BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.coscup.org//coscup-2026//talk//BGYZ3B
BEGIN:VTIMEZONE
TZID:CST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:CST
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-coscup-2026-BGYZ3B@pretalx.coscup.org
DTSTART;TZID=CST:20260809T112000
DTEND;TZID=CST:20260809T115000
DESCRIPTION:GPUs are expensive. Kubernetes doesn't share them well yet\, DR
 A is still work in progress.\n\nHAMi: Heterogeneous GPU sharing for Kubern
 etes.\n\nWhat I'll Tell You:\n\nHow it hijacks CUDA calls without touching
  your app code\nWhy memory isolation matters\nSome real production use cas
 es\nReal Results: Teams cut GPU costs 40-60%\n\nWho Should Show Up: K8s op
 erators\, platform engineers\, anyone watching GPUs sit idle\n\ngithub.com
 /Project-HAMi/HAMi
DTSTAMP:20260713T142747Z
LOCATION:TR214
SUMMARY:Running Multiple AI Workloads on One GPU with HAMi: Architecture an
 d Gotchas - Reza jelveh
URL:https://pretalx.coscup.org/coscup-2026/talk/BGYZ3B/
END:VEVENT
END:VCALENDAR