BEGIN:VCALENDAR
VERSION:2.0
METHOD:PUBLISH
CALSCALE:GREGORIAN
PRODID:-//WordPress - MECv7.34.2//EN
X-ORIGINAL-URL:https://ecocloud.epfl.ch/
X-WR-CALNAME:EcoCloud
X-WR-CALDESC:
X-WR-TIMEZONE:Europe/Paris
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20260329T030000
RRULE:FREQ=YEARLY;BYMONTH=03;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20261025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=4SU
END:STANDARD
END:VTIMEZONE
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-PUBLISHED-TTL:PT1H
X-MS-OLK-FORCEINSPECTOROPEN:TRUE
BEGIN:VEVENT
CLASS:PUBLIC
UID:MEC-8417b4b7a64929b9e6dea71c8c247959@ecocloud.epfl.ch
DTSTART;TZID=Europe/Paris:20261008T143000
DTEND;TZID=Europe/Paris:20261008T150000
DTSTAMP:20260720T172529Z
CREATED:20260720
LAST-MODIFIED:20260720
PRIORITY:5
SEQUENCE:14
TRANSP:OPAQUE
SUMMARY:Beyond the Thermal Wall: Intelligent Cooling Architecture for the AI Era
DESCRIPTION:\nMaurizio Miozza, CCO of GemaTEG\n\nAI workloads are driving chip power densities beyond the reach of conventional liquid-cooling architectures. As GPUs exceed 2 kW and chiplet designs integrate increasingly dense HBM memory stacks, traditional thermal solutions are hitting hard limits, with direct consequences for energy consumption, infrastructure cost, and datacenter scalability.\nThe presentation examines chiplet-level independent thermal control as an architectural response to the constraints of conventional cooling. Rather than treating a compute package as a single thermal entity, independent management of GPU dies and HBM memory stacks enables cooling resources to be allocated where and when they are needed most. The approach supports facility water temperatures of 40–45°C without mechanical refrigeration, reducing cooling-related energy consumption, while preserving silicon performance.\nBy introducing localized temperature control, the architecture adds a new dimension to datacenter operation. Combined with continuous telemetry and infrastructure-wide orchestration, thermal management evolves from a passive heat-rejection function into an active optimization layer, capable of anticipating thermal excursions, optimizing cooling asset utilization, and minimizing downtime through predictive maintenance.\nResults are drawn from DaTEG® 2.0, a patented implementation of the architecture developed by GemaTEG, and from deployment experience with the GemaTEG Supervisor™ platform. The findings suggest that closing the gap between AI compute density and cooling infrastructure capability will require not just improved heat removal, but coordinated thermal intelligence designed to evolve alongside the hardware it serves.\n \nMaurizio Miozza, Co-Founder and CCO of GemaTEG, has three decades of expertise in semiconductor technologies, especially in silicon. His approach integrates business development with engineering, guided by a mantra of Integration.\nMaurizio leads efforts on CPUs and GPUs, overseeing teams in Italy, Switzerland, and the USA, and fostering key technical and strategic partnerships. His dual role in research and business led to the creation of a groundbreaking TOF detector and drove UmbraGroup’s expansion in the US. Renowned for market foresight, Maurizio’s leadership focuses on risk management, trust-building, and mindset transformation, turning potential into achievement.\n
URL:https://ecocloud.epfl.ch/mcevents/intelligent-cooling/
CATEGORIES:event2026
ATTACH;FMTTYPE=image/jpeg:https://ecocloud.epfl.ch/wp-content/uploads/2026/07/miozzathumbnail.jpg
END:VEVENT
END:VCALENDAR