The generative AI landscape is rapidly evolving, with new large language models (LLMs), visual language models (VLMs), and vision language action (VLA) models emerging daily. To stay at the forefront of this transformative era, developers need a platform powerful enough to seamlessly deploy the latest models from the cloud to the edge with optimized inferencing and open ML frameworks using CUDA.
To support emerging applications for generative AI workloads in robotics and multimodal agents, NVIDIA is refreshing the NVIDIA Jetson Orin Nano Developer Kit through a software update with an unprecedented boost in performance of up to 1.7x at an amazing new price of $249. In recognition of the superb boost in performance and accessibility of advanced AI, the Jetson Orin Nano Developer Kit is being renamed as the NVIDIA Jetson Orin Nano Super Developer Kit.
Jetson Orin Nano Developer Kit can be upgraded to Jetson Orin Nano Super Developer Kit with just a software update.
In this post, learn about the new capabilities of the developer kit and how you can seamlessly tap into the boosted performance.
The new Jetson Orin Nano Super Developer Kit
The Jetson Orin Nano Super Developer Kit now offers the following improvements:
- 1.7x higher generative AI model performance.
- 67 Sparse TOPs, a significant increase from the previous 40 Sparse TOPs
- 102 GB/s of memory bandwidth, a significant leap from the previous 65 GB/s memory bandwidth.
- 1.7 GHz of CPU clock speed, up from 1.5 GHz.
- New reduced price of $249, down from $499.
With the same hardware architecture, this performance boost is enabled by a new power mode which increases the GPU, memory, and CPU clocks. All previous Jetson Orin Nano Developer Kits can use the new power mode by upgrading to the latest version of JetPack.
With this performance boost, the Jetson Orin Nano Super Developer Kit delivers up to 70% higher generative AI performance, enabling you to run an emerging class of newer Transformer-based models. This compact yet powerful system can effortlessly handle a wide range of LLMs, VLMs, and Vision Transformers (ViTs), from smaller models to those with up to 8B parameters, such as the Llama-3.1-8B model.
Jetson supports the same ML frameworks and optimized inferencing infrastructure as other platforms, such as HuggingFace Transformers, Ollama, llama.cpp, vLLM, MLC, NVIDIA TensorRT-LLM, and more. This makes it an ideal choice for you to quickly adapt and deploy scalable solutions between the cloud, edge, and PC.
Jetson Orin Nano Super Developer Kit configuration
| NVIDIA Jetson Orin Nano Developer Kit (original) | NVIDIA Jetson Orin Nano Super Developer Kit | |
| GPU | NVIDIA Ampere architecture 1,024 CUDA Cores 32 Tensor Cores 635 MHz | NVIDIA Ampere architecture 1,024 CUDA Cores 32 Tensor Cores 1,020 MHz |
| AI PERF | 40 INT8 TOPS (Sparse) 20 INT8 TOPS (Dense) 10 FP16 TFLOPs | 67 TOPS (Sparse) 33 TOPS (Dense) 17 FP16 TFLOPs |
| CPU | 6-core Arm Cortex-A78AE v8.2 64-bit CPU 1.5 GHz | 6-core Arm Cortex-A78AE v8.2 64-bit CPU 1.7 GHz |
| Memory | 8GB 128-bit LPDDR5 68 GB/s | 8GB 128-bit LPDDR5 102 GB/s |
| MODULE POWER | 7W | 15W | 7W | 15W | 25W |
Runs a wide range of LLMs, VLMs, and ViTs
The NVIDIA Jetson Orin Nano Super Developer Kit offers performance that is a game-changer if you want to bring generative AI capabilities to your embedded applications or experience cutting-edge generative AI models in an affordable computer.
LLMs, VLMs, and vision transformers (ViTs) are transformative AI architectures driving innovation across domains. Foundation LLMs excel at generalized language processing and generating human-like text, enabling natural human/machine interfaces when coupled with streaming speech recognition and synthesis pipelines such as NVIDIA Riva.
Having advanced rapidly to compete with larger models through techniques like pruning and distillation, small language models (SLMs) are optimized for edge use cases, and are generally referred to as being <7B parameters in size.
