GPU execution
Threads, blocks, and warps as the execution model behind CUDA performance behavior.
Static Research Blog
D.W.LEE
Exploring AI systems through GPU computing, memory architecture, and computer systems research.
This site collects short study notes, paper reviews, and implementation writeups. The focus is on how workloads map onto hardware, where performance goes, and which tradeoffs matter in practice.
Current scope: the PDF CV is available now. GitHub and email remain unpublished until verified public values are available.
Research Focus
Memory architecture
Bandwidth, locality, reuse, and data movement across the hierarchy that constrains AI workloads.
Systems evaluation
Paper reading and project notes that connect implementation choices to observable runtime cost.
Selected Study Notes
Study Note
GPU Execution Model: Threads, Blocks, and Warps
How CUDA execution maps onto warps, divergence, and memory access patterns.
Study Note
Memory Bandwidth and AI Workloads
Why bandwidth and data movement often dominate AI accelerator performance.
Selected Paper Reviews
Paper Review
Newton: Accelerator-in-Memory Architecture for Machine Learning
A short review of how near-memory computation changes the cost of machine learning systems.
Selected Projects
Project
CS336 Assignment 1: Implementation Notes and Takeaways
Implementation notes from a language-model assignment with attention to shape discipline and measurement.