KTH Royal Institute of Technology
Company Website
Full Time
Closes: 8 January 2024
Applications have closed
Postdoc in GPU algorithms for molecular dynamics with GROMACS

We are looking for an HPC expert to contribute to one or multiple projects related to the GROMACS molecular simulation software. GROMACS is one of the most important HPC applications in the world and runs on all modern CPU and GPU architectures. GROMACS uses a hierarchical parallelization with a wide range of parallelization techniques from its SIMD library to GPU abstraction layer with CUDA, SYCL, and OpenCL backends, OpenMP, MPI. As the computational power increases, new algorithms are needed to make efficient use of the hardware. We are collaborating with world-leading experts in academia and at hardware vendors to develop efficient and scalable algorithms for molecular dynamics (MD). This work will be related to collaborative, international projects potentially including fast multipole electrostatics, fine-grained message-driven algorithms and machine learning potentials. All projects mainly target GPUs. Although MD is the base algorithm, these projects do not require deep knowledge of MD. But as MD is the target, iteration rates are often measured in hundreds of microseconds, which means that latency reduction and critical path optimizations are crucial. The GROMACS environment at KTH is shared between the Applied Physics department, where MD experts are located, and the high performance compute center, PDC, where HPC experts are located.

Results of this work will have high impact, both through the world wide use of the GROMACS package as well as through scientific publications.