DEV Community

Cover image for First Look at NVIDIA Jetson Orin Nano Super - The Most Affordable Generative AI Supercomputer
Ajeet Singh Raina
Ajeet Singh Raina

Posted on • Edited on

First Look at NVIDIA Jetson Orin Nano Super - The Most Affordable Generative AI Supercomputer

Table of Contents

NVIDIA has just reinvented edge computing with its latest offering - the Jetson Orin Nano Super Developer Kit. This isn't just an incremental update; it's a significant leap forward in bringing generative AI capabilities to the edge at an unprecedented price point of $249.

Comparing Jetson Nano Vs Jetson Orin Nano Super

The NVIDIA Jetson Orin Nano Super Developer Kit is a compact, yet powerful computer that redefines generative AI for small edge devices.

It delivers up to 67 TOPS of AI performance—a 1.7X improvement over its predecessor—to seamlessly run a wide variety of generative AI models, like vision transformers, large language models, vision-language models, and more.

It provides developers, students, and makers with the most affordable and accessible platform with the support of the NVIDIA AI software and a broad AI software ecosystem to democratize generative AI at the edge. Existing Jetson Orin Nano Developer Kit users can experience this performance boost with just a software upgrade, so everyone can
now unlock new possibilities with generative AI.

Learn about Jetson AI Lab

jetson_orin

Let's dive into the specs and see how it compares to its predecessors:

Feature Orin Nano Original Orin Nano Super Improvement
GPU Architecture NVIDIA Ampere (1024 CUDA cores, 32 Tensor cores) @ 635 MHz NVIDIA Ampere (1024 CUDA cores, 32 Tensor cores) @ 1020 MHz 1.6x GPU Clock
AI Performance 40 TOPS (Sparse) / 20 TOPS (Dense) 67 TOPS (Sparse) / 33 TOPS (Dense) 1.7x AI Performance
CPU 6-core Arm Cortex-A78AE @ 1.5 GHz 6-core Arm Cortex-A78AE @ 1.7 GHz 1.13x CPU Clock
Memory 8GB 128-bit LPDDR5 @ 68 GB/s 8GB 128-bit LPDDR5 @ 102 GB/s 1.5x Memory Bandwidth
Module Power 7W/15W 7W/15W/25W Additional Power Mode

How powerful is NVIDIA Jetson Orin Super?

orin_super_two

The most striking aspect of the Super variant is its performance improvements:

  • 1.7x increase in AI compute performance (67 TOPS vs 40 TOPS)
  • 1.5x increase in memory bandwidth (102 GB/s vs 68 GB/s)
  • Higher GPU and CPU clock speeds for better overall performance

Generative AI Capabilities

The NVIDIA Jetson™ platform runs the NVIDIA AI software stack, with a variety of available use-case-specific application frameworks. These include NVIDIA Isaac™ for robotics, NVIDIA Metropolis for vision AI, and NVIDIA Holoscan for sensor processing. You can save significant time with NVIDIA Omniverse™ Replicator for synthetic data generation (SDG) and NVIDIA TAO Toolkit for fine-tuning pretrained AI models from the NVIDIA® NGC™ catalog.

One of the most impressive aspects of the Orin Nano Super is its ability to run various types of generative AI models:
Large Language Models (LLMs):

Model Performance Gain
Llama-3.1 8B 1.37x
Llama 3.2 3B 1.55x
Qwen2.5 7B 1.53x
Gemma2 2B 1.63x
Gemma2 9B 1.28x
Phi 3.5 3B 1.54x
SmoLLM2 1.7B 1.57x

Vision Language Models (VLMs):

Model Performance Gain
VILA 1.5 3B 1.51x
VILA 1.5 8B 1.45x
LLAVA 1.6 7B 1.36x
Qwen2-VL-2B 1.57x
InternVL2.5-4B 2.04x
PaliGemma2-3B 1.58x
SmoLVLM-2B 1.59x

Vision Transformers

Model Performance Gain
clip-vit-base-patch32 1.60x
clip-vit-base-patch16 1.69x
DINOv2-base-patch14 1.68x
SAM2 base 1.43x
Grounding-DINO 1.52x
vit-base-patch16-224 1.61x
vit-base-patch32-224 1.60x

