Скачать или смотреть GPU-Accelerated Electronic Structure-Based Molecular Dynamics (ISC21, NVIDIA Virtual Booth)

Robert Schade presents our paper:

Enabling GPU-Accelerated Electronic Structure-Based Molecular Dynamics Simulations with Hundreds of Millions of Atoms

By Robert Schade, Tobias Kenter, Hossam Elgabarty, Michael Lass, Ole Schütt, Alfio Lazzaro, Hans Pabst, Stephan Mohr, Jürg Hutter, Thomas D. Kühne, Christian Plessl

Preprint: https://arxiv.org/abs/2104.08245
Link to event page: https://resources.nvidia.com/en-us-vi... Schade presents our paper:

Enabling GPU-Accelerated Electronic Structure-Based Molecular Dynamics Simulations with Hundreds of Millions of Atoms

By Robert Schade, Tobias Kenter, Hossam Elgabarty, Michael Lass, Ole Schütt, Alfio Lazzaro, Hans Pabst, Stephan Mohr, Jürg Hutter, Thomas D. Kühne, Christian Plessl

Preprint: https://arxiv.org/abs/2104.08245
Link to event page: https://resources.nvidia.com/en-us-vi...

Abstract:

We push the boundaries of electronic structure based molecular dynamics (AIMD) beyond 100 million atoms. We combine innovations in linear-scaling AIMD, efficient and approximate sparse linear algebra, low and mixed-precision floating-point computation on tensor cores of NVIDIA GPUs, and a compensation scheme for the errors introduced by numerical approximations. The core of our work is the non-orthogonalized local submatrix (NOLSM) method, which scales very favorably to massively parallel systems and is ideally suited to hardware accelerators. We demonstrate that the NOLSM method achieves 324 PFLOP/s in mixed FP16/FP32 precision corresponding to 67.7% of peak on 1536 NVIDIA A100 GPUs.Robert Schade presents our paper:

Enabling GPU-Accelerated Electronic Structure-Based Molecular Dynamics Simulations with Hundreds of Millions of Atoms

By Robert Schade, Tobias Kenter, Hossam Elgabarty, Michael Lass, Ole Schütt, Alfio Lazzaro, Hans Pabst, Stephan Mohr, Jürg Hutter, Thomas D. Kühne, Christian Plessl

Preprint: https://arxiv.org/abs/2104.08245
Link to event page: https://resources.nvidia.com/en-us-vi...

Abstract:

We push the boundaries of electronic structure based molecular dynamics (AIMD) beyond 100 million atoms. We combine innovations in linear-scaling AIMD, efficient and approximate sparse linear algebra, low and mixed-precision floating-point computation on tensor cores of NVIDIA GPUs, and a compensation scheme for the errors introduced by numerical approximations. The core of our work is the non-orthogonalized local submatrix (NOLSM) method, which scales very favorably to massively parallel systems and is ideally suited to hardware accelerators. We demonstrate that the NOLSM method achieves 324 PFLOP/s in mixed FP16/FP32 precision corresponding to 67.7% of peak on 1536 NVIDIA A100 GPUs.ise-5?lx=IwPvNh

Abstract:

We push the boundaries of electronic structure based molecular dynamics (AIMD) beyond 100 million atoms. We combine innovations in linear-scaling AIMD, efficient and approximate sparse linear algebra, low and mixed-precision floating-point computation on tensor cores of NVIDIA GPUs, and a compensation scheme for the errors introduced by numerical approximations. The core of our work is the non-orthogonalized local submatrix (NOLSM) method, which scales very favorably to massively parallel systems and is ideally suited to hardware accelerators. We demonstrate that the NOLSM method achieves 324 PFLOP/s in mixed FP16/FP32 precision corresponding to 67.7% of peak on 1536 NVIDIA A100 GPUs.