The 27th LSF Community Summit took place on the 1^st of June at the Santa Clara, Hilton.   We took the 120+ attendees on a journey from foundational concepts through to the future of AI-driven operations.   From understanding the platform, operationalizing it effectively, and exploring how it is evolving to meet the demands of modern HPC and AI environments.

The morning sessions focused on building that foundation with experts from IBM.

• Michael Spriggs, STSM, opened with a deep dive into LSF core concepts, providing a detailed look at cluster architecture, job lifecycle, and scheduling behavior. He moved beyond theory to explain how policies, resources, and prioritization interact in real deployments. A key theme was extensibility—capabilities such as esub, elim, and execution hooks demonstrated how LSF can be adapted to highly specific workload requirements. The session reinforced the idea that LSF is not just a scheduler, but a flexible platform designed to support complex and heterogeneous environments.
• John Welch, HPC Specialist, followed by shifting attention to lifecycle management, a critical but often underappreciated aspect of running LSF in production. He outlined the three-tier distribution strategy—Base Installations, Service Packs, and Fix Patches—and described how each plays a role in maintaining stability while enabling innovation. His talk emphasized the importance of structured maintenance practices and aligning updates with operational risk. The message was clear: managing LSF effectively requires treating it as enterprise infrastructure, not just software.
• Demin Zhang, Technical Account Manager, then brought an operational lens to the discussion with a session on running a healthy cluster. His focus on proactive monitoring, regular maintenance, and continuous optimization resonated strongly with day-to-day HPC operations. Rather than reacting to issues, he advocated for identifying and addressing them early, supported by tools such as the LSF Performance Analysis Tool. The session highlighted that consistent performance and reliability are the result of disciplined operational practices.
• George Gao, Senior Architect, then looked ahead to the tools and techniques needed to manage increasingly complex environments. He covered essential administrative capabilities, including queue analysis, resource monitoring, and fairshare tuning, before introducing AI-driven innovations such as LSF Simulator and LSF Predictor. These tools leverage historical data to improve scheduling accuracy and capacity planning, signaling a shift toward predictive and automated operations. His discussion of the emerging Agentic Framework underscored LSF’s trajectory toward more autonomous cluster management.
• Marek Sadowski, Alejandro Palumbo and Zack Williams then closed the morning session with a preview of how Instana and Watsonx.orchestrate can be used to manage and automate an LSF environment.

Taken together, the morning sessions painted a picture of a platform built on strong fundamentals but evolving steadily toward greater intelligence and automation—an essential combination in today’s mixed HPC and AI landscape.

The afternoon sessions extended this narrative, combining strategic vision, product updates, and customer experiences to show how LSF is being applied at scale.

• Sripriya Srinivasan, General Manager, IBM Software opened with a keynote that positioned IBM Spectrum LSF as an intelligent control plane for modern compute orchestration. Reflecting on its 30-year evolution, she highlighted how LSF has transitioned from a traditional batch scheduler into a central component of AI-era infrastructure. Her examples, spanning industries such as semiconductor and life sciences as well as IBM’s own internal workloads, reinforced LSF’s role in managing large-scale, business-critical environments. The forward-looking vision focused on AI-driven optimization and increasingly autonomous operations.
• Bill McMillan, Principal Product Manager, followed with a preview of LSF 10 Service Pack 16, outlining a set of enhancements focused on improving efficiency, stability, and usability. Updates such as enhanced memory reporting, waste analysis, and improved runtime management for critical workloads address practical challenges faced by operators. Additional improvements in GPU support, job filtering, and the Resource Connector reflect the ongoing evolution of LSF to support modern, heterogeneous infrastructures. The release represents steady, targeted progress in areas that directly impact operational effectiveness.
• The customer sessions brought this evolution to life. Bill Steinmetz, Distinguished Engineer at NVIDIA set the tone with a discussion of scaling EDA workloads across multiple data centers using LSF and MultiCluster. Their architecture demonstrated how advanced scheduling, containerization, and cluster affinity can be combined to improve utilization and resilience at global scale.
• Dan Coops, Verification Manager, IBM followed with their “Cloudburst” initiative, showcasing how hybrid HPC can be used to extend on-premises capacity into the cloud. By leveraging IBM Cloud HPC, they were able to dynamically scale resources while balancing cost and performance, providing a clear example of how organizations can modernize infrastructure while maintaining continuity.
• Gabor Samu, Senior Product Manager, he discussed the integration of quantum computing with HPC using LSF, highlighting the benefits of hybrid quantum-classical workflows in fields like semiconductor development and materials discovery, and showcased IBM’s efforts in developing open-source tools and plugins for managing quantum resources and orchestrating hybrid workflows.
• Microsoft 2025 Journey to LSF was presented by IBM Expert Labs, highlighted the impact of adopting LSF at hyperscale. Within a relatively short timeframe, they achieved large-scale production deployment, significantly increasing throughput and reducing downtime. Even small per-job efficiency gains translated into substantial overall improvements, illustrating the value of optimization at scale.
• Jeff Lau and Robert Ikeoka from Synopsys concluded the customer presentations with a transformation story focused on improving utilization and efficiency. By moving to a hub-and-spoke spot/shared model, they consolidated fragmented resources and dramatically improved performance and turnaround times. This demonstrated how architectural changes, combined with LSF capabilities, can unlock significant operational gains.
• The summit concluded with a forward-looking discussion from McMillan and Spriggs, focusing on future directions for LSF. Topics included enhancements to the information model, improved multi-cluster services, tighter integration with AI workloads, and the increasing role of AI in managing LSF itself. The overarching theme was simplification and automation, with a clear trajectory toward more intelligent, self-managing systems.

Overall, the summit reinforced a consistent message: LSF continues to evolve from a traditional workload scheduler into a central orchestration platform for hybrid, large-scale, and AI-driven compute environments. With strong customer validation and a roadmap focused on automation and intelligence, it is well positioned to support the next generation of HPC and AI workloads.