Dr. Wenzhe Guo
Senior R&D Engineer
imec
Dr. Wenzhe Guo is a Senior R&D Engineer in the Compute System Architecture department at imec, specializing in performance modeling and system-level design for large-scale AI workloads. He received his M.S. and Ph.D. in Electrical and Computer Engineering from King Abdullah University of Science and Technology (KAUST) in 2023, where his research focused on energy-efficient AI computing inspired by brain-like principles. At imec, he leads the development of imec.kelis, an analytical virtual-datacenter modeling tool that identifies compute, memory, and network bottlenecks for LLM training and inference at scale. His work spans accelerator modeling, distributed parallelism, and cross-stack hardware–software co-optimization.
Presentations
Introducing imec.kelis: A Datacenter Performance Model for AI Densification
AI-driven datacentre densification depends on understanding where the next bottleneck will arise—compute, memory, or network. As large‑scale large language models (LLMs) continue to proliferate, their enormous compute and bandwidth demands make system‑level design increasingly complex. Achieving high efficiency requires coordinated, cross‑stack co‑optimization spanning hardware technology, interconnects, model architecture, and distributed parallelism. Yet exploring this space experimentally is prohibitively time‑ and energy‑intensive. Imec.kelis is a virtual datacenter modeling tool that converts full LLM workloads into fast, explainable predictions across GPU architectures, node configurations, and cluster scales. Within minutes, it reveals performance limits, communication hotspots, and scaling behavior, enabling architects to test “what-if” scenarios before deployment. Kelis supports both datacenter-level and processor-level exploration, helping teams identify bottlenecks and maximize throughput within fixed power, thermal, and space constraints.