Table 3. Best production time (ns per day) for the different test systems (AMOEBA force field) using various methods. Number of atoms and optimal number of cores are given for each systems. All timings are given for Intel Haswell processors. Reference canonical Tinker CPU times are given for Open-MP computations using 8 cores. All computations were performed using a RESPA (2 fs) integrator if not specified otherwise. ASPC = Always Stable Predictor Corrector.53 N.A. = Non Applicable due to memory limitations. GPU production times were obtained using the Tinker-OpenMM software34 (CUDA 7.5), the JI/DIIS solver and a GTX 1080 NVIDIA card.
Systems | Ubiquitin | DHFR | COX-2 | STMV | Ribosome |
Number of atoms | 9737 | 23 558 | 174 219 | 1 066 628 | 3 484 755 |
Tinker-HP number of CPU cores | 480 | 680(960) | 2400 | 10 800 | 10 800 |
PCG (10–5 D, ASPC) | 8.4 | 6.3(7.2) | 1.6 | 0.45 | 0.18 |
TPCG2 | 10.42 | 7.81(8.93) | 1.98 | 0.56 | 0.22 |
TPCG2/RESPA (3 fs) | 15.62 | 11.71(13.39) | 2.98 | 0.84 | 0.34 |
CPU OPEN-MP | 0.43 | 0.21 | 0.024 | 0.0007 | N.A. |
GPU (GTX 1080) | 10.97 | 7.85 | 1.15 | N.A. | N.A. |
Systems (water boxes) | Puddle | Pond | Lake | Sea | Ocean |
Number of atoms | 96 000 | 288 000 | 864 000 | 7 776 000 | 23.3 × 106 |
Tinker-HP number of CPU cores | 1440 | 2400 | 7200 | 7104 | 12 288 |
PCG (10–5 D, ASPC) | 2.54 | 1.3 | 0.52 | 0.062 | 0.0077 |
TPCG2 | 3.10 | 1.59 | 0.63 | 0.076 | 0.01 |
TPCG2/RESPA (3 fs) | 4.65 | 2.38 | 0.95 | 0.11 | 0.014 |
CPU OPEN-MP | 0.050 | 0.014 | 0.003 | N.A. | N.A. |
GPU (GTX 1080) | 2.06 | 0.80 | 0.21 | N.A. | N.A. |