July 22, 2026 -- Quside announces the availability of quside-curand, a C++/CUDA library that makes hardware-generated quantum entropy available to CUDA applications through a familiar cuRAND-style host-generator interface. The integration connects the Quside Garnet Plus A41 quantum random number generator (QRNG) with GPU-accelerated workflows used in high-performance computing, quantitative finance, simulation, and security-sensitive applications.

Developers can use quside-curand to generate uniform floats, uniform doubles, or raw 32-bit words from the Quside QRNG and pass them into existing CUDA workflows. The library follows the conventions of the cuRAND host API, including familiar creation, generation, destruction, status, and error-handling patterns, reducing the amount of application code that needs to change.

A complementary approach to GPU pseudorandomness

quside-curand is designed to complement, not replace, GPU pseudorandom number generators. Quantum entropy can be used for seeds, keys, independent starting states, and selected high-consequence variates, while cuRAND continues to generate the very large streams required by performance-intensive GPU workloads.

Randomness is a core dependency in Monte Carlo methods, distributed simulations, risk models, randomized algorithms, and cryptographic operations. GPU pseudorandom number generators are optimized for speed and reproducibility, but their output remains deterministic and depends on the quality, uniqueness, and management of their seeds. At scale, reused seeds, correlated streams, and limited visibility into the provenance of entropy can create operational and security risks.

“Randomness is a fundamental input for GPU-accelerated workloads, but not every application has the same requirements for speed, provenance, and unpredictability. quside-curand gives developers a practical way to combine locally generated, verifiable quantum entropy with the performance of CUDA, applying each source where it delivers the most value”  – Jose Ramón Martínez, VP Product.

The release includes:

• A cuRAND-style host API for requesting quantum-generated uniform floats, doubles, and raw 32-bit values.
• A GPU Monte Carlo example that estimates pi using Quside entropy and CUDA compute kernels.
• Transparent performance measurements that help developers understand the trade-off between physical entropy generation and arithmetic GPU generation.
• Validation utilities that can feed raw or conditioned QRNG output into commonly used test suites for engineering analysis.
• CMake integration, including a static library, public header, and package configuration for straightforward inclusion in C++/CUDA projects.

To ease integration, quside-curand is API-compatible in style with the curandGenerator_t implementation. Entropy is generated on the host by the Quside QRNG and then transferred to GPU memory, following the same host-to-device workflow used by applications that generate data in host memory today. Existing GPU kernels do not need to change.

A practical hybrid model allows organizations to apply quantum entropy selectively. Direct QRNG output can be used where provenance, independence, or unpredictability is a primary requirement. For high-volume workloads, the QRNG can provide high-quality seed material while GPU generators expand the bulk stream at native CUDA performance.

Both quside-curand and Quside Garnet Plus A41 hardware are available today. The initial release supports the Quside QRNG SDK, NVIDIA GPUs with CUDA Toolkit 12.x, and CMake 3.20 or later. Developers can request access, technical documentation, and evaluation support through Quside.