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:20220812T074335Z
LOCATION:Singapore Room
DTSTART;TZID=Europe/Stockholm:20220629T150000
DTEND;TZID=Europe/Stockholm:20220629T153000
UID:submissions.pasc-conference.org_PASC22_sess155_msa105@linklings.com
SUMMARY:MG-GCN: Scalable Multi-GPU GCN Training Framework
DESCRIPTION:Minisymposium\n\nMG-GCN: Scalable Multi-GPU GCN Training Frame
 work\n\nBalin, Sancak, Catalyurek\n\nFull batch training of Graph Convolut
 ional Network (GCN) modelsis not feasible on a single GPU for large graphs
  containing tens of millionsof vertices or more. Recent work has shown tha
 t, for the graphs used in themachine learning community, communication bec
 omes a bottleneck and scalingis blocked outside of the single machine regi
 me. Thus, we propose MG-GCN,a multi-GPU GCN training framework taking adva
 ntage of the high-speedcommunication links between the GPUs present in mul
 ti-GPU systems.MG-GCN employs multiple High-Performance Computing optimiza
 tions,including efficient re-use of memory buffers to reducethe memory foo
 tprint of training GNN models, as well as communication andcomputation ove
 rlap. These optimizations enable execution on larger datasets,that general
 ly do not fit into memory of a single GPU in state-of-the-art implementati
 ons. Furthermore, they contribute to achieve superior speedup compared to 
 thestate-of-the-art. For example, MG-GCN achieves super-linear speedup wit
 h respectto DGL, on the Reddit graph on both DGX-1 (V100) and DGX-A100.\n\
 nDomain: Computer Science and Applied Mathematics
END:VEVENT
END:VCALENDAR