Today, it’s common for open LLMs and SLMs to have been trained for agentic tool use and function calling with structured I/O, which connects LLM parsing and reasoning to real-world systems.
Support for LLM fine-tuning and memory-efficient LoRAs have also improved across many ML frameworks, including NVIDIA NeMo, enabling well-understood recipes for the alignment and specialization of SLMs in application-specific domains.
ViTs harness the power of Transformers for vision by tokenizing patches of pixels into embeddings. They have achieved state-of-the-art performance across fundamental tasks:
- Feature extraction
- Classification
- Detection
- Segmentation
- Pose estimation
They also extend to higher-dimensional modalities such as 3D point clouds and video with NVIDIA Cosmos Tokenizers. ViTs prompt creative new hybrid models that blend bespoke vision capabilities with open-vocabulary language support and dynamic runtime queries against a wide variety of subject matter and object classes, without needing additional fine-tuning.
VLMs such as VILA fuse ViTs and LLMs across visual and textual modalities, empowering models to understand and generate detailed scene descriptions, detecting objects of interest and extracting text with OCR. They can apply multimodal reasoning in response to user queries about images or video sequences.
Having undergone similar minification efforts to SLMs, VLMs are increasingly useful for edge applications when applied to the remote monitoring of camera streams with flexible event-based alerts and summarization, such as in the reference VLM Workflow in Jetson Platform Services.
Together, these technologies underpin advances in generative AI and multimodal systems, and real-world applications. The new NVIDIA Jetson Orin Nano Super delivers substantial performance increases on a wide variety of LLMs, SLMs, ViTs, and VLMs and significantly lowers the barrier of entry for gaining hands-on experience with the latest Transformer models as a gateway to physical AI. We benchmarked some of the popular LLMs, VLMs and ViTs below and showcase the speed up the Jetson Orin Nano Super developer kit provides over the predecessor.
| Model | Jetson Orin Nano (original) | Jetson Orin Nano Super | Perf Gain (X) |
| Llama 3.1 8B | 14 | 19.14 | 1.37 |
| Llama 3.2 3B | 27.7 | 43.07 | 1.55 |
| Qwen2.5 7B | 14.2 | 21.75 | 1.53 |
| Gemma 2 2B | 21.5 | 34.97 | 1.63 |
| Gemma 2 9B | 7.2 | 9.21 | 1.28 |
| Phi 3.5 3B | 24.7 | 38.1 | 1.54 |
| SmolLM2 | 41 | 64.5 | 1.57 |
* LLM generation performance (tokens per second) with INT4 quantization using MLC API.
| Model | Jetson Orin Nano (original) | Jetson Orin Nano Super | Perf Gain (X) |
| VILA 1.5 3B | 0.7 | 1.06 | 1.51 |
| VILA 1.5 8B | 0.574 | 0.83 | 1.45 |
| LLAVA 1.6 7B | 0.412 | 0.57 | 1.38 |
| Qwen2 VL 2B | 2.8 | 4.4 | 1.57 |
| InternVL2.5 4B | 2.5 | 5.1 | 2.04 |
| PaliGemma2 3B | 13.7 | 21.6 | 1.58 |
| SmolVLM 2B | 8.1 | 12.9 | 1.59 |
*All VILA and LLAVA models were run with INT4 precision using MLC while the rest of the models were run in FP4 precision with Hugging Face Transformers.
| Model | Jetson Orin Nano (original) | Jetson Orin Nano Super | Perf Gain (X) |
| clip-vit-base-patch32 | 196 | 314 | 1.60 |
| clip-vit-base-patch16 | 95 | 161 | 1.69 |
| DINOv2-base-patch14 | 75 | 126 | 1.68 |
| SAM2 base | 4.42 | 6.34 | 1.43 |
| Grounding DINO | 4.11 | 6.23 | 1.52 |
| vit-base-patch16-224 | 98 | 158 | 1.61 |
| vit-base-patch32-224 | 171 | 273 | 1.60 |
*All ViT models were run with FP16 precision using NVIDIA TensorRT (measurements are in FPS).
Getting started with Jetson Orin Nano Super Developer Kit
To enjoy the boosted performance, download the SD Card image from the JetPack SDK page and follow the Getting Started Guide.
