NVIDIA-AI-Blueprints/portfolio-optimization

Powered by NVIDIA cuOpt: a GPU-accelerated portfolio optimization toolkit for building, backtesting, and scaling Mean-CVaR and Mean-Variance investment workflows with CUDA-X Data Science.

What it solves

This project addresses the conflict between computational speed and model complexity in financial portfolio optimization. It transforms slow, batch-processed robust analysis—such as Mean-CVaR (Conditional Value-at-Risk) and large-scale simulations—into a fast, iterative workflow for dynamic decision-making.

How it works

The solution uses a GPU-accelerated pipeline built on the NVIDIA CUDA ecosystem:

  1. Scenario Generation: Uses RAPIDS cuML for GPU-accelerated Kernel Density Estimation (KDE) to learn and sample return distributions, achieving up to 100x speedups.
  2. Optimization: Employs NVIDIA cuOpt solvers to handle complex, scenario-based Mean-CVaR portfolio optimization problems, delivering up to 160x speedups over CPU-based solvers.
  3. Backtesting: Utilizes CUDA-X Data Science and the HPC SDK to rigorously test and refine trading strategies.

Who it’s for

Financial analysts, quantitative researchers, and developers building high-performance trading strategies who need to perform large-scale simulations and risk modeling at production speeds.

Highlights

  • Massive Speedups: Up to 160x faster optimization and 100x faster scenario generation.
  • Risk Modeling: Enables the use of Conditional Value-at-Risk (CVaR) at production speeds.
  • Iterative Workflow: Supports dynamic, data-driven optimization cycles rather than slow batch processing.
  • Interactive Tools: Includes Jupyter notebooks for walkthroughs and a Streamlit app for dynamic rebalancing demonstrations.

Related

  • Project
    NVIDIA/cuopt-examples

    A collection of examples and notebooks demonstrating how to use NVIDIA cuOpt, a GPU-accelerated engine for solving complex decision optimization problems like vehicle routing and linear programming.

  • Project
    NVIDIA/cuopt

    A GPU-accelerated optimization engine for solving large-scale linear, quadratic, and vehicle routing problems in near real-time.

  • Project
    NVIDIA-NeMo/Curator

    A GPU-accelerated data curation framework for building repeatable pipelines to load, filter, deduplicate, and transform massive text, image, video, and audio datasets for AI training.

  • Project
    KomputeProject/kompute

    A cross-vendor GPU compute framework based on Vulkan that enables high-performance acceleration for machine learning and data processing across AMD, NVIDIA, and Qualcomm hardware.

  • Project
    NVIDIA/cuml

    A GPU-accelerated machine learning library that provides scikit-learn-style APIs and a transparent acceleration module to speed up ML workloads on NVIDIA GPUs.