Edge Computing

Production-Ready, Enterprise-Grade Software on NVIDIA IGX Platform, Support for NVIDIA RTX 6000 ADA, and More

Decorative image of green icons on a black screen behind IGX hardware.

Real-time AI at the edge is crucial for medical, industrial, and scientific computing because these mission-critical applications require immediate data processing, low latency, and high reliability to ensure timely and accurate decision-making. The challenges involve not only high-bandwidth sensor processing and AI computation on the hardware platform but also the need for enterprise-level AI software to support the entire edge computing software stack.

Purpose-built for industrial and medical environments, NVIDIA IGX empowers organizations with the performance, durability, security, and safety required for AI at the edge. Rather than spending years designing custom systems and tailoring AI models, IGX accelerates the development of advanced AI solutions across industries with GPU-accelerated computing on NVIDIA AI Enterprise, significantly reducing time and costs.

At Computex 2024, IGX makes a major platform update with production-ready NVIDIA AI Enterprise IGX, support for NVIDIA RTX 6000 Ada dGPU, expansion of IGX systems, and more. All the enhancements come with up to 10 years of hardware and software support, enabling your organizations to confidently deliver AI safely and securely to support human and machine collaboration.

NVIDIA IGX software stack is now production-ready

With the public release of NVIDIA IGX-SW 1.0, the software stack on NVIDIA IGX is now production-ready. You can also take the NVIDIA AI Enterprise IGX long-term support and deploy your products with confidence backed by NVIDIA.

To address the increasing demand for multimodal generative AI deployment and high-bandwidth signal processing, this release brings in support for the NVIDIA RTX 6000 ADA graphics card, extending the AI compute up to 1705 TOPS. This is a 7x increase in AI performance compared with using an onboard iGPU. With such high-performance AI compute at your disposal, you can easily run demanding AI workloads and support generative AI applications. 

The release also streamlines the installation process by providing the capability to install all firmware with a simple-to-use web interface: select the firmware needed and choose Install.

Download the NVIDIA IGX-SW 1.0 public release from the Download Center, where all the software stacks are freely available for development and evaluation. 

For production, the software stack is supported by NVIDIA AI Enterprise IGX, providing enterprise-grade, long-term support with access to NVIDIA experts. With a subscription to NVIDIA AI Enterprise IGX, you get access to the following options: 

  • Production branch: Stay with the latest NVIDIA AI stack and tap into advancements and optimizations. A new production branch is released every six months and each branch is maintained for nine months. This is ideal for use cases such as retail and robotics to ensure that you can bring the latest AI models and generative AI workloads.
  • Long-term support branch: Version-locked and maintained and supported for 10 years. A new long-term-supported branch is released every 2.5 years with a new version of the stack. This is ideal for regulated use cases such as medical and scientific computing.

Both branches offer continuous monitoring for security vulnerabilities, API stability, bug fixes, technical support, and timely resolution from NVIDIA experts and engineers. For more information, see Powering Mission-Critical AI at the Edge with NVIDIA AI Enterprise IGX

To get started, download the firmware, OS, and compute stack to the NVIDIA IGX Orin Developer Kit. For more information about installation, see the NVIDIA IGX User Guide.

To get started with NVIDIA AI Enterprise IGX, request an NVIDIA AI Enterprise IGX software trial.

Enterprise-ready application frameworks designed for medical, industrial, and robotics use

NVIDIA IGX supports all the SDKs, application frameworks, and tools available for the edge from NVIDIA. 

Diagram shows NVIDIA Metropolis, NVIDIA Holoscan, and NVIDIA Isaac in the application layer, NVIDIA AI Enterprise IGX in the next layer, and NVIDIA IGX hardware (industrial grade) as the foundation of the stack.
Figure 1. NVIDIA application frameworks on IGX Orin hardware

It includes application frameworks such as the following:

Use pretrained models from NGC to cut down model development time, synthetic data generators to create synthetic data for model training, and NVIDIA TAO Toolkit [LINK] to quickly train and optimize the models to achieve the best model performance.

Multimodal AI and generative AI are becoming increasingly important to handle unstructured data and diverse environments. 

  • In the medical space, medical diagnostics, an AI copilot for surgery, AI surgical robots, and the AI agent for patient care are now possible. 
  • In manufacturing instrumentation, real-time wafer inspection and high bandwidth sensor processing is possible. 
  • In scientific computing, radar processing and radio astronomy will be powered by AI. 
  • In industrial robotics, factory automation and robotic collaboration will be safely guarded with AI. 

You can find many of those application examples and generative AI models in the /nvidia-holohub GitHub repo.

Expansion of the IGX family with NVIDIA Certified Systems

The NVIDIA IGX Orin platform introduces a brand new product, the IGX Orin 500 system-on-module. It provides flexibility to OEMs to design carrier-board and custom configurations to meet customer needs. Without sacrificing enterprise software support, IGX Orin 500 features the same iGPU, CPU, memory, and storage offered by IGX Orin 700 (previously named IGX Orin Boardkit) and the capability to enable functional safety with additional sMCU for robotics and smart agriculture applications.

Photo of the NVIDIA IGX Orin hardware on a stage with a black background.
Figure 2. NVIDIA IGX Orin product family
Table 1. Key feature comparisons
  IGX Orin 700 IGX Orin 500
Performance Up to 1705 TOPs* 248 TOPS
Specs IGPU: 2048-core NVIDIA Ampere with 64 Tensor Core
CPU: 12-core Arm Cortex A78AE CPU
Memory: 64 GB LPDDR5 with ECC
Storage: 64 GB eMMC
Power Up to 125W (without dGPU)
400W (with dGPU)
I/O Throughput 2X100 Gb/s Up to 10 Gb/s
Optional dGPU Yes No
Integrated Dual 100 Gb/s ConnectX-7 Yes No
BMC Yes No
Functional Safety Yes Yes (requires sMCU on carrier board)
Carrier Board Customization No Yes
Product Lifecycle and Enterprise SW Support 10 Years (until 2033)
Value Prop Same NVIDIA Certification process to ensure enterprise-level software support
AI safety and functional safety for industries

*1705 TOPS = 248 TOPS (GPU) + 1457 TOPS (6000 Ada dGPU)

By using the NVIDIA IGX Orin 500 Design Guide, OEM and ODM partners can get started with custom carrier board design. For all IGX systems, OEM full systems are required to pass the same NVIDIA certification process to ensure enterprise-level software support and the best AI software performance. This rigorous testing and validation reduces the risk and time involved in deploying specialized infrastructure that can support complex, computationally intensive, generative AI workloads.

NVIDIA-Certified IGX systems will soon be available from select OEM partners. For more information about the NVIDIA IGX platform and the latest breakthroughs, see the NVIDIA IGX page and the IGX Download Center.

Discuss (0)