GPU-Accelerated Numerical Optimization for Portfolio Construction
Overview
What this challenge is about.
Implement a primal-dual interior-point solver for box-constrained QP in CUDA. Use cuSPARSE for sparse matrix-vector multiply and cuSOLVER for the inner Newton system. Benchmark on 30 days of anonymized portfolio-construction problems (4,200 assets, 18,000 constraints). Compare against Mosek and OSQP on CPU. Report wall time, accuracy (gap to KKT), and stability across days. Deliver CUDA source, a 7-page benchmark + accuracy report, and a deployment recommendation (including cost analysis: GPU node vs CPU cluster).
The Brief
What you'll do, and what you'll demonstrate.
Cut daily portfolio-construction QP solve time from 38 minutes to under 5 minutes using GPU-accelerated interior-point methods without sacrificing KKT accuracy.
Earning criteria — what you'll demonstrate
- Implement a primal-dual interior-point method for constrained QP
- Use cuSPARSE and cuSOLVER for the dominant sparse-linear-algebra kernels
- Benchmark GPU vs CPU solvers fairly with identical accuracy targets
- Reason about the GPU vs CPU cost trade for daily production workloads
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career mappings coming soon.