What is ONNX GO?
ONNX GO, developed by Graiphic, is an innovative software solution designed for dynamic orchestration and real-time execution of numerical and AI computational graphs based on ONNX (Open Neural Network Exchange). ONNX is a universal file format designed to represent machine learning models. ONNX GO extends this capability, providing an intuitive, flexible, and powerful environment for creating, modifying, and executing computational graphs dynamically at runtime.
Why is ONNX GO Revolutionary?
Unlike traditional static graph approaches where models and computational workflows are fixed after deployment, ONNX GO introduces dynamic orchestration that enables real-time adjustments and optimizations. This breakthrough allows models and algorithms to be reconfigured instantly according to changing conditions, significantly improving operational adaptability.
The orchestration process occurs in two main stages:
-
Graph Design (ONNX): Computational workflows are built visually using LabVIEW, a graphical programming language. These workflows are then exported as standard ONNX graphs.
-
Runtime Execution (ONNX Runtime): At runtime, the ONNX graphs are dynamically loaded and executed with ONNX Runtime, optimizing execution across various hardware platforms including CPU, GPU, and FPGA.
Which Problems Does ONNX GO Solve?
ONNX GO addresses several critical industry challenges:
-
Complexity of AI Deployment: It simplifies deployment by enabling users without traditional programming expertise to create and manage advanced computational workflows through visual programming.
-
Inflexibility of Traditional Models: It overcomes the limitations of static pipelines by allowing real-time modifications and dynamic reconfiguration of models.
-
Performance and Energy Efficiency: By intelligently allocating hardware resources, ONNX GO reduces compute time and power consumption compared to conventional methods.
What Makes ONNX GO Unique?
-
100% Native ONNX Architecture: ONNX GO works directly on ONNX graphs without conversion layers or wrappers.
-
Visual Programming with LabVIEW: No other ONNX-based solution provides such a mature and intuitive graphical interface for both design and execution.
-
Dynamic Multi-Hardware Orchestration: It supports real-time execution across diverse hardware providers such as Intel, NVIDIA, ARM, and FPGA, with dynamic selection and load balancing.
-
Beyond AI: ONNX GO generalizes the use of computational graphs beyond machine learning to signal processing, automation, robotics, instrumentation, and scientific computing.
-
Technological Sovereignty: ONNX GO is a strategic European solution that ensures independence from cloud giants and foreign runtime dependencies.
Illustrations of the Two-Stage Orchestration Process
1. Performance Comparison: ONNX Runtime vs OpenCV
ONNX Runtime delivers approximately 40 percent faster processing than OpenCV in a typical edge detection task.
-
OpenCV: 219.9 ms
-
ONNX Runtime: 147.6 ms
2. Visual Graph Design with LabVIEW
The user creates a computational graph using LabVIEW’s graphical programming environment, specifying operations such as sine, log, or multiplication.
3. Real-Time Execution with ONNX Runtime
The designed ONNX model is executed using ONNX Runtime, with seamless integration into LabVIEW and support for hardware-specific optimization.
4. Full ONNX Graph Orchestration Pipeline
ONNX GO combines design and runtime into one fluid, dynamic pipeline that makes building and deploying AI and numerical models faster and more efficient.
ONNX GO transforms how engineers, scientists, and industrial users approach AI and numerical computation. It replaces rigid pipelines with flexible, visual, and high-performance orchestration. By combining native ONNX compatibility with runtime efficiency and graphical usability, ONNX GO is setting a new standard for accessible and sovereign technology.
A full video presentation is available here: