GROMACS version: 2018
GROMACS modification: No
I would like to install an old version of GROMACS, I don’t particularly care as long as it is before 2020, but, can’t get any to install. I have tried 2016, 2018 to no avail. The issue always comes down to GPU’s.
Can I just install a CPU only version of GROMACS? I never use GPU’s ever, so it is a little frustrating to constantly run into an install wall due do GPU’s. There must be an option when using cmake .. to tell it no GPU whatsoever…
I am slowly going insane from seeing
NVCC/C compiler combination does not seem to be supported. CUDA frequently
does not support the latest versions of the host compiler, so you might
want to try an earlier C/C++ compiler version and make sure your CUDA
compiler and driver are as recent as possible.
Call Stack (most recent call first):
CMakeLists.txt:582 (gmx_gpu_setup)
and other gpu related issues
In old versions, use -DGMX_GPU=OFF to disable anything related to GPU detection/utilization. Defaulting to cmake .. lacks a lot of control over important options.
1 Like
Thanks!
I also just found that yum install gromacs installs gromacs 2018.8 which also gets me out of a bind (alchemical decoupling in 2020 on has a PBC bug :( )
Note that package manager versions of GROMACS are not optimized for your hardware and you will almost certainly be sacrificing a lot of performance. Installing from source is always best.
yep. I was able to install 2018 with GPU off, so it should be fairly fast
It turns out the package manager version is for GPU and so I can’t use it. That said, I installed both 2016.4 and 2018 successfully, turning GPU off, however, both installs give me this when I run it
WARNING: Using the slow plain C kernels. This should
not happen during routine usage on supported platforms.
And both run at 25% the speed of the “correctly” installed 2021.3 version I also installed myself some months ago.
I installed 2016.4 using
cmake .. -DGMX_MPI=OFF -DGMX_GPU=OFF -DCMAKE_INSTALL_PREFIX=/apps/gromacs-2016.4 -DGMX_BUILD_OWN_FFTW=ON
and 2018 using
cmake .. -DGMX_GPU=OFF -DCMAKE_INSTALL_PREFIX=/apps/gromacs-2018 -DGMX_BUILD_OWN_FFTW=ON
I have installed gromacs many times on my personal Ubuntu computers, and Compute Canada, and never had the slow plain C kernel warning before :( it is unbearably slow as well.
(With some homework, I see that this means it has not automatically done my SIMD selection for me)
Any suggestions?
Under normal circumstances, cmake should correctly detect the appropriate SIMD instructions for your hardware. The output of configuration would be useful here to see what it found and didn’t find. You can manually set the appropriate SIMD instructions with the -DGMX_SIMD option, but it’s really bizarre that this would be necessary, so something strange is going on.
When I run GROMACS it mentions both that it is using the slow plain C kernel, and also that I should be using SIMD=AVX2_128, and I do have that. I have tried setting DGMX_SIMD=AVX2_128 and got an error message, I then also tried DGMX_SIMD=auto. I also tried -DGMX_USE_RDTSCP=ON. setting SIMD to auto, and using RDTSCP=ON led to an install with SIMD=none, and using plain C kernel.
