# Hybridizer

> Hybridizer is a compiler toolchain that turns C#/.NET code into native GPU-accelerated code, especially CUDA kernels for NVIDIA GPUs, without requiring developers to rewrite their algorithms in CUDA C++.

Hybridizer targets .NET teams that need to accelerate numerically intensive workloads — Monte Carlo simulations, pricing engines, risk calculations, actuarial models, seismic processing, and industrial simulation — while keeping their existing C# codebase intact.

The core value: write once in idiomatic C#, compile to GPU-native code using standard build tooling. The generated CUDA source is inspectable, profilable with NVIDIA Nsight, and auditable for regulated environments.

## Key concepts

- **Single-source acceleration**: mark a C# method with `[EntryPoint]`, invoke it via `HybRunner`. No separate CUDA file, no separate algorithm.
- **Backends**: CUDA (primary, NVIDIA GPU), OMP (CPU multithread, used for validation), AVX/AVX2/AVX512 (CPU SIMD).
- **Generated artifacts**: Hybridizer produces `.cu` source files. Enterprise users can inspect, audit, and manually optimize these.
- **Constraints**: no heap allocation in device code, no reference types in kernels, use structs and flat arrays. Virtual dispatch and delegates are supported but expensive on GPU.

## Who uses it

- Quantitative finance teams (banks, hedge funds, asset managers) running pricing, risk, and Monte Carlo in .NET
- Insurance and actuarial teams in regulated environments that require auditable generated code
- Industrial simulation and geoscience teams (energy, seismic, CFD) with large C# numerical codebases
- Current enterprise customers include BNP Paribas and AON

## Editions

- **Free**: functional GPU acceleration, Visual Studio integration, community support. No time limit for evaluation and prototyping.
- **Enterprise**: commercial usage rights, generated CUDA source visibility, 24/7 support, code auditing, optimization consulting, cross-platform deployment guidance.

## Docs and resources

- Homepage: https://hybridizer.io
- Documentation: https://docs.hybridizer.io
- Blog (tutorials and benchmarks): https://hybridizer.io/blog/
- Download / free edition: https://hybridizer.io/download/
- Pricing: https://hybridizer.io/pricing/
- Quantitative finance page: https://hybridizer.io/quant/
- Energy and simulation page: https://hybridizer.io/energy/
- Hybridizer vs CUDA rewrite: https://hybridizer.io/blog/hybridizer-vs-a-cuda-rewrite-when-dotnet-teams-should-not-start-from-scratch/
- First 15 minutes guide: https://hybridizer.io/blog/first-15-minutes-with-hybridizer/
- How to evaluate GPU fit for a .NET loop: https://hybridizer.io/blog/how-to-know-if-your-dotnet-loop-belongs-on-the-gpu/
- Why generated CUDA visibility matters: https://hybridizer.io/blog/why-generated-cuda-visibility-matters-for-production-dotnet-teams/
- Hybridizer vs ILGPU comparison: https://hybridizer.io/hybridizer-vs-ilgpu/