Sunil Kumar

Ph.D. | HPC & Systems Researcher

Indraprastha Institute of Information Technology, Delhi (IIIT-Delhi)  ·  Google Ph.D. Fellow

I am a researcher working on computer systems, with a focus on high-performance computing, power management, and performance optimization of modern CPU/GPU systems.

My research focuses on improving power efficiency (performance/watt) by exploiting techniques such as dynamic voltage and frequency scaling (DVFS), uncore frequency scaling, power capping, and runtime-level power management. I characterize diverse workloads using hardware performance counters and design adaptive runtime systems that respond to application behavior.

My work spans co-running HPC workloads, multicore and multiprocessor systems, CPU/GPU performance analysis, and resource management under power constraints. I am particularly interested in the interaction between workload characteristics, processor frequency settings, power budgets, throughput, and fairness.

Research interests

High-Performance Computing · CPU/GPU Performance Optimization · Power Management · DVFS · Uncore Frequency Scaling · Power Capping · Runtime Systems · PMU/PMC Profiling · Performance/Watt

News

2026 Full paper accepted at HiPC'26.

2026 Submitted Ph.D. thesis in July 2026.

2026 Full paper published in ACM TACO.

2025 Full paper accepted at Euro-Par'25.

2025 Short paper accepted in the Euro-Par'25 PhD Symposium.

2025 Artifact accepted at Euro-Par'25.

2025 Received student scholarship for Euro-Par'25.

2024 Poster accepted at HiPC SRS'24.

2022 Received the Google Ph.D. Fellowship.

2021 Cuttlefish accepted at SC'21.

Selected publications

Reward-Driven Power Scheduling under Power Capping with DVFS-UFS Co-Optimization

Sunil Kumar and Vivek Kumar · HiPC'26 · 2026

Details & publications →

Power Scheduling for Maximizing Throughput and Fairness in Co-running Applications

Sunil Kumar, Vivek Kumar, and Sridutt Bhalachandra · ACM TACO · 2026

Paper →

KarmaPM: Reward-Driven Power Manager

Sunil Kumar and Vivek Kumar · Euro-Par'25 · 2025

Paper →