I/O and Connectivity

Interface Specification
Camera 2x MIPI CSI-2 22-pin Camera Connectors
PCIe M.2 Key M x4 PCIe Gen 3
Additional PCIe M.2 Key M x2 PCIe Gen3
Expansion M.2 Key E PCIe (x1), USB 2.0, UART, I2S, and I2C
USB 4x USB 3.2 Gen2 Type A + 1x Type C for Debug
Network 1x GbE Connector
Display DisplayPort 1.2 (+MST)
Storage microSD slot (UHS-1 cards up to SDR104 mode)
GPIO 40-Pin Expansion Header

Developer-Friendly Features

Feature Description
Software Stack Full support for TensorRT-LLM
Framework Compatibility Native compatibility with popular frameworks
Jetson Ecosystem Jetson Software Stack & Microservices support
Deployment Pre-built containers for rapid deployment

AI Development Tools

Tool Description
TensorRT Optimization Optimized inference using TensorRT
Quantization Support INT8/FP16 quantization support
Multi-Model Inference Ability to run multiple models simultaneously
Containerization Docker container support for easy deployment

Getting Started with Jetson Orin Super

Image8

This guide will walk you through setting up Ollama on your Jetson device, integrating it with Open WebUI, and configuring the system for optimal GPU utilization. Whether you're a developer or an AI enthusiast, this setup allows you to harness the full potential of LLMs right on your Jetson device.

Pre-requisite

  1. Jetson Orin Nano
  2. A DC power supply
  3. 64GB/128GB SD card
  4. WiFi Adapter
  5. Wireless Keyboard
  6. Wireless mouse

Software

  • Download Jetson SD card image

Ensure that you download the latest JetPack 6.2 SDK from this link. Your Jetson Orin Nano Developer Kit comes with an old firmware flashed at the factory, which is NOT compatible with JetPack 6.x. Click here to download

  • Etcher installed on your local system

Download Jetson SDK using this link

Preparing Your Jetson Prin Nano

  1. Unzip the SD card image
  2. Insert SD card into your system.
  3. Bring up Etcher to flash SD card image into the SD card

Prerequisite

  • Ensure that you have Jetpack 6.0 installed on your Jetson Orin Nano device. You can download the SDK Manager on the remote Windows or Linux and follow the tutorial from the official NVIDIA Developer site.

To flash the Orin Nano using the SDK Manager, it must first be put into “recovery mode.” To do that, attach a jumper or jumper wire between the FC_REC and GND pins (pins 2 and 3) on the underside of the Orin Nano card. Refer this blog to know more

Image8

jumper-pins

Step 1. Verify L4T Version

To check the L4T (Linux for Tegra) version on your NVIDIA Jetson device (e.g., Jetson Nano, Jetson Xavier), follow these steps:

Run the following command to retrieve your current L4T version.

head -n 1 /etc/nv_tegra_release

Here are the list of supported L4T versions:

  • 35.3.1
  • 35.4.1
  • 35.5.0
  • 36.3.0

If your L4T version does not match the supported versions listed above, you may need to re-flash the system on your NVIDIA Jetson device. You might need to use SDK Manager on another computer to re-flash the device. You can download the SDK Manager and follow the tutorial from the official NVIDIA Developer site.

Step 2. Keep apt up to date:

   sudo apt update && sudo apt upgrade
Enter fullscreen mode Exit fullscreen mode

Step 3. Install jetpack:

   sudo apt install jetpack
Enter fullscreen mode Exit fullscreen mode

Step 4. Add users

Add your user to the docker group and restart the Docker service to apply the change:

   sudo usermod -aG docker $USER && \
   newgrp docker && \
   sudo systemctl daemon-reload && \
   sudo systemctl restart docker
Enter fullscreen mode Exit fullscreen mode

Step 5. Install jetson-examples:

   pip3 install jetson-examples
Enter fullscreen mode Exit fullscreen mode

Step 6. Reboot system

   sudo reboot
Enter fullscreen mode Exit fullscreen mode

Step 7. Install Ollama

   reComputer run ollama
Enter fullscreen mode Exit fullscreen mode

Optional: If you run the above command via ssh and encounter the error command not found: reComputer, you can resolve this by executing the following command:

   source ~/.profile
Enter fullscreen mode Exit fullscreen mode

Step 8. Run a model

The smallest LLaMA model available for download is TinyLlama, a compact 1.1 billion parameter model. Despite its reduced size, TinyLlama demonstrates remarkable performance across various tasks, making it suitable for applications with limited computational resources. You can access TinyLlama through its GitHub repository or via Hugging Face.

