Large Language Models: Finding Their Place in the HPC Ecosystem

Key findings from a recent Hyperion Research study indicate that HPC storage solutions, and associated storage and system interconnects, continue to be critical for HPC infrastructure to deliver optimal capabilities and provide the fastest time to results for the systems' users. Data-intensive workloads driven by new AI/ML/DL workloads, increasing scale of traditional HPC modelling and simulation, emerging edge computing, and emerging composable systems are placing greater demands and requirements on HPC storage systems. Insights into the critical factors driving these and other trends are detailed in the 2022 iteration of Hyperion Research's annual MCS end users' study, 2022 HPC Multi-Client Study: Trends and Forecasts in HPC Storage and Interconnects. Key findings from the report are summarized in this document.

Expertise is now a major concern for both HPC and AI users, outranked only by budget concerns among top barriers to expanding HPC on-premises. A third of HPC sites (33.1%) reported that lack of knowledge or skilled support staff was one of their top three barriers. Only a third of respondents (35.4%) report that they do not have any staffing concerns within the next year. When asked about barriers to furthering AI capabilities, AI-specific expertise was identified as a significant concern. Other significant barriers also included access to AI expertise (48.9%), skills in AI model development (47.2%), and skills in AI programming (36.1%). This data is from an annual study that is part of the eighth edition of Hyperion Research's HPC end-user-based tracking of the HPC marketplace. It included 181 HPC end-user sites with 3,830 HPC systems.