Highlights

Below is a selected list of highlights generated for VTK-m related R&D.

Mode Analysis with Poincaré Plots The accelerated Poincaré plotting implemented with VTK-m has been enhanced to enable mode analysis. This mode analysis is allowing fusion scientists to understand homoclinic tangles, a feature in the magnetic confinement in tokamak fusion reactors that is the theoretical cause of significant electron energy loss.
WarpX on Frontier with In Situ Visualization VTK-m was used for an In Situ visualization of a 4.6 billion element WarpX laser wakefield simulation running on Frontier with 4416 GPUs on 552 nodes.
VTK-m Rendering at Scale on Frontier VTK-m rendering was demonstrated at scale on the Frontier supercomputer. Frontier is the world’s first exascale supercomputer, and a VTK-m rendering benchmark was able to run on the majority of its compute nodes.
Uncertainty Visualization of Marching Cubes FunMC², a filter for computing probable contours in the face of uncertainty, was implemented in VTK-m. Uncertainty adds significant computational load to operations like Marching Cubes, which had to consider multiple cases simultaneously. These algorithms were accelerated in VTK-m and demonstrated large speedups on GPU systems.
Accelerating ParaView on Crusher ParaView was demonstrated running on Crusher and using VTK-m to accelerate operations on its GPUS. Crusher is the ORNL early-access cluster for Frontier, the first supercomputer to report over an exaFLOPS of computation. These machines rely on GPUs for the bulk of their computation power. When ParaView was run on crusher, select filters are automatically replaced with VTK-m implementations that run accelerated on the GPU.
Accelerated Poincaré Plots VTK-m has reduced the time to compute a Poincaré plot from 2 hours down to 7 minutes. Poincaré plots are instrumental to the WDMApp project to understand the energy transport that occurs as energetic particles interact with components in the ITER reactor. A single Poincaré plot shows the cross section of many particle traces that iterate several times around the fusion reactor, which is computationally intensive to generate. This efficiency was achieved by leveraging VTK-m’s GPU particle tracing capabilities.
VTK-m Demonstrates Performance Portability A literature survey reveals VTK-m algorithms run as efficiently as targeted implementations. When comparing an algorithm implemented in VTK-m to an algorithm written for a specific device, one would expect the targeted algorithm to run faster than the general VTK-m implementation. But in reviewing peer-reviewed papers providing such a comparison, on average the general VTK-m algorithm performed about the same as the targeted implementation.
In Situ Vector Field Reduction Using Lagrangian Basis Flows Lagrangian Basis Flows are an alternate representation of flow fields that record displacement rather than velocity. VTK-m is used to compute Lagrangian Basis Flows in situ with simulation to reduce I/O demands and improve post hoc analysis accuracy.
In Situ Visualization in WDMApp Using VTK-m Integrating VTK-m with simulations for the planned experimental ITER fusion reactor. Images are generated in situ while the simulation is running.
Flow Visualization in WarpX Laser Wakefield Simulations Unlike for a typical fluid velocity field, particle trajectories for relativistic plasma and accelerator physics must be inferred from electromagnetic fields as well as the particles’ momentum. VTK-m’s particle tracing was customize for the specifics of WarpX particle trajectories.
In Situ/In Transit Visualization with WarpX VTK-m was used to perform in situ visualization and rendering during a simulation of a 10-stage laser-wakefield accelerator. The simulation was run on 600 GPUs on Summit at OLCF. This visualization was featured during the keynote talk of ISAV 2020.
Large-Scale In Situ Visualization of Raleigh-Taylor Instability with Ascent and VTK-m Coupled to a simulation through Ascent, VTK-m visualized a 97.8 billion element Raleigh-Taylor instability simulation on 4096 nodes of the Sierra computer at Lawrence Livermore National Laboratory.
Scientific Discovery via Visualization Using Accelerated Computing A short retrospective and timeline of the origins of VTK-m.
Enabling in situ visualization of petascale time-dependent CFD simulations Coupled to a PyFR CFD simulation through ParaView/Catalyst, VTK-m was used to visualize flow over a low pressure turbine linear cascade. The simulation was done in situ with the simulation with the data remaining entirely on the GPU for maximum efficiency and performance.
Live Demonstration of In Situ Visualization on Accelerator Processors A demonstration of live in situ visualization on the Titan supercomputer. PyFR, a CFD simulation that was coupled with ParaView/Catalyst, ran on 256 GPUs of the Titan supercomputer at Oak Ridge National Laboratory, visualized its results with VTK-m, and displayed its results on a live client application running at the SC 2015 show floor.