NVIDIA released the Jetson Orin Nano development kit at the GTC 2023 spring session to improve the performance benchmark of entry-level AI devices. Let’s take a look at its specification analysis first.
A new benchmark for entry-level AI performance
The Jetson Orin Nano 8GB computing module can provide 40TOPS of INT8 data type AI computing performance under 15W power consumption mode, which is about twice as high as Jetson Xavier NX’s performance of only 21TOPS. Compared with Jetson Nano’s INT16 data type performance of only 0.5TOPS, the gap between the two is even more 80 times spectrum. In addition, Jetson Orin Nano 8’s FP32 data type CUDA core performance is also 5 times that of Jetson Nano, 6 sets of Arm Cortex -A78AE processor core also has nearly 7 times processor performance.
On the other hand, the Jetson Orin Nano 8GB computing module can also switch to the 7W power consumption mode, bringing more flexible application methods. Overall, Jetson Orin Nano can bring up to 50 times the power efficiency of Jetson Nano.
Benefiting from the improved performance of the Jetson Orin Nano platform and the ability to support more mainstream AI computer vision models, developers can use the Jetson Orin Nano development kit to create applications such as entry-level AI robots, drones, and cameras with more functions.
- Vision model supported by Jetson Orin Nano platform
- NVIDIA PeopleNet v2.3: Streamlined People Detection
- NVIDIA PeopleNet v2.5: Highest Accuracy People Detection
- NVIDIA Action Recognition Net: 2D and 3D Person Pose Recognition
- NVIDIA LPRNet: License Plate Recognition
- NVIDIA DashCamNet, BodyPoseNet: People Pose Prediction
- ResNet-50 (224×224): Image Recognition
Rich I/O functions
The I/O function of the Jetson Orin Nano development kit is similar to that of the Jetson Xavier NX development kit, and it is much richer than the Jetson Nano development kit. It can not only support microSD memory cards, but also install M.2 solid-state drives to meet higher throughput. computing needs.
Jetson Orin Nano Development Kit Specifications and Features at a Glance
Computing module part: Jetson Orin Nano 8GB computing module
Processor: 6 Arm Cortex-A78AE v8.2 64-bit cores (1.5MB L2 + 4MB L3)
GPU: 1024 sets of NVIDIA Ampere architecture CUDA cores + 32 sets of Tensor cores
Memory: 8GB 128-bit LPDDR5 (bandwidth = 68 GB/s)
Storage media: M.2 NVMe solid state drive, microSD memory card
Power consumption: 7~15W
Motherboard part: NVIDIA Reference Carrier Board
Camera terminal: 2 sets of MIPI CSI-2 22-pin Camera Connectors
M.2：Key M PCIe Gen 3×4、M.2 Key M PCIe Gen3x2、M.2 Key E PCIe (x1) / USB 2.0 / UART / I2S / I2C
USB: 4 groups of USB 3.2 Gen2, USB Type-C (Debug, Device Mode only)
Network: GbE Ethernet
Display: DisplayPort 1.2 (supports MST multi-stream transmission)
microSD card reader: support UHS-1 SDR104 mode
Other terminals: 40-Pin expansion terminal (support UART, SPI, I2S, I2C, GPIO), 12-Pin button terminal, 4-Pin fan terminal, DC power input terminal
Dimensions: 100 x 79 x 21mm (including base, computing module and heat sink)
The author will conduct an actual performance test on the Jetson Orin Nano development kit in the next article. If there is no accident, I plan to conduct a “cough cough” test, so stay tuned for our follow-up reports.
series of articles
NVIDIA Introduces Jetson Orin Nano Development Kit, Dramatically Boosts Performance of AI Entry Devices
Jetson Orin Nano development kit hands-on play (1): specification introduction and unpacking (this article)
Jetson Orin Nano development kit hands-on play (2): Performance measurement (working)
Jetson Orin Nano development kit hands-on play (3): One More Thing (working, publication time to be determined)