BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG E 1.1
DTSTART;TZID=Europe/Stockholm:20190613T114500
DTEND;TZID=Europe/Stockholm:20190613T121500
UID:submissions.pasc-conference.org_PASC19_sess159_msa213@linklings.com
SUMMARY:Accelerating Eulerian Gyrokinetic Codes on the Latest GPU Platform
 s
DESCRIPTION:Minisymposium\nPhysics\n\nAccelerating Eulerian Gyrokinetic Co
 des on the Latest GPU Platforms\n\nAsahi, Ina, Maeyama, Idomura, Watanabe\
 n\nPlasma turbulence simulation for the magnetic fusion plasmas is often v
 ery costly, since we have to solve 5D kinetic equations. Even with several
  simplifications like adiabatic response of electrons or concentric circul
 ar magnetic geometry, peta-scale supercomputers are necessary for producti
 on runs. In order to model more realistic plasmas, we need to employ kinet
 ic electron models and deal with a more complex magnetic geometry. Both of
  them result in a significant increase of computational costs, which can o
 nly be managed with the upgrades of both hardware and software sides. As a
  first step to upgrade our codes to perform simulations on a foreseen exa-
 scale platform, we have ported gyrokinetic codes GT5D [Idomura, CPC, 2008]
  and GKV [Watanabe, NF, 2006] to GPU platforms in different approaches: Op
 enACC and CUDA. Both are Eulearian codes, but GT5D and GKV solve full-F an
 d delta-f gyrokinetic equations, respectively. Consequently, they employ d
 ifferent numerical approaches which result in different types of cost dist
 ributions: clear hot spot in GT5D and well-balanced costs in GKV. The form
 er is suitable for hot-spotting porting, while the latter requires porting
  of the entire code. In the talk, we demonstrate the acceleration of GT5D 
 with OpenACC-based porting and of GKV with CUDA-based porting.
END:VEVENT
END:VCALENDAR