This research introduces CENT, a CXL-enabled system that eliminates GPU dependency for LLM inference while optimizing memory resources and performance.

• Leverages Processing-In-Memory (PIM) technology to address the memory bottlenecks in LLM inference
• Utilizes CXL interconnect to create a flexible, scalable system architecture
• Tackles the challenges of large context windows (up to 1 million tokens) through innovative memory management
• Provides a cost-effective alternative to traditional GPU-based inference systems

This engineering breakthrough matters because it offers a path to more efficient, accessible LLM deployment at scale without the power and cost constraints of GPU-dependent architectures.

PIM Is All You Need: A CXL-Enabled GPU-Free System for Large Language Model Inference