FICO adds Nvidia GPU power to Xpress for 50x speed-up

FICO has added GPU acceleration from Nvidia to the latest version of its Xpress Optimisation suite, which it says delivers speed improvements of up to 50 times for very large optimisation problems.

The Xpress 9.8 release introduces a GPU-accelerated implementation of FICO's hybrid gradient algorithm. The company said this targets extremely large and dense optimisation models that contain tens or hundreds of millions of non-zero elements in their matrices.

The implementation uses Nvidia's CUDA-X libraries. CUDA-X is a collection of GPU-accelerated libraries and frameworks that support data processing, artificial intelligence and high-performance computing on Nvidia hardware.

GPU-based solver

FICO said the GPU-accelerated hybrid gradient algorithm can deliver up to 50-fold speed gains in some scenarios. The algorithm uses lower memory overhead than some traditional techniques. This suits large-scale problems that would otherwise require significant CPU memory and time.

The company has made GPU acceleration for the hybrid gradient algorithm available as a beta feature in Xpress 9.8. The feature runs on any Nvidia GPU that supports CUDA Compute Capability 7.5 or later.

FICO positions Xpress as a suite with a broad range of optimisation algorithms. These cover linear, mixed-integer and nonlinear problems.

Bill Waid, Chief Product and Technology Officer at FICO, said the company views the latest release as part of a longer development path. "With more than 40 years of continuous innovation in the optimization solver industry, FICO Xpress continues to evolve by integrating cutting-edge technologies that address our customers' need for solving larger, more complex problems at unprecedented speed," said Bill Waid, chief product and technology officer at FICO. "The GPU-accelerated hybrid gradient algorithm represents an important addition to Xpress's comprehensive suite of optimization algorithms, enabling organizations to solve large-scale problems faster and unlock new possibilities for data-driven decision making."

Nvidia partnership

Nvidia has worked in recent years with several optimisation software vendors that are adapting solvers for GPUs. The company sees optimisation as a significant use case for its accelerated computing strategy.

"NVIDIA CUDA-X libraries and accelerated computing can help enterprises solve the world's largest, most complex optimization problems," said Alex Fender, Director of Decision Optimization at NVIDIA. "The ecosystem has quickly embraced GPU solvers, and by leveraging NVIDIA CUDA-X libraries, FICO Xpress 9.8 is helping customers achieve breakthroughs in speed and scale, transforming operational efficiency and decision making."

Performance improvements

Alongside the new GPU feature, FICO has made several CPU-based improvements in Xpress 9.8. The company said the mixed-integer programming engine now runs 14% faster overall than in the previous year's release.

On models that take at least 100 seconds to solve, FICO reports average speed improvements of 24% for the mixed-integer programming engine. These models include more complex integer and binary variables that often arise in scheduling, routing and resource allocation problems.

The firm also reports substantial gains in its global solver. It said the global solver is now 68% faster overall compared with the prior version. On models that require at least 100 seconds, FICO said the global solver runs 5.3 times faster.

FICO has framed these enhancements as part of a broad optimisation toolkit within Xpress. The suite includes Xpress Solver, which provides access to multiple algorithm types for different mathematical programming formulations.

Target users

Xpress targets operations research professionals and data scientists who build optimisation models for business applications. Typical uses include supply chain planning, logistics, production scheduling, pricing and workforce planning.

The software lets users formulate models in high-level terms that represent real-world trade-offs and sensitivities. It also supports what-if scenario analysis and comparison of different strategies within decision workflows.

FICO said Xpress 9.8 is available now with GPU acceleration for the hybrid gradient algorithm offered as a beta feature. The company is offering a complimentary 60-day trial licence of the suite for new users.

FICO indicated that it plans further development across both GPU and CPU paths as organisations increase the size and complexity of optimisation models in production systems.