Agentic AI speeds up design of diverse computing systems efficiently
Agentic AI for Heterogeneous System Design Across Chip, Scale-Up, and Scale-Out Domains
Hardware ArchitectureDistributed, Parallel, and Cluster Computing
Summary
Designing modern computing systems is complicated because there are many types of chips, memory, and network connections to choose from, and no single setup works best for all tasks. The authors created ArchAgent, a smart system that breaks down the design challenge into smaller parts and uses specialized helpers to find the best combination. It checks its ideas with accurate models rather than guessing. ArchAgent finds better designs faster than older methods, saving time and energy while improving performance.
What this means in practice
- •For data center architects: Optimize hardware configurations across diverse compute chips and interconnects to improve AI and HPC workload performance under power constraints.
- •For chip system engineers: Automatically generate heterogeneous system designs that balance chip, scale-up, and scale-out considerations more efficiently than manual or hill climbing methods.
Authors
Murali Emani, Kashyap R Manjusha
Abstract
The design space of modern AI/HPC systems has expanded rapidly across compute architectures, including GPUs, dataflow processors, and RISC-V accelerators; memory tech- nologies, including HBM, GDDR, and SRAM; and interconnects such as NVLink, UALink, Infinity Fabric, InfiniBand, Slingshot, and Ultra Ethernet. No single architecture is optimal across LLM training, LLM serving, and AI-coupled HPC workloads, while the resulting combinatorial design space is too large for manual exploration. We present ArchAgent, a hierarchical multi- agent framework in which an LLM-based planner decomposes system design into chip, scale-up, and scale-out subproblems, each handled by specialist agents. A deterministic analytical evaluator closes the loop, ensuring that all performance estimates are derived from analytical models rather than generated by the LLM. On a mixed AI+HPC workload under a 1 MW power budget, ArchAgent discovers heterogeneous designs that achieve a 1.93x weighted geometric mean speedup over a homogeneous GPU baseline, while requiring only 18 evaluations compared with 100 evaluations for hill climbing, which achieves 1.65x speedup.