

180.6

(6.1×)

 

MICnative

7.6

(5.0×)

26.8

(7.0×)

310.5

(3.6×)

  1. The MIC native mode (MICnative) used an optimized numbers of threads, which were the largest numbers available for each protein size (small = 240 threads, medium = 171 threads, and large = 38 threads).