Registered Data

[02001] GPU batched sparse solver for XGC fusion plasma collision operator

• Session Time & Room : 4E (Aug.24, 17:40-19:20) @G305
• Type : Contributed Talk
• Abstract : Batched linear solvers solve many small related but independent problems. They are beneficial for GPUs, which require substantial amounts of work to operate efficiently. The XGC gyrokinetic particle-in-cell code for modeling magnetically confined fusion plasma devices employs a LAPACK CPU solver for the collision operator. We describe how Ginkgo's batched solver can be integrated into the collision operator and accelerate the simulation process. We present comparisons for the solve times on A100 GPUs with CPUs.
• Classification : 15-04, 35-04, 76-10
• Format : Talk at Waseda University
• Author(s) :
  • Paul Lin (Lawrence Berkeley National Laboratory)
  • Aditya Kashi (Oak Ridge National Laboratory)
  • Pratik Nayak (Karlsruhe Institute of Technology)
  • Dhruva Kulkarni (Lawrence Berkeley National Laboratory)
  • Aaron Scheinberg (Jubilee Development)
  • Hartwig Anzt (University of Tennessee)