PHD STUDENT · TENNESSEE TECH
·
BAY AREA, CA
Amr Akmal Abouelmagd
PhD Student, Computer Science / HPC & AI Systems Researcher / Software Engineer
I work on the systems software behind large-scale HPC and AI: fault tolerance, KV-cache architecture and performance, and MPI and collective communication for AI that keep making progress, even under failures. I'm a PhD student at Tennessee Tech, advised by Prof. Anthony Skjellum. Previously, I researched GPU performance on the AMD MI300A (El Capitan's APU) and NVIDIA H100 at Lawrence Livermore National Laboratory, and spent 2.5 years building production distributed systems at Incorta.
Open to Summer 2027 internships in HPC & AI systems
§ 01
News
// recentSep 2026
First-author paper “Deriving Vendor-Agnostic GPU Metrics from AMD MI300A and NVIDIA H100 Hardware Counters” accepted at the ProTools workshop at SC'26.
Sep 2026
OpenCCL poster, proposing an open standard that unifies collective communication libraries (*CCLs), accepted at SC'26.
Aug 2026
Completed my M.S. at Tennessee Tech (4.0 GPA) and my research internship at LLNL; started my PhD as a Graduate Research Assistant with Prof. Anthony Skjellum.
Dec 2025
Paper “GPU Partitioning, Power, and Performance of the AMD MI300A” accepted at SCA / HPC Asia 2026.
Nov 2025
Top 5 Finalist in the SC'25 Graduate Student Research Competition for MI300A APU performance analysis.
2024
Received the Outstanding Student Paper Award at IEEE HPEC 2024.
§ 02
Research interests
// focus areas01Fault Tolerance in HPC Systems
02KV-Cache Architecture & Performance
03MPI & Collective Communication for AI — Fault-Tolerant, Strong Progress
04Distributed Systems & Parallel Computing
05Systems Performance Analysis
06GPU Performance — Partitioning, Power & Cross-Vendor Hardware Counters
§ 03
Experience
// research + industryTennessee Technological University
Cookeville, TN
Aug 2026 — Present
Graduate Research Assistant · Advisor: Prof. Anthony Skjellum
- Research on fault tolerance in HPC systems, KV-cache architecture and performance, and MPI for AI: communication that survives failures and makes strong progress.
- Contributing to OpenCCL, a proposed open standard and reference implementation that unifies collective communication libraries (*CCLs) for performance, portability, and productivity; co-author of the SC'26 poster.
Lawrence Livermore National Lab
Computation Directorate · Livermore, CA
Jan 2026 — Aug 2026
Graduate Research Intern
- Derived vendor-agnostic GPU performance metrics from AMD MI300A and NVIDIA H100 hardware counters, enabling like-for-like comparison across vendors; first-author paper accepted at the ProTools workshop at SC'26.
- Collected and analyzed hardware counters using Caliper, Thicket, rocprofiler-sdk, and Nsight Compute, with Benchpark and Spack for reproducible experiments.
Lawrence Livermore National Lab
Computation Directorate · Livermore, CA
May 2025 — Aug 2025
Graduate Research Intern
- Designed and ran benchmarking experiments on the AMD MI300A APU across SPX, TPX, and CPX partitioning modes using RAJA Performance Suite kernels, surfacing up to 30% execution-time variance between configurations.
- Conducted systems-level analysis of GPU runtime scheduling, dynamic power sharing, and heterogeneous memory behavior.
- First-author paper accepted at SCA/HPC Asia 2026, targeting optimization of LLNL's El Capitan exascale supercomputer; Top 5 Finalist in the SC'25 Graduate Student Research Competition.
Incorta
Alexandria, Egypt
Aug 2021 — Feb 2024
Software Engineer II · R&D Software Engineer · Graduate Student Intern
- Delivered a 2× indexing speedup and 7× query-latency reduction on datasets exceeding 1 billion records in a distributed analytics platform built on Kubernetes and ZooKeeper, improving SLA compliance.
- Raised average CPU utilization from 40% to 85% by resolving execution bottlenecks, reducing cluster over-provisioning.
- Designed an internal caching layer that cut cloud object-storage I/O and infrastructure cost with no hardware changes.
- Improved monitoring and telemetry, reducing incident MTTR across microservices.
JavaPythonGoKubernetesZooKeeperMySQLDistributed Systems
§ 04
Publications
// 9 publications · 5 first-author01
Deriving Vendor-Agnostic GPU Metrics from AMD MI300A and NVIDIA H100 Hardware Counters
02
OpenCCL: A Proposed Open Standard and Reference Implementation Unifying *CCLs for Performance, Portability, and Productivity
03
GPU Partitioning, Power, and Performance of the AMD MI300A
04
Using Hardware Metrics to Understand Performance of RAJA Suite Kernels in Different GPU Modes on MI300A
05
Load Imbalance in HPC Applications: Improved Profiling and New Ways to Use Wasted Cycles
06
A Survey of Optimization Approaches for MPI Alltoall and MPI Alltoallv
07
Leveraging the Power of AI and Social Interactions to Restore Trust in Public Polls
08
Emerging Paradigms for Securing Federated Learning Systems
09
Cycle-Stealing in Load-Imbalanced HPC Applications
§ 05
Education
// PhD in progressTennessee Technological University
Cookeville, TN
Aug 2026 — Present
Ph.D. in Computer Science
- Advised by Prof. Anthony Skjellum. Research on HPC and AI systems: fault tolerance, KV-cache architecture and performance, and MPI for AI. Expected completion August 2030.
Tennessee Technological University
Cookeville, TN
Jan 2024 — Aug 2026
M.S. in Computer Science
- GPA 4.0 / 4.0, completed August 2026. Research in high-performance computing, GPU performance analysis, and distributed systems.
Alexandria University
Faculty of Engineering · Alexandria, Egypt
Sep 2016 — Jun 2021
B.S. in Computer Engineering
§ 06
Selected projects
// systems / parallelDistributed Key-Value Store
Leader-coordinated store with auto-partitioning and live rebalancing. Fault tolerance via the Bully election algorithm for zero-data-loss failover; multithreaded reads with mutexes and atomics.
C++MultithreadingDistributed
BusTub Database Engine
Buffer-pool manager with page-replacement policies, plus a thread-safe LRU-K eviction algorithm that maximizes memory utilization under concurrent workloads.
C++MakefileCMU
CUDA Parallel Softmax
Optimized softmax kernel using shared memory and block-level reduction to cut global-memory latency; throughput scaled via memory coalescing and parallel row-wise processing.
C++CUDAGPU
§ 07
Skills & tooling
Languages
Parallel & GPU programming
Performance analysis
Benchmarking & reproducibility
Distributed systems & infra
Developer tools
§ 08
Awards & achievements
Top 5 · Graduate Student Research CompetitionACM/IEEE SuperComputing (SC'25)
2025
Outstanding Student Paper AwardIEEE HPEC 2024
2024
10th Place · AlexCPC — qualified for ECPCEgyptian Collegiate Programming Contest
2018
Cloud DevOps NanodegreeUdacity
§ 09
Get in touch
I'm open to Summer 2027 internships in HPC and AI systems, and always happy to talk about research and collaborations. Email is the fastest way to reach me.