Minseung Jung

• 20-hour+ week student positions responding to the computing needs of the Georgia Tech Community • Supported client with software installations, imaging windows, virtual machines, school accounts, authentication systems, etc. • Analyzed and resolved workstation, application, and peripherals related problems.

• Developing advanced sampling techniques to optimize multi-GPU workload simulations, effectively reducing workload sizes while preserving key performance characteristics and GPU counter profiles. • Analyzing critical GPU counters and work features, ensuring that the sampling process accurately reflects both computations and communication errors between GPUs in a multi-GPU environment. • Minimizing communication and computation errors in multi-GPU simulations by designing efficient sampling strategies that account for inter-GPU data transfers and synchronization overheads, enhancing overall simulation accuracy and speed.

• Developed a robust framework for detecting data duplication, CPU, GPU, and FPGA telemetry from Data Centers within Azure Synapse environments. Leveraged the HyperLogLog algorithm to improve data quality and reliability for downstream applications. • Designed a real-time dashboard for monitoring data duplication, delivering clear, actionable insights and enabling timely interventions to maintain data integrity and track HW malfunctions, specifically CPU and GPU issues. • Optimized ML model training by ensuring clean data inputs through efficient data duplication detection, enhancing model accuracy and overall system performance.

• Conducted in-depth workload profiling & tracing on CPU architectures, focusing on multi-threaded applications with high performance memory systems, particularly in realm of Graph Analytics applications. • Conducted performance analyzing across diverse system configurations, utilizing advanced techniques in memory and instruction tracing analysis (tracing for multi-threaded application), leading to the identification of optimal system configuration for enhanced application throughput. • Contributed invaluable insights into performance enhancement strategies, drawing upon expertise in Graph algorithm, x86 ISA, and CPU vectorization methods, cementing a reputation for driving impactful improvements. • Developed bespoke CPU profiling tools and refined performance analysis workflows, showcasing proficiency in C++, coupled with an understanding of CPU parallelization software and multicore architectures, and ran a variety of simulator to evaluate performance.

• Supervisor: Professor Hyesoon Kim • Explored the feasibility of generating representative proxy benchmarks for proprietary/sensitive programs • Learned how to profile Machine Learning programs and read hardware performance counters • Investigated the privacy aware tracing mechanism to prevent side-channel attacks • Developed benchmarks for AI workloads that leverage PyTorch ET capabilities to record model information, hide model sensitive information (algorithms), and reproduce original performance • Used NVIDIA Nsight Compute to evaluate performance metrics and extend with analysis scripts for post-processing results

• Aided 6-hour-long weekly Lab Sessions by guiding students in class and graded lab reports and exams • Coached students to build various components and testing the circuits • Worked alongside 10+ other TAs (including Graduate TAs) and assisted 520+ students and clarified their questions