gmx --version spits out
GROMACS: gmx, version 2018
Executable: /apps/gromacs-2018/bin/gmx
Data prefix: /apps/gromacs-2018
Working dir: /home/bkelly/test_gromacs
Command line:
gmx --version
GROMACS version: 2018
Precision: single
Memory model: 64 bit
MPI library: thread_mpi
OpenMP support: enabled (GMX_OPENMP_MAX_THREADS = 64)
GPU support: disabled
SIMD instructions: NONE
FFT library: fftw-3.3.5
RDTSCP usage: enabled
TNG support: enabled
Hwloc support: disabled
Tracing support: disabled
Built on: 2022-03-26 00:46:35
Built by: bkelly@c110370 [CMAKE]
Build OS/arch: Linux 3.10.0-1160.36.2.el7.x86_64 x86_64
Build CPU vendor: Unknown
Build CPU brand: Unknown
Build CPU family: 0 Model: 0 Stepping: 0
Build CPU features: Unknown
C compiler: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-cc GNU 9.3.0
C compiler flags: -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -O3 -DNDEBUG -funroll-all-loops -fexcess-precision=fast
C++ compiler: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-c++ GNU 9.3.0
C++ compiler flags: -fvisibility-inlines-hidden -std=c++17 -fmessage-length=0 -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -std=c++11 -O3 -DNDEBUG -funroll-all-loops -fexcess-precision=fast
which is alot messier than the 2021.3 version I have which is
GROMACS: gmx, version 2021.3
Executable: /usr/local/gromacs/bin/gmx
Data prefix: /usr/local/gromacs
Working dir: /home/bkelly/test_gromacs
Command line:
gmx --version
GROMACS version: 2021.3
Verified release checksum is c5bf577cc74de0e05106b7b6426476abb7f6530be7b4a2c64f637d6a6eca8fcb
Precision: mixed
Memory model: 64 bit
MPI library: thread_mpi
OpenMP support: enabled (GMX_OPENMP_MAX_THREADS = 64)
GPU support: disabled
SIMD instructions: AVX2_256
FFT library: fftw-3.3.8-sse2-avx-avx2-avx2_128
RDTSCP usage: enabled
TNG support: enabled
Hwloc support: disabled
Tracing support: disabled
C compiler: /opt/rh/devtoolset-9/root/usr/bin/cc GNU 9.3.1
C compiler flags: -mavx2 -mfma -Wno-missing-field-initializers -fexcess-precision=fast -funroll-all-loops -O3 -DNDEBUG
C++ compiler: /opt/rh/devtoolset-9/root/usr/bin/c++ GNU 9.3.1
C++ compiler flags: -mavx2 -mfma -Wno-missing-field-initializers -fexcess-precision=fast -funroll-all-loops -fopenmp -O3 -DNDEBUG
I wonder if it is because these versions(2016.4 and 2018) are installing using the C compiler in psiconda, which was created when I installed PSI4. I may need to fix my paths :S
ALso note the FFTW versions… I think the FFTW library is a different, less sophisticated one that was installed with PSI4?
Yes, get out of that psiconda environment. That build is a mess.
I am at a bit of a loss on that part since I thought by specifying the compiler location, it would boycott psi4conda, but that doesn’t seem to be the case… I am using
cmake .. -DGMX_BUILD_OWN_FFTW=ON -DGMX_GPU=OFF -DCMAKE_INSTALL_PREFIX=/apps/ -DCMAKE_C_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/cc -DCMAKE_CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++
but during make I can see it still going through psi4conda for compiler info… for example, a small piece (note, the install is 2022, but it is the same for other versions)
libtool: compile: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-cc -DHAVE_CONFIG_H -I. -I/apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2 -I../../.. -I /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 -isystem /home/bkelly/psi4conda/include -march=core-avx2 -mfma -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -MT n1fv_2.lo -MD -MP -MF .deps/n1fv_2.Tpo -c /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2/n1fv_2.c -fPIC -DPIC -o n1fv_2.o
libtool: compile: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-cc -DHAVE_CONFIG_H -I. -I/apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2 -I../../.. -I /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 -isystem /home/bkelly/psi4conda/include -march=core-avx2 -mfma -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -MT n1fv_3.lo -MD -MP -MF .deps/n1fv_3.Tpo -c /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2/n1fv_3.c -fPIC -DPIC -o n1fv_3.o
libtool: compile: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-cc -DHAVE_CONFIG_H -I. -I/apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2 -I../../.. -I /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 -isystem /home/bkelly/psi4conda/include -march=core-avx2 -mfma -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -MT n1fv_7.lo -MD -MP -MF .deps/n1fv_7.Tpo -c /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2/n1fv_7.c -fPIC -DPIC -o n1fv_7.o
libtool: compile: /home/bkelly/psi4conda/bin/x86_64-conda-linux-gnu-cc -DHAVE_CONFIG_H -I. -I/apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2 -I../../.. -I /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 -isystem /home/bkelly/psi4conda/include -march=core-avx2 -mfma -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/bkelly/psi4conda/include -MT n1fv_4.lo -MD -MP -MF .deps/n1fv_4.Tpo -c /apps/gromacs-2022/build/src/external/build-fftw/fftwBuild-prefix/src/fftwBuild/dft/simd/avx2/n1fv_4.c -fPIC -DPIC -o n1fv_4.o
I can also see it is grabbing BLAS and LAPACK from psi4conda
Found BLAS: /home/bkelly/psi4conda/lib/libmkl_intel_lp64.so;/home/bkelly/psi4conda/lib/libmkl_intel_thread.so;/home/bkelly/psi4conda/lib/libmkl_core.so;/home/bkelly/psi4conda/lib/libiomp5.so;-lpthread;-lm;-ldl
-- Looking for cheev_
-- Looking for cheev_ - found
-- Found LAPACK: /home/bkelly/psi4conda/lib/libmkl_intel_lp64.so;/home/bkelly/psi4conda/lib/libmkl_intel_thread.so;/home/bkelly/psi4conda/lib/libmkl_core.so;/home/bkelly/psi4conda/lib/libiomp5.so;-lpthread;-lm;-ldl;-lpthread;-lm;-ldl
You need to completely deactivate the environment. As I recall, Psi4 inserts itself into your startup scripts and automatically sets PATH information. You need to disable that. I have had similar problems compiling other software so I have had to disable this environment entirely to make them work.
I think I have it all sorted, and have gromacs-2018.6 running.
In the .bashrc I had to
#. $HOME/psi4conda/etc/profile.d/conda.sh
#conda activate
#. $HOME/psi4conda/etc/profile.d/conda.sh
#conda activate
then to compile, I did have to manually specify the SIMD… If I did not, it was None, this was odd, but, at least using the following, it worked
cmake .. -DGMX_BUILD_OWN_FFTW=ON -DGMX_GPU=OFF -DCMAKE_INSTALL_PREFIX=/apps/gromacs-2018.6 -DGMX_SIMD=AVX2_256
so now I have
GROMACS: gmx, version 2018.6
Executable: /apps/gromacs-2018.6/bin/gmx
Data prefix: /apps/gromacs-2018.6
Working dir: /apps/gromacs-2018.6/build
Command line:
gmx --version
GROMACS version: 2018.6
Precision: single
Memory model: 64 bit
MPI library: thread_mpi
OpenMP support: enabled (GMX_OPENMP_MAX_THREADS = 64)
GPU support: disabled
SIMD instructions: AVX2_256
FFT library: fftw-3.3.8-sse2-avx-avx2-avx2_128-avx512
RDTSCP usage: enabled
TNG support: enabled
Hwloc support: disabled
Tracing support: disabled
Built on: 2022-03-28 16:03:30
Built by: bkelly@c110370 [CMAKE]
Build OS/arch: Linux 3.10.0-1160.36.2.el7.x86_64 x86_64
Build CPU vendor: Unknown
Build CPU brand: Unknown
Build CPU family: 0 Model: 0 Stepping: 0
Build CPU features: Unknown
C compiler: /opt/rh/devtoolset-9/root/usr/bin/cc GNU 9.3.1
C compiler flags: -march=core-avx2 -O3 -DNDEBUG -funroll-all-loops -fexcess-precision=fast
C++ compiler: /opt/rh/devtoolset-9/root/usr/bin/c++ GNU 9.3.1
C++ compiler flags: -march=core-avx2 -std=c++11 -O3 -DNDEBUG -funroll-all-loops -fexcess-precision=fast
Thanks for the help!