NVIDIA released a SD card image based on JetPack 6.1 with support for boosted performance. You can also install JetPack 6.1 with boosted performance for Jetson Orin Nano Developer Kit using SDK Manager. Make sure that you have updated the SDK Manager and select JetPack 6.1 (rev. 1) while installing.
When you’re up and running with JetPack, change the power mode using the following command to unlock the super performance. Mode 2 is MAXN mode, which brings the super performance.
sudo nvpmodel -m 2
You can also change the power mode using the Power Mode Selector on the right side of the Ubuntu desktop’s top bar.
Experience generative AI on Jetson Orin Nano Super Developer Kit
NVIDIA offers a range of tutorials and pre-built containers in the Jetson AI Lab for exploring generative AI on the Jetson Orin Nano Developer Kit. If you’re interested in robotics, be sure to explore the LeRobot tutorial. For those looking to create a generative AI chatbot, there’s a dedicated tutorial for building one.
Hugging Face LeRobot
NVIDIA has partnered with Hugging Face to accelerate robotic research on the LeRobot open AI platform. You can run HuggingFace LeRobot on Jetson Orin Nano Super Developer Kit, which runs generative AI models for predicting actions for a particular task from visual inputs and prior trajectories.
Generative AI–driven chatbot with Ollama
Run a generative AI chatbot on Jetson Orin Nano Super Developer Kit. This chatbot features Ollama with Open WebUI, a widely used, open-source, chatbot server interface that connects to locally running LLMs. It uses retrieval-augmented generation (RAG) to further enhance the user experience and capabilities.
Advancing generative AI with Jetson AI Lab
The Jetson AI Lab is the hub for discovering and experimenting with the latest generative AI technologies optimized for edge devices.
By fostering an openly collaborative and community-driven environment, NVIDIA works alongside developers and partners to advance open-source edge AI and robot learning. With our comprehensive support for popular machine learning frameworks and optimized inferencing microservices on Jetson devices, you can quickly build and deploy the latest research innovations and models on your Jetson computers to keep up with the rapid pace of innovation.
Figure 7 shows examples of the NanoOWL open-vocabulary real-time object detection ViT.
Figure 8 shows the NanoDB multimodal vector database with interactive txt2img and img2img similarity search.
Jetson AI Lab offers almost 50 easy-to-follow tutorials and prebuilt containers to get developers of any experience level to quickly get started with on-device LLMs, SLMs, and multimodal VLMs, along with variants for VLAs, diffusion policies, and speech models. They’re all deployed locally with the same optimized inferencing infrastructure used in the cloud.
The collaborative, community-driven tutorials and resources of the Jetson AI Lab significantly reduces the barriers to entry for deploying advanced generative AI to the edge.
All Jetson Orin Nano series and Jetson Orin NX series modules gets a super performance boost
NVIDIA is also refreshing the performance across the Jetson Orin Nano series and Jetson Orin NX series. From small AI cameras to large autonomous machines, every edge device needs the compute capability to run generative AI models.
With the same module hardware, you can take advantage of the up to 1.7x increased performance across the different modules.
- Orin Nano series: Improvements to the frequencies and performance on the GPU, CPU, and memory.
- Orin NX series: Improvements across the GPU and DLA.
Support and documentation for boosted performance for Jetson Orin Nano and Orin NX production modules will be released in the first half of January with JetPack 6.1.1.
Due to the growing customer demand for Orin, NVIDIA also recently announced the extension of the product lifecycle of Jetson Orin through 2032. With this super performance boost, the Orin Nano series and Orin NX series are the ideal platforms for both current and future models.
Jumpstart your generative AI developer today
The Jetson Orin Nano Super Developer Kit is your ultimate platform for leading the way in generative AI development for edge computing. Now is the perfect time to get started and join the vibrant and diverse community of advanced developers and researchers working together with NVIDIA to address real-world challenges with physical AI.
For existing Jetson Orin Nano Developer Kit users, upgrade your JetPack SDK to unlock boosted performance today.
Stay up to date by subscribing to our newsletter, and follow NVIDIA Robotics on LinkedIn, Instagram, X, and Facebook. For more information, explore our documentation or join the Robotics community on our developer forums, Discord, and YouTube channels.
