BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000712Z
LOCATION:505
DTSTART;TZID=America/Denver:20231112T095600
DTEND;TZID=America/Denver:20231112T100100
UID:submissions.supercomputing.org_SC23_sess421_ws_rsdha109@linklings.com
SUMMARY:NVMe-Backed GNN Training on GPU Leveraging a Paged UVM Memory Syst
 em
DESCRIPTION:Benjamin Wagley (Colorado School of Mines), Pak Markthub (NVID
 IA Corporation), and Bo Wu and Mehmet Belviranli (Colorado School of Mines
 )\n\nGraph Neural Networks (GNNs) are powerful machine learning models tha
 t learn on graph data by extracting embeddings that represent vertex and e
 dge features, as well as graph topology. With graph data scale increasing,
  and high memory pressure generated from GNN feature data, we turn to out-
 of-core training methods on many real world graphs. Current state-of-the-a
 rt methods for large-graph GNN training leverage mini-batches, distributed
  or parallel environments, and memory-aware partitioning and sampling.  Th
 ese methods however require custom training architectures and pipelines. H
 ere, we propose Kirin, a framework for large-graph out-of-core training on
  a single machine with a single GPU on pre-sampled graphs. Kirin leverages
  Dragon-direct, allowing for NVMe-backed tensors for out-of-core training 
 through driver managed allocations. Building on UVM, Dragon-direct utilize
 s a page-based unified memory system, resulting in memory-management that 
 is largely invisible to the user. We showcase Kirin and analyze its perfor
 mance and effectiveness for GNN workloads.\n\nTag: Accelerators, Edge Comp
 uting, Heterogeneous Computing\n\nRegistration Category: Workshop Reg Pass
 \n\nSession Chairs: Ali Akoglu (University of Arizona), Mehmet E Belviranl
 i (Colorado School of Mines), and Seyong Lee (Oak Ridge National Laborator
 y (ORNL))\n\n
END:VEVENT
END:VCALENDAR