Team Matador (Texas Tech) at the 2026 Winter Classic Invitational Student Cluster Competition: Joseph, Benjamin, Pawanjit and Naeem.
The Winter Classic is a series of weekly modules. Each one is run by an HPC center or vendor and each uses a different real machine: build and tune workflows on supercomputers, run real HPC workloads as fast as possible, and stay within a power cap.
Stack: C, Fortran, Python, Linux, Slurm, MPI, OpenMP, OpenACC, Apptainer, CUDA-Q, PennyLane
Final presentation: Final Presentation.pdf
| Week | Module | Host / system | Headline result |
|---|---|---|---|
| 00 | HPC crash course: parallelism, OpenMP/OpenACC, MPI | XSEDE/ACCESS bootcamp, PSC Bridges-2 | Exam, plus Laplace solvers in OpenMP, OpenACC, Fortran and MPI |
| 01 | HPL & HPCG benchmarking | HPE Cray EX (HPE Vlabs) | HPL 6.2 TFLOP/s (75.8% of peak) on 2 nodes; HPCG 4% → 17% on 4 AMD EPYC nodes |
| 02 | Quantum circuit simulation with CUDA-Q | PSC Bridges-2 (V100 / H100) | 31-qubit GHZ ~1,000× faster on GPU than CPU; tensornet-mps up to 1,760 qubits; splitting the Hamiltonian over 2 GPUs halved runtime |
| 03 | Troubleshooting: system testing of container images | ORNL OLCF Odo | Flagged 4 of 8 images (3 broken, 1 degraded) with HPL, LAMMPS and OSU |
| 04 | Quantum transfer learning | ORNL OLCF Odo + IQM Garnet | Swept 1–8 GPUs, qubits and shots; 5× faster training; ran on a real IQM quantum computer |
Each commit adds one topic, in the order we worked through the competition. To review what was learned step by step:
git log --reverse --onelineweek00-hpc-crash-course/ notes, exam, openmp/ openacc/ fortran/ mpi/ jobs/
week01-hpe-benchmarks/ hpl/ (HPL.dat, test.bat, tuning tools) hpcg/
week02-psc-cudaq/ cudaq_psc/ (GHZ backends, tensornet limits, observe_async, Q6)
week03-olcf-system-testing/ scripts/ (Apptainer jobs, automation, parser) results/
week04-olcf-quantum-transfer-learning/ scripts/ (sweeps, expanded dataset, qml.specs) results/ docs/
Some original scripts were lost after the competition. They have been rebuilt from the submitted reports and the final presentation, and every rebuilt file says so in its header. All numbers in the READMEs come from the submitted reports, not from new runs. Files still missing are listed in week 01.
Never commit cluster passwords or API tokens. keys.sh and similar files are git-ignored.