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:20230831T095747Z
LOCATION:Sanada I
DTSTART;TZID=Europe/Stockholm:20230628T153000
DTEND;TZID=Europe/Stockholm:20230628T160000
UID:submissions.pasc-conference.org_PASC23_sess138_msa271@linklings.com
SUMMARY:Tradeoffs in Low-Power Accelerators Design for Large-Scale Interfe
 rometers
DESCRIPTION:Minisymposium\n\nDenisa Constantinescu, Benoît Denkinger, Migu
 el Peon Quiros, and David Atienza (EPFL)\n\nLarge-scale scientific infrast
 ructures like SKAO—the world’s largest radio observatory for the coming de
 cades—are generating massive-scale data streams of multi-Tb/s to be proces
 sed using complex interferometry algorithms. Concretely, SKAO is expected 
 to generate over 710 petabytes per year of data products with a limited en
 ergy budget caped to 1 MWatt per site. We address the computing sustainabi
 lity of this scientific endeavour from a holistic hardware-software codesi
 gn point of view. Our work analyses the tradeoffs of using novel reconfigu
 rable domain-specific processor designs tailored for the interferometry do
 main in an application-specific multiprocessor system. We aim to design cu
 stom energy-efficient hardware architectures and heterogeneous scheduling 
 techniques that will help SKAO meet its accuracy, energy efficiency, perfo
 rmance, and scalability requirements while keeping in check the overall co
 st. One of our biggest challenges is that high-performance software decisi
 ons cannot be fixed early in SKAO construction. The adoption and usefulnes
 s of custom hardware designs depend on how well we can bridge the programm
 ability and performance gap between general-purpose processors and domain-
 specific processors for the interferometry domain. Thus, a top-down and bo
 ttom-up approach in collaboration with astronomers and computer architects
  will be required to converge to impactful, low-power, high-performance co
 mputing solutions.\n\nDomain: Computer Science, Machine Learning, and Appl
 ied Mathematics 
\n\nSession Chair: Emma Tolley (EPFL)
END:VEVENT
END:VCALENDAR