Not stupid at all, but I think most of the information on this is in the Gromacs-LS manual if you give it a close read, including the FAQ section -
GROMACS-LS is very slow because it doesn’t use most of the optimizations that Gromacs has built in (no SIMD acceleration on CPUs, no GPUs) and it runs on only one core. Another reason is that there’s no PME for electrostatics, so you have to use a larger short-range electrostatic cutoff to get accurate forces - have your read about that parameter in the Gromacs-LS doc? It’s pretty important to get right.
The good news is that if you have a cluster or computer with multiple cores, you can break up that trr file into chunks (Gromacs-LS provides a tool for this), use a bash script to call the program on each chunk (one for each cpu core you have is probably best), and then combine the results at the end during the processing step.