NVIDIA announces Jetson Orin Nano Super developer kit, a small AI board with 1.7 times the AI performance at almost half the price of conventional boards
NVIDIA has announced the Jetson Orin Nano Super developer kit, a small AI board kit for developers. Compared to the previous model, the AI performance has been improved by up to 70%, and the price has been reduced by almost half from the previous model's $499 (about 76,000 yen) to $249 (about 38,000 yen).
Jetson Orin Nano Super Developer Kit | NVIDIA
NVIDIA Unveils Its Most Affordable Generative AI Supercomputer | NVIDIA Blog
https://blogs.nvidia.com/blog/jetson-generative-ai-supercomputer/
In the following video, NVIDIA CEO Jensen Huang introduces the Jetson Orin Nano Super.
Introducing NVIDIA Jetson Orin™ Nano Super: The World's Most Affordable Generative AI Computer - YouTube
The photo was taken in the kitchen of CEO Huang's villa, where the company has previously announced graphics boards and chips. The kitchen is decorated for Christmas.
CEO Juan removes the tray from the oven.
On the tray was a Jetson Orin Nano Super board.
The Jetson Orin Nano Super is a board for developers interested in developing skills in generative AI, robotics, and image and video recognition. NVIDIA claims that the generative AI inference performance of the previous generation product was 40 TOPS (INT8), while the Jetson Orin Nano Super announced this time has 67 TOPS (INT8), a 70% improvement.
In addition, because TOPS performance alone is not enough to improve the performance of generative AI models, the memory bandwidth of the Jetson Orin Nano Super has been increased to 102GB/s, approximately 1.5 times the previous 68GB/s. In addition, the CPU frequency has been improved to 1.7GHz from 1.5GHz of the previous generation. The main specifications of the Jetson Orin Nano Super are as follows:
| Jetson Orin Nano Super | |
| CPU | Arm Cortex-A78AE (Armv8.2 64-bit) x 6 cores Cache: L2: 1.5MB L3: 4MB |
| GPU | NVIDIA Ampere Architecture (1024 CUDA cores + 32 Tensor cores) |
| Memory | 8GB 128-bit LPDDR5 |
| Storage | SD card and NVMe compatible |
| I/O | 2x MIPI CSI-2 22-pin camera connectors 2 x USB 3.2 Gen2 Type-A ports USB Type-C (UFP) Gigabit Ethernet DisplayPort 1.2 40-pin expansion header |
| Video Encoding | 1080p30fps (supported by CPU 1-2 cores) |
| Video Decoding | 4K60fps(H.265) x1 4K30fps(H.265) x2 1080p60fps(H.265) x5 108030fps(H.265) x11 |
| size | 103mm x 90.5mm x 34.77mm |
| Power Consumption | 7 to 25 W |
NVIDIA claims that the Jetson Orin Nano Super can now run Transformer-based models such as large-scale language models and visual language models with up to 8B parameters. It also maintains compatibility with major ML frameworks such as HuggingFace Transformers , Ollama , llama.cpp , vLLM , MLC , and NVIDIA TensorRT-LLM . In addition, the Jetson Orin Nano Super developer kit is priced at $249, almost half the price of the previous generation's $499.
According to NVIDIA, the performance improvements are achieved through new power modes that increase the GPU, memory, and CPU clocks, and existing Jetson Orin Nano developer kit users can get the same performance improvements by upgrading to the latest JetPack.
Related Posts:
