Introduction GPU Servers Research Conclusion GPU Servers for Research in Quantum Fluids L. Galantucci HPC & Quantum Summit QEII Centre, London, 5 February 2019
Introduction GPU Servers Research Conclusion Overview Introduction 1 GPU Servers 2 Research 3 Conclusion 4
Introduction GPU Servers Research Conclusion Overview Introduction 1 GPU Servers 2 Research 3 Conclusion 4
Introduction GPU Servers Research Conclusion JQC Durham - Newcastle 22 staff 23 Post Doc 28 PhD Grants ∼ 15M£(EPSRC)
Introduction GPU Servers Research Conclusion Newcastle University School of Mathematics, Statistics and Physics 9 staff 4 Post Doc 5 PhD Bose Einstein Quantum Quantum Superfluid Condensates Helium Turbulence Vortices
Introduction GPU Servers Research Conclusion Quantum fluids Group School of Mathematics, Statistics and Physics Prof. Barenghi Dr. Parker Quantum Vortex Reconnections in Trapped Atomic BECs EP/R005192/1 10/2017 - 09/2020 374,496 £ ∼ 45,000 £ equipment
Introduction GPU Servers Research Conclusion Overview Introduction 1 GPU Servers 2 Research 3 Conclusion 4
Introduction GPU Servers Research Conclusion Tesla V100 GPUs Server comparison BEC IBM POWER9 GPU Server experiments in Trento 2 × 16-core 2.60 GHz 4 × 10 5 → 4 × 10 7 atoms POWER9 × 100 computing power 256 GB DDR4 Memory memory � 64 GB 4x NVIDIA Tesla V100 GPU past: 24 CPU node, 128 GB NVLINK SXM2 Tesla V100 GPUs 5120 cores 7.850 GFLOPS 16 GB Memory [NVIDIA Tesla V100 SXM2 Module] HBM2 Mem → 900 GB/sec
Introduction GPU Servers Research Conclusion NVLINK 2.0 NVIDIA - IBM NVIDIA - INTEL NVLINK 2.0: 300 GB/s NVLINK 1.0: 160 GB/s - GPU - GPU - GPU - GPU PCIe: 32GB/s GPU - CPU - GPU - CPU
Introduction GPU Servers Research Conclusion Overview Introduction 1 GPU Servers 2 Research 3 Conclusion 4
Introduction GPU Servers Research Conclusion Quantum Vortex Reconnections Quantum vortices one-dimensional objects hollow core source of rotational motion Importance of Reconnections redistribute energy quantum turbulence dissipate energy
Introduction GPU Servers Research Conclusion Plasmas Magnetic Flux Tubes reconnections of magnetic field lines [Zhike et al. , Nat. Com. (2016)] anomalous heating of solar corona [Cirtain et al. , Nature (2013)] explosive events, solar and stellar flares [Che et al. , Nature (2011)]
Introduction GPU Servers Research Conclusion Polymers and DNA long chains of atoms [Marenduzzo et al. , PNAS (2009)] [Marenda et al. , Nat Comm (2018)]
Introduction GPU Servers Research Conclusion Optical beams lines of zero intensity optical vortices phase singularities [Dennis et al. , Nature (2010)]
Introduction GPU Servers Research Conclusion Nematic Liquid Crystals Topological Defects [Chuang et al. , Science (1991)]
Introduction GPU Servers Research Conclusion Classical and Quantum Fluids vortices [Hussain et al. , Phys. Fluids (2011)] [LG et al. , subm. PNAS (2018)] vortex tubes vortex lines
Introduction GPU Servers Research Conclusion Quantum Vortex Reconnections Visualization collaboration with BEC Centre, Trento core ∼ 0.5 µ m, size ∼ 50 µ m innovative stroboscopic technique orientation of vortices rebound reconnection
Introduction GPU Servers Research Conclusion Quantum Vortex Reconnections Universal aspects scaling distance δ ( t ) energy dissipation ǫ Controlled Vortex Generation laser beams atomtronics
Introduction GPU Servers Research Conclusion Overview Introduction 1 GPU Servers 2 Research 3 Conclusion 4
Introduction GPU Servers Research Conclusion Summary IBM POWER9 GPU Server 4 × Tesla V100 GPUs NVLINK 2.0 speed up ∼ 80 × Quantum Vortex Reconnections visualization fundamental aspects controlled vortex generation
Recommend
More recommend