Let's run the tinyllama model and perform tasks like generating Python code:

ollama run tinyllama
>>> > Can you write a Python script to calculate the factorial of a number?
Sure! Here’s the code:

def factorial(n):
    if n == 0 or n == 1:
        return 1
    else:
        return n * factorial(n - 1)

num = int(input("Enter a number: "))
print(f"The factorial of {num} is {factorial(num)}")
Enter fullscreen mode Exit fullscreen mode

Step 9. Install models (e.g. llama3.2) from Ollama Library

ollama pull llama3.2
Enter fullscreen mode Exit fullscreen mode

Step 9. Install and run Open WebUI through Docker

docker run -d -p 3000:8080 --gpus all --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:cuda
Enter fullscreen mode Exit fullscreen mode

Step 10. Install and run Open WebUI through docker

Once the installation is finished, you can access the GUI by visiting YOUR_SERVER_IP:3000 in your browser.

Access the API endpoints by navigating to YOUR_SERVER_IP/ollama/docs#/. For comprehensive documentation, please refer to the official resources: the Ollama API Documentation (recommended) and Open WebUI API Endpoints.

Using GPU

This installation method uses a single container image that bundles Open WebUI with Ollama, allowing for a streamlined setup via a single command. Choose the appropriate command based on your hardware setup:

sudo docker run -d -p 3000:8080 --gpus=all -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:ollama
Enter fullscreen mode Exit fullscreen mode

Using CPU only

For CPU Only: If you're not using a GPU, use this command instead:

sudo docker run -d -p 3000:8080 -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:ollama
Enter fullscreen mode Exit fullscreen mode

Both commands facilitate a built-in, hassle-free installation of both Open WebUI and Ollama, ensuring that you can get everything up and running swiftly.

Once configured, Open WebUI can be accessed at http://localhost:3000, while Ollama operates at http://localhost:11434. This setup provides a seamless and GPU-accelerated environment for running and managing LLMs locally on NVIDIA Jetson devices.

Conclusion

The Jetson Orin Nano Super Developer Kit represents a significant milestone in edge AI computing. It brings datacenter-class AI capabilities to the edge at an unprecedented price point, making it an ideal platform for developers, researchers, and businesses looking to deploy advanced AI applications at the edge.

The combination of increased AI performance, enhanced memory bandwidth, and broad model support makes it a compelling choice for anyone serious about edge AI development. At $249, it's not just a product - it's a revolution in accessible AI computing.

References

Top comments (9)

Collapse
 
ajeetraina profile image
Ajeet Singh Raina • Edited

Quick Update[17 feb 2025]::

I highly recommend using SSD instead of microSD card for AI workloads. I ordered "Crucial P3 500GB PCIe 3.0 3D NAND NVMe M.2 SSD, up to 3500MB/s - CT500P3SSD8" from Amazon. Will keep you all posted on the performance.

Expecting NVMe SSDs to reach atleast 3500 MB/s+ (35x faster than SD cards). Need a faster boot times, quicker model loading, and improved system responsiveness.

Collapse
 
ajeetraina profile image
Ajeet Singh Raina • Edited

Quick Update: 22- Feb 2025:

Image description

My containers just got a power boost! ⚡ Upgraded my Jetson Orin Nano with a Crucial P3 NVMe SSD—more speed, more storage, and extra energy for AI workloads! 🚀 #JetsonOrin #EdgeAI #Containers #NVMePower

Collapse
 
nathaniel_palmieri_9d5830 profile image
Nathaniel Palmieri

Thank you for your insights.

Collapse
 
exaptedai777 profile image
Exapted Aint

Will this run any GPU-requiring code or is it limited LLM installs only?

Collapse
 
ajeetraina profile image
Ajeet Singh Raina • Edited

It’s not just restricted to LLM but run any GPU requiring code.

Collapse
 
darkpowerxo profile image
Sam Abtahi • Edited

Anyone knows where to buy this in canada? Other than the the 999$ option on amazon instead of the 250 ?

Collapse
 
ajeetraina profile image
Ajeet Singh Raina • Edited
Collapse
 
click2install profile image
click2install

Triton makes it super easy to spin up a model using various backends.

Collapse
 
ajeetraina profile image
Ajeet Singh Raina

That's on my wishlist:

docker run --gpus all --rm -p 8000:8000 -p 8001:8001 -p 8002:8002 \
    -v /path/to/model/repository:/models \
    nvcr.io/nvidia/tritonserver:23.12-py3 \
    tritonserver --model-repository=/models
Enter fullscreen mode Exit fullscreen mode