BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:What impact can Integrated Photonics have on data center architect
 ure?
DTSTART:20171208T110000
DTEND:20171208T120000
DTSTAMP:20260510T101039Z
UID:7ad44d011aa5beffdc27b99ba5c76909b7a0c97dcf7c152ad4f604a4
CATEGORIES:Conferences - Seminars
DESCRIPTION:Sébastien Rumley\, Research Scientist in the Lightwave Resear
 ch Laboratory\, Columbia University\, New York\nBio: Prior joining Columbi
 a in 2012\, he was at EPFL where he got his M.S and Ph.D degrees in commun
 ication systems. His research focuses on multilayer\, cross-scale modeling
  and optimization of large scale interconnection networks. This includes a
 nalysis of nanophotonic devices and the integration thereof in next genera
 tion computing systems\, network topology design and dimensioning\, charac
 terization of data-movement requirements and end-to-end evaluation of inte
 rconnect power consumption. He is also interested in novel\, post-Moore’
 s Law computer architectures relying on photonic connectivity. Dr. Rumley 
 is co-author of over 70 publications in the fields of optical interconnect
 s and optical networks. He has served or is serving as Program Committee M
 ember for the SuperComputing\, ISC-HPC and NETWORKS conferences\, as well 
 as for various workshops (HiPINEB\, Exacomm\, HUCAA\, AISTECS\, OPTICS\, H
 CPM\, IA^3). He is a co-recipient of the best-student paper award of the 2
 016 SuperComputing edition.\nAbstract: Big data analytics applications tha
 t rely on machine and deep learning techniques are seismically changing th
 e landscape of datacenter architectures. Image or speech recognition tasks
  are now routinely executed by datacenters. These tasks\, however\, demand
  much more computing power than traditional ones: tens of GFLOP for recogn
 izing one image\, compared to tens of MFLOP for an SQL query. As a result\
 , Graphics processing units (GPU) are literally invading datacenters\, and
  will likely be followed by highly machine learning optimized hardware as 
 Google’s Tensor Processing Unit (TPU). Yet the emergence of this novel c
 omputing hardware is only one facet of the ongoing data center transformat
 ion. Another transformation must occur in terms of interconnections. Hence
 \, the performance of machine learning optimized chips is increasingly lim
 ited by off-chip communications. The concept of disaggregated datacenter\,
  proposed by many actors (HPE’s The Machine/Moonshot\, Intel RSD\, Open 
 Compute) as a way to use IT hardware more efficiently\, is also bumping on
 to the high cost and power consumption of interconnects. In general\, inte
 r-component communications are increasingly acting as a bottleneck to data
 center performance.\n\nIntegrated photonics is (and has been) frequently e
 voked as a way to alleviate major bandwidth bottlenecks. And silicon photo
 nics has been for years presented as the default path to low cost\, low po
 wer photonics. In this talk\, we will review the progresses realized in si
 licon photonics transceiver fabrication in the last years\, and show how t
 hese transceivers can be closely integrated with conventional chips. The c
 oncept of Optically Connected Multi Chip Module (OC-MCM)\, composed of a s
 ilicon interposer with embedded optical connectivity\, and carrying an hig
 h performance ASIC as a CPU\, a GPU\, an FPGA\, or several memory dies\, w
 ill be presented. The expected figures of merit in connectivity terms of s
 uch OC-MCMs will be summarized. We will then envisage how the datacenter a
 rchitecture can be reorganized around discrete compute or memory building 
 blocks\, each block being assembled as one OC-MCM. Distance independence o
 f optics allows in principle a CPU block to directly communicate with a me
 mory block located several meters apart. Treating such a remote memory as 
 a local one can be valuable when executing tasks with high memory capacity
  or bandwidth needs. A CPU block cannot be directly connected to every mem
 ory block\, however\, so the right trade-off in terms of connectivity must
  be identified. More generally\, we will discuss the impact this OC-MCM ap
 proach can have on system design. Finally\, we will see how the connection
 s between OC-MCM can be reconfigured by means of optical switches. We will
  show that this reconfigurability can be exploited to adapt the hardware a
 rchitecture to specific need\, e.g. machine learning model training\, but 
 also point out some of the challenges it raises\, in particular in terms o
 f scheduling.
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR