# Gmx 2025.4 in Ubuntu 26.04

**URL:** <https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437>\
**Category:** User discussions\
**Created:** [May 5, 2026, 3:57pm UTC](https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437 "2026-05-05T15:57:30Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![slovas](https://avatars.discourse-cdn.com/v4/letter/s/e274bd/32.png) [@slovas](https://gromacs.bioexcel.eu/u/slovas)\
**Post date:** [May 5, 2026, 3:57pm UTC](https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437/1 "2026-05-05T15:57:30Z")

</div>

GROMACS version:2025.4  
GROMACS modification: No

I have updated my computer to Ubuntu 26.04, and have 7.0.0-15-generic kernel, cuda-13.2  
and compiled gmx 2025.4 using gcc13 and nvcc 13.1.r13.1/compiler.37061995\_0.

I can prepare the input (pdb2gmx, editconf, solvate…) with the newly compiled code, but when I issue the gmx mdrun… it just hangs and do nothing.

Here are the commands I used for compiling the code:

cmake .. -DCMAKE\_C\_COMPILER=gcc-13 -DCMAKE\_CXX\_COMPILER=g+±13 -DCMAKE\_CUDA\_HOST\_COMPILER=/usr/bin/g+±13 -DCMAKE\_CUDA\_COMPILER=/usr/local/cuda-13.1/bin/nvcc -DCUDAToolkit\_ROOT=/usr/local/cuda-13.1 -DGMX\_SIMD=AVX2\_256 -DGMX\_GPU=CUDA -DCMAKE\_CUDA\_ARCHITECTURES=120 -DGMX\_OPENMP=ON -DGMX\_MPI=OFF -DGMX\_DOUBLE=OFF -DGMX\_BUILD\_OWN\_FFTW=ON -DBUILD\_SHARED\_LIBS=ON -DGMX\_HWLOC=ON -DGMX\_DEFAULT\_SUFFIX=OFF -DCMAKE\_INSTALL\_PREFIX=/opt/gromacs-2025 -DCMAKE\_BUILD\_TYPE=Release

make -j 20 \>& make.log &  
sudo make install -j 20

I should add that this was working all right with Ubuntu 25.10

I would appreciate any suggestions.

Sandor

---

<div class="post-metadata">

**Author:** ![al42and](https://dub1.discourse-cdn.com/flex017/user_avatar/gromacs.bioexcel.eu/al42and/32/1393_2.png) [@al42and](https://gromacs.bioexcel.eu/u/al42and)\
**Post date:** [May 5, 2026, 5:27pm UTC](https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437/2 "2026-05-05T17:27:36Z")

</div>

Hi Sandor,

By “it just hangs and do nothing” do you mean it does not even print the version header? Is the `md.log` file completely empty?

Which command-line options are you using? Have you tried running without GPU (do `export CUDA_VISIBLE_DEVICES=` before launching `gmx mdrun`; to revert do `unset CUDA_VISIBLE_DEVICES`)?

Sidenote 1: `-DCMAKE_CUDA_ARCHITECTURES` is only supported in GROMACS 2026. For 2025 you should use `-DGMX_CUDA_TARGET_SM=120`. Should not matter for your problem, though.; without the correct option it will just take slightly longer to compile.

Sidenote 2: You say “cuda-13.2”, but the command line shows paths to cuda-13.1. Which one is the typo?

---

<div class="post-metadata">

**Author:** ![slovas](https://avatars.discourse-cdn.com/v4/letter/s/e274bd/32.png) [@slovas](https://gromacs.bioexcel.eu/u/slovas)\
**Post date:** [May 5, 2026, 5:43pm UTC](https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437/3 "2026-05-05T17:43:09Z")

</div>

Not even a log file created. It does not show up in nvidia-smi

ok the nvidia-driver is 13-2, but the available cuda is 13-1

---

<div class="post-metadata">

**Author:** ![slovas](https://avatars.discourse-cdn.com/v4/letter/s/e274bd/32.png) [@slovas](https://gromacs.bioexcel.eu/u/slovas)\
**Post date:** [May 8, 2026, 8:44pm UTC](https://gromacs.bioexcel.eu/t/gmx-2025-4-in-ubuntu-26-04/13437/4 "2026-05-08T20:44:34Z")

</div>

In the previous cuda versions I always used -DGMX\_HWLOC=ON option and that usually gave me ~5% performance improvement. With the CUDA driver: 13.20 and CUDA runtime: 13.10 the hwloc should off (I think this is the default) and everything is working fine.