THUWAL — King Abdullah College of Science and Expertise (KAUST), Saudi Arabia’s main analysis establishment, and Cerebras Methods, a US-based pioneer in accelerating generative AI, have collaborated on seismic processing, utilizing the Condor Galaxy AI supercomputer, which was constructed by Cerebras and their strategic associate G42 out of Abu Dhabi, incomes them a spot as finalists for the esteemed 2023 Gordon Bell Prize.was
The Gordon Bell Prize is a prestigious award named after Gordon Bell, a pioneer within the area of pc science and high-performance computing. The prize, in its 35th 12 months, is awarded yearly to acknowledge excellent achievements within the area of high-performance computing (HPC).
The groundbreaking analysis by KAUST and Cerebras revolves round multi-dimensional seismic processing, a self-discipline very important for understanding Earth’s assets and steering the world towards a sustainable, low-carbon future. This revolutionary collaboration harnessed the ability of Cerebras Methods’ cutting-edge expertise to ship seismic processing at speeds and accuracy ranges beforehand thought-about unattainable.
Key to this achievement is the event of a Tile Low-Rank Matrix-Vector Multiplication (TLR-MVM) kernel, uniquely tailor-made to use the superior structure of Cerebras CS-2 programs throughout the Condor Galaxy AI supercomputer. The consequence was a seismic processing resolution that reached production-grade accuracy with a report sustained reminiscence bandwidth of 92.58 petabytes per second (PB/s) with bandwidth usually being the bottleneck in in the present day’s processor-rich architectures. This achievement underscores the transformative potential of AI-customized architectures in ushering in a brand new period of seismic algorithms.
Andrew Feldman, the co-founder and CEO of Cerebras Methods, expressed his delight on this groundbreaking collaboration, stating, “In partnership with KAUST researchers and utilizing the Condor Galaxy AI supercomputer, we’re honored to be acknowledged by Gordon Bell for setting a brand new report in what is feasible for multi-dimensional seismic processing. This work will unlock world-changing developments throughout local weather and climate modeling, computational astronomy, wi-fi communication, seismic imaging, and extra.”
Lead writer Hatem Ltaief, Principal Analysis Scientist of KAUST’s Excessive Computing Analysis Heart (ECRC), expressed his pleasure with the analysis achievement: “Disaggregated reminiscence requires fine-grained algorithmic improvements. Working with Cerebras engineers to deploy them and extract the {hardware}’s full potential was a dream.”
ECRC Director David Keyes added, “It’s thrilling to find the flexibility of wafer-scale {hardware} past neural community coaching for which it was conceived and optimized. We be part of different examples of such architectural crossover in Gordon Bell Prize historical past.”
Different KAUST co-investigators embrace Professor Matteo Ravasi of Earth and Environmental Sciences and Engineering and 2023 KAUST Pc Science PhD graduate Yuxi Hong, who now workout routines his HPC expertise in Lawrence Berkeley Nationwide Laboratory’s Exascale Computing Program.
The work carried out by KAUST and Cerebras, and executed on Condor Galaxy -1, the AI Supercomputer constructed by Cerebras and G42, demonstrates the transformative potential of Cerebras CS-2 as a substitute that may obtain record-breaking efficiency for traditionally memory-bound functions.
Extra info on this multi-dimensional seismic processing workload will be discovered at https://repository.kaust.edu.sa/handle/10754/694388. This analysis was made potential by G42’s grant of system time on the Condor Galaxy 1 AI supercomputer.