Even I am facing a similar problem with 2020 versions. I have tried installing with combinations like cuda-9, 10, 11 with different versions of gcc like 5.5, 7.3 and 7.5. But, no luck. All the time, I get GPU detection deactivated only. Can you provide specific versions of the software you have used to compile it. I was able to use GPUs for the 2019 installations with same parameters used for 2020.
I found the gromacs 2020 with same parameters installation detects 2 x RTX 2080 Ti and it has 32 Xeon cores Gold 6130 processors with 64 threads. it is very strange. The only change is 1080 Ti driver version is 384.98 and 2080 Ti driver version is 410.93. I even tried compiling on skylake node with SIMD as AVX256. But, still it gives GPU detection deactivated on the broadwell nodes.
Please find my cmake command below.
My machine is E5 -2667V4 16 xeon cores with 32 threads with 2 x 1080 Ti cards.
cmake … -DGMX_GPU=on -DGMX_FFT_LIBRARY=fftw3 -DCMAKE_INSTALL_PREFIX="/apps/gromacs/2020.3/avx-256/" -DCMAKE_C_COMPILER="/apps/compilers/gcc/7.5/bin/gcc" -DCMAKE_CXX_COMPILER="/apps/compilers/gcc/7.5/bin/c++" -DBUILD_SHARED_LIBS=OFF -DCMAKE_PREFIX_PATH="/apps/libs/fftw/3.3.8"
Any help/ inputs to solve the problem will be appreciated.