Skip to content
← ProjectsIn progress

GPU Workload Profiling & Optimization

Full-stack profiling and optimization of GPU workloads — my final-year project, turning measurements into concrete speedups.

Final-Year Project · 2026–2027
  • systems
  • performance

Problem

GPU workloads are easy to run and hard to make fast — occupancy, memory bandwidth, and kernel latency all pull against each other, and intuition is a poor guide. This final-year project profiles GPU workloads end to end and lets the measurements, not guesswork, drive each optimization. It’s the centerpiece of my systems work, and this page grows into a full case study as results land.

Results — before / after

Before / after
MetricBeforeAfter
Kernel latency (p99)
Memory throughput
SM occupancy