UniMem - Redesigning Disaggregated Memory within a Unified Local-Remote Memory Hierarchy
Explores UniMem, a novel disaggregated memory system that optimizes remote memory access, local cache utilization, and page migration for improved performance in large-scale computing environments.
Optimize Your AI Cloud Infrastructure: A Hardware Perspective
Dive into hardware-level optimization techniques for AI cloud infrastructure, exploring GPU clusters, NUMA topology, and performance tuning strategies using H100 clusters and MLPerf benchmarks.
TAO: THP Allocator Optimizations
Explore THP Allocator Optimizations with ZHAO Yu, delving into advanced Linux memory management techniques and performance enhancements.
Leverage Topology Modeling and Topology-Aware Scheduling to Accelerate LLM Training
Explore topology modeling and scheduling techniques to optimize LLM training performance. Learn to address network bottlenecks and enhance resource allocation for AI workloads in Kubernetes environments.
Boosting Training and Inference Performance via Topology-Aware Scheduling of Heterogeneous Resources
Discover how ByteDance optimizes LLM workload performance through enhanced topology-aware scheduling, focusing on die-level affinity, RDMA optimization, and GPU-RDMA integration in Kubernetes environments.
Service Profiling Based Management and Scheduling in Kubernetes
Discover how to optimize Kubernetes resource management through service profiling, utilizing historical patterns for efficient workload scheduling and improved cluster performance.