Cost-Efficient LLM Training with Lifetime-Aware Tensor Offloading via GPUDirect Storage
- Ziqi Yuan
- Haoyang Zhang
- et al.
- 2025
- NeurIPS 2025
Dr. Seetharami R. Seelam is a Distinguished Engineer at IBM and a leader in large-scale systems architecture spanning hybrid cloud, high-performance computing (HPC), AI, and quantum computing. He currently serves as Chief Architect of Classical System Co-Design for Quantum-Centric Supercomputing (QCSC), where he applies deep systems expertise to accelerate and scale classical components of quantum algorithms. His work focuses on demonstrating quantum advantage by advancing classical subroutines of QCSC algorithms, developing integrated GPU-based solutions for outer decoders, and defining reference system architectures that unify quantum, HPC, and AI with an eye toward the scaling requirements of Starling and Blue Jay systems. He also leads collaborative efforts with partners and national laboratories on next-generation HPC and quantum architectures.
Previously, Dr. Seelam led a worldwide team at IBM Research to define strategy and execution plans for workloads including HPC, and AI IBM’s Hybrid Cloud, driving innovation roadmaps spanning multiple generations of compute, networking, storage, accelerators, and resource scheduling for large-scale, geographically distributed cloud systems.
Dr. Seelam has also made significant contributions to academia. He previously taught at New York University and currently teaches at Columbia University, bridging research, industry, and education in systems, cloud, and quantum computing.
An accomplished inventor and researcher, Dr. Seelam has filed more than 30 patents (15 issued) and published over 35 papers, receiving two Best Paper Awards and an Outstanding Paper Award. He is a sought-after speaker and keynote presenter at major academic and industry venues, including Supercomputing, GPU Technology Conference, IBM Think, and most recently FAST and ACM Middleware, speaking on topics such as cloud and AI infrastructure, DevOps, microservices, Kubernetes, and container technologies.