As of early 2021, please use Tinker9 for GPU simulations. Tinker9 GPU molecular dynamics
Tinker GPU simulations should use Tinker9. As of 2021, we switched to Tinker9 for GPU simulations, which is similar/consistent to Tinker CPU in terms of usage/setup, is faster than OpenMM for AMOEBA and some fixed charged force fields, and has all our new development for AMOEBA+.
The setup and key files used by Tinker CPU (on this wiki page) can be directly applied to Tinker9. Set "openmp-threads 1" in the key file (this is no longer needed).
Free to academic and nonprofit org
About the compilation of Tinker9, please refer to the successful build on the GitHub site (https://github.com/TinkerTools/tinker9/discussions/121)
!!! Note: as of today (Feb 22, 2021), tinker9 has to be compiled on an older machine (CPU) in order to generally run on newer machines. An alternative is to build an executable on each machine.
Example setup, xyz and key files for protein simulations: https://github.com/TinkerTools/tinker9/blob/master/example/
Recommend to use RESPA integrator with 2fs time step and write out less frequent (e.g. every 2 ps). On RTX3070, you should be able to achieve ~40ns/day for DHFR. Use MonteCarlo or Langvin piston for pressure control in NPT More details see Tutorials
To run tinker9 is almost the same as to run the canonical tinker. After compilation, an executable called tinker9 exists in the build directory. Here is the help information if you directly execute ./tinker9
SYNOPSIS tinker9 PROGRAM [ args... ] PROGRAMS AVAILABLE analyze bar dynamic info minimize testgrad help
So it is just adding tinker9 in front of the canonical tinker run commands.
Here are the compiled executables ready to use on our clusters.
/home/liuchw/Softwares/tinkers/Tinker9-latest/build_cuda11.2/tinker9 #this one runs on 3070/3080 cards
/home/liuchw/Softwares/tinkers/Tinker9-latest/build_cuda10.2/tinker9 #this one runs on the rest of the cards (not all of the cards have been tested)
An example MD command line: /home/liuchw/Documents/Github.leucinw/Tinker9/build/tinker9 dynamic box.xyz 500000 2 2 2 298 >& mdlog&
Tinker9 Run scripts:
In Ren lab: bash; source /home/liuchw/.bashrc.tinker9 (this will set up the right build on each node).
Option 1: manually select the build according to the GPU card (RTX 30xx vs others)
#!/bin/bash export TINKER9=/home/liuchw/Softwares/tinkers/Tinker9-latest/ export CUDA_DEVICE_ORDER=PCI_BUS_ID export CUDA_VISIBLE_DEVICES=0 # device number; can use 1 or 2 if there are multiple GPU cards # for 3070/3080/3090 nodes (check use `nvidia-smi` ) $TINKER9/build_cuda11.2/tinker9 dynamic your.xyz -k your.key 1000 2.0 2.0 4 298.15 1.0 # for other GPU nodes (we tested on 1070/1080/1080Ti/2070/2080) #$TINKER9/build_cuda10.2/tinker9 dynamic your.xyz -k your.key 1000 2.0 2.0 4 298.15 1.0
Option 2: automatically select the compatible build from multiple builds. Either source the following lines or paste them in your run script will work. Aliases such as dynamic_gpu, analyze_gpu, and bar_gpu are made for convenience.
#!/usr/bin/bash VAL=`nvidia-smi &> /dev/null; echo $?` # check existence if [ $VAL != 0 ]; then echo -e "\e[101mCUDA utility not installed on `hostname`\e[0m" else export TINKER9=/home/liuchw/Documents/Github.leucinw/Tinker9 VERSION=`nvidia-smi | grep "CUDA Version" | cut -c70-73` #check CUDA Version if [ $VERSION == 10.2 ]; then export TINKER9EXE=$TINKER9/build_1 #make aliases alias dynamic_gpu=$TINKER9EXE/dynamic9.sh alias analyze_gpu=$TINKER9EXE/analyze9.sh alias bar_gpu=$TINKER9EXE/bar9.sh elif [ $VERSION == 11.2 ]; then export TINKER9EXE=$TINKER9/build #make aliases alias dynamic_gpu=$TINKER9EXE/dynamic9.sh alias analyze_gpu=$TINKER9EXE/analyze9.sh alias bar_gpu=$TINKER9EXE/bar9.sh else echo -e "\e[101mTinker9 not supported for CUDA $VERSION\e[0m" fi fi