BREAKING: The first native GPU-accelerated computation addon for Node.js - A new era is dawning - 2025.09.15
Hello World! My name is Zsolt Tövis and I officially announce that on September 15, 2025, I successfully ran computational tasks on the GPU in a Node.js environment using a native addon. This is a milestone in the JavaScript world, as developers can now directly access the power of the GPU without relying on Python or other external libraries.
This repo is not a project, but an official statement to document this historic event and lay the foundation for future development. I will regularly update information about the development process, share challenges, and milestones.
I have solved the creation of a native GPU addon that enables computational tasks to run on the GPU in Node.js. GPU-accelerated computations are already working, and the addon can handle multiple platforms and different GPU backends. This is no longer just a plan, but reality.
- Native addon for Node.js that allows direct GPU access
- Multi-platform support (Windows, Linux, macOS)
- GPU backend support (Vulkan, DirectX, Metal, OpenGL, WebGL)
- GPU-accelerated computational tasks
- Asynchronous operations with async/await support
This development is a revolutionary step in the Node.js world, as it allows developers to harness the power of the GPU without relying on Python or other external libraries. It unifies frontend and backend development in a single language, simplifying development processes and reducing complexity.
Instead of launching a project that simply provides GPU access, I will showcase various projects that leverage this technology. These projects span different fields, including machine learning, image processing, and other compute-intensive tasks.
- The Node.js world will never be the same
- Thousands of projects are waiting to leverage this paradigm shift
- Follow the updates, because the future is now
The initial proof of concept has been successfully implemented, demonstrating the feasibility of GPU-accelerated
computations in Node.js. This includes basic tensor operations, neural network layers, and activation functions running
on the GPU. I publish proof of concept code in the proof-of-concept folder of this repository. In
addition, I will continuously upload comparisons and benchmarks for each completed function.
If you want to support the development of this project, you can:
- Star the repository to show your interest
- Share your ideas and feedback
- Spread the word about this project
- Donate to support further development
Are you representing a company or large enterprise? You are welcome to invest in the future of native GPU computing for Node.js. Corporate sponsors and organizations can make significant donations or contact me directly to discuss business opportunities, partnerships, or custom enterprise solutions. For all business inquiries, please reach out via email or GitHub.
- GitHub Sponsors
- PayPal
- OpenCollective: This repo needs 100 stars to enable this option.
Another milestone: I have successfully implemented Autograd, Dense layers, and Optimizer for neural networks, enabling backpropagation and learning on the GPU in Node.js. I have also implemented the complete neural network training process, including forward and backward passes and optimizer steps.
- Autograd: automatic differentiation for neural networks
- Dense layer: fully connected layer for neural networks
- Optimizer: optimization algorithms for neural networks
- Complete neural network training pipeline: learning and inference on GPU
- Layers functions:
createDenseLayer,denseForward,denseForwardAutograd,denseParameters,denseParamCount - Autograd activations:
reluAutograd,sigmoidAutograd,tanhAutograd,reluBackward,sigmoidBackward,tanhBackward - Optimizers:
createSGDOptimizer,createAdamOptimizer,sgdStep,adamStep,sgdZeroGrad,adamZeroGrad
Another milestone: I have successfully implemented neural network activation functions, tensor manipulations, and loss functions. This proves that native GPU acceleration is not just theory, but practical application.
- Neural network activations:
elu,hardSigmoid,leakyRelu,relu,selu,sigmoid,softmax,softplus,softsign,tanh - Tensor manipulations:
cast,concat,expandDims,gather,oneHot,reshape,slice,squeeze,stack,tile - Aggregation functions:
argmax,argmin,cumsum,max,mean,min,norm,prod,reduceMax,reduceMean,reduceMin,reduceProd,reduceSum,sum - Comparison functions:
equal,greater,greaterEqual,less,lessEqual,notEqual - Loss functions:
binaryCrossEntropy,binaryCrossEntropyAutograd,huberLoss,meanSquaredError,meanSquaredErrorAutograd,softmaxCrossEntropy
I have successfully run computational tasks on the GPU in Node.js using a native addon. This is no longer just a plan, these are facts.
I decided that the first library to leverage native GPU acceleration in Node.js will be a TensorFlow.js-compatible API. This will allow developers to easily switch to Node.js with native GPU acceleration without significant changes to their code.
Tensor API development has started. What is already done and works on GPU in Node.js:
- Creations:
create,eye,fill,fromArray,linspace,ones,random,randomNormal,randomUniform,range,zeros - Core operations:
add,divide,minimum,maximum,modulo,multiply,power,subtract - Math functions:
abs,acos,asin,atan,atan2,ceil,cos,exp,floor,log,log1p,neg,round,sin,sqrt,tan - Logical operations:
logicalAnd,logicalNot,logicalOr - Linear algebra:
det,inverse,matmul,qr,svd,trace,transpose - Utilities:
dispose,getData,where
⭐ Star this repo to witness the birth of native GPU computing in Node.js!