ilgpu.net

What It Is

ILGPU is a modern GPU JIT compiler for .NET applications. Its goal is to let developers write high-performance GPU kernels in C# or F# without having to work directly with C++, CUDA, or OpenCL. The project is written entirely in C#, and the materials emphasize that the current version has no native dependencies, which makes it more portable for .NET applications.

Core Capabilities

Its core value is bringing GPU kernel compilation, distribution, and execution into the .NET ecosystem. ILGPU supports concepts such as Context, Accelerator, MemoryBuffers, ArrayViews, Kernels, Shared Memory, and Math Functions. Functions within kernel scope do not require extra annotations and can operate on value types. The CPU Accelerator is a major highlight: it can execute kernels in single-threaded or multi-threaded mode, making it useful for debugging, testing, or simulating a target platform. Shared memory, atomic operations, warp shuffles, low-level intrinsics, high-performance math functions, multidimensional indexing, and implicitly grouped kernels are also listed among its features, indicating that it is not merely a high-level wrapper.

Support and Ecosystem

In terms of languages and platforms, the materials explicitly mention C# and F#, with support for .NET 4.7 and .NET Standard 2.1, such as .NET Core 3.1/.NET 5.0. Ecosystem entry points include Documentation, GitHub, and Discord. Project news also mentions regular developer discussions and GitHub milestone tracking. The documentation covers GPU fundamentals, getting started guides, memory, kernels, shared memory, debugging, profiling, and upgrade guides, with a fairly complete structure; however, some example content still appears to be planned or in progress.

Pricing and Open Source

ILGPU is free and open source under the University of Illinois/NCSA Open Source License. The project is primarily supported by G-Research and accepts contributions or small donations. For individual research, open-source projects, and cost-sensitive teams, it offers very strong value.

Pros, Cons, and Who It’s For

Its strengths are clear: it is .NET-friendly, supports CPU-based debugging, has a well-defined cross-platform approach, and exposes a broad range of low-level GPU capabilities. Its limitations are also explicit: exception control flow is not supported, reference types are currently unsupported, and lambda/delegate support is still planned for the future. GPU hardware debugging and profiling only have basic support, with the official recommendation leaning toward CPU debugging. It is well suited to teams with .NET experience that need high-performance parallel computing or algorithm prototyping. It is not a good fit for directly migrating business code that strongly depends on full .NET runtime features.

Access from China

The collected materials do not provide information about access from mainland China, mirrors, payment, or network availability, so this is unknown. Possible alternatives include CUDA, OpenCL, and C++ AMP. Teams operating in China’s network environment should verify access to GitHub, the documentation site, and Discord in advance.