Memory Wall Mitigation and Acceleration of AI Workloads Using CXL Near Memory Computing
Open Compute Project via YouTube
Overview
Learn about CXL-based Near Memory Compute architectures in this technical talk that explores solutions for hyperscale workload acceleration. Discover how Marvell's integrated solutions address memory capacity and bandwidth challenges for recommendation systems, big data analytics, in-memory vector databases, and KV caches. Explore the implementation of NMC accelerators featuring high-performance compute cores with SIMD vector processing and efficient engines for data movement, compression, and encryption. Understand how these technologies work together to mitigate memory wall issues and enhance AI accelerator performance in heterogeneous computing environments.
Syllabus
Memory wall mitigation and acceleration of AI workloads, and in memory databases using CXL Ne
Taught by
Open Compute Project