NVIDIA Jetson TX2 Doubles Intelligence

NVIDIA introduced the Jetson TX2 and JetPack 3.0 AI SDK. Jetson is the world's leading low-power embedded platform that provides server-level AI computing performance for edge devices anywhere. The Jetson TX2 features an integrated 256-core NVIDIA Pascal GPU, a six-core ARMv8 64-bit CPU complex, 8GB of LPDDR4 memory, and a 128-bit interface. The CPU complex combines a dual-core NVIDIA Denver 2 with a quad-core ARM Cortex-A57. The Jetson TX2 module (shown in Figure 1) is suitable for small size, weight and power consumption in size, weight and power (SWaP) sizes of 50 x 87 mm, 85 g and 7.5 watt typical energy consumption.

Internet of Things (IoT) devices are often used as simple gateways for relaying data. They rely on cloud connections to perform heavy work and digital processing. Edge computing is an emerging paradigm that uses local computing to implement data source analysis. With more than TFLOP / s performance, the Jetson TX2 is ideally suited for deploying advanced AI to remote field locations that lack or costly Internet connections. Jetson TX2 also provides near real-time responsiveness and minimal latency for smart machines that require mission-critical autonomous functions.

The Jetson TX2 is based on a 16nm NVIDIA Tegra "Parker" system-on-chip (SoC) (a block diagram is shown in Figure 2). Jetson TX2's deep learning extrapolates energy efficiency twice that of its predecessor, the Jetson TX1, and performs better than the Intel Xeon Server CPU. This jump in efficiency redefines the possibility of extending advanced AI from the cloud to the edge.

Figure 2: NVIDIA Jetson TX2 Tegra "Parker" SoC block diagram with NVIDIA Pascal GPU, NVIDIA Denver 2 + ARM Cortex-A57 CPU cluster and multimedia acceleration engine (click image for full resolution). Figure 2: NVIDIA Jetson TX2 Tegra "Parker" SoC block diagram with NVIDIA Pascal GPU, NVIDIA Denver 2 + ARM Cortex-A57 CPU cluster and multimedia acceleration engine (click image for full resolution).

The Jetson TX2 has multiple multimedia streaming engines that keep their Pascal GPUs providing data by offloading sensor acquisition and distribution. These multimedia engines include six dedicated MIPI CSI-2 camera ports, each providing 2.5 Gb/s of bandwidth, dual image service processor (ISP) providing 1.4 Gigapix/s processing power, and 4K video support for H.265 The decoder is 60 frames per second.

Jetson TX2 uses NVIDIA cuDNN and TensorRT libraries to accelerate leading-edge deep neural network (DNN) architectures, supporting recurrent neural networks (RNN) , long-term short-term memory networks (LSTM), and online reinforcement learning . Its dual CAN bus controller allows the autopilot integration to control robots and drones using DNNs to sense the surrounding world and operate safely in a dynamic environment. The Jetson TX2 software is available via NVIDIA's JetPack 3.0 and Linux For Tegra (L4T) Board Support Packages (BSP).

Table 1 compares the characteristics of Jetson TX2 and previous generation Jetson TX1.


NVIDIA
Jetson TX1 NVIDIA
Jetson TX2 Central Processor ARM Cortex-A57 (Quad) @ 1.73GHz ARM Cortex-A57 (Quad) @ 2GHz +
NVIDIA Denver2 (Dual Core) @ 2GHz GPU 256 Core Maxwell @ 998MHz 256 Core Pascal @ 1300MHz Memory 4GB 64bit LPDDR4 @ 1600MHz | 25.6GB / s 8GB 128bit LPDDR4 @ 1866Mhz | 59.7GB / s Storage 16GB eMMC 5.1 32GB eMMC 5.1 Encoder * 4Kp30, (2x) 1080p60 4Kp60, (3x) 4Kp30, (8x) 1080p30 Decoder* 4Kp60, (4x) 1080p60 (2x) 4Kp60 Cameras † 12 lanes MIPI CSI-2 | 1.5 Gb/s per channel | 14 million Pixel/second ISP 12 lanes MIPI CSI-2 | 2.5 Gb/sec per channel | 14 MP/sec ISP display 2x HDMI 2.0 / DP 1.2 / eDP 1.2 | 2x MIPI DSI wireless 802.11a / b / g / n / ac 2 × 2 867Mbps | Bluetooth 4.0 802.11a / b / g / n / ac 2 × 2 867Mbps | Bluetooth 4.1 Ethernet 10/100/1000 BASE-T Ethernet USB USB 3.0 + USB 2.0 of PCIe Gen 2 | 1 × 4 + 1x1 Gen 2 | 1x4 + 1x1 or 2x1 + 1x2 Can not Support Dual CAN Bus Controller Miscellaneous I/O UART, SPI, I2C, I2S, GPIO Socket 400-pin Samtec Board to Board Connector, 50x87mm warm current -25°C to 80°C power †† 10W 7.5W Price 1K Unit 299 USD 1K Unit 399 USD Table 1: Comparison of Jetson TX1 and Jetson TX2.
* Supported video codecs: H.264, H.265, VP8, VP9
† MIPI CSI-2 Bifurcation: Up to 6 2-channel or 3-channel 4-channel cameras
‡ Operating temperature range, TTP maximum junction temperature.
典型 Typical power consumption under load, input ~5.5-19.6 VDC, Jetson TX2: Maximum Q value curve. Double performance and twice the efficiency

In my article on JetPack 2.3 , I demonstrated how NVIDIA TensorRT improves Jetson TX1 deep learning inference performance, which is 18 times more efficient than a desktop CPU. TensorRT optimizes the production network by using graph optimization, kernel fusion, half-precision floating point calculation (FP16) and architecture tuning to significantly improve performance . In addition to utilizing the Jetson TX2's hardware support for the FP16, NVIDIA TensorRT can also process multiple images simultaneously in batches for higher performance.

Together, Jetson TX2 and JetPack 3.0 take the Jetson platform's performance and efficiency to a whole new level, providing users with twice the performance of Jetson TX1 or twice the performance of AI applications. This unique feature makes the Jetson TX2 ideal for products that require high-efficiency AI at the edge and high-performance products near the edge. The Jetson TX2 is also compatible with the Jetson TX1 and offers easy upgrade opportunities for products designed using the Jetson TX1.

To benchmark the performance of the Jetson TX2 and JetPack 3.0, we compared it to the server-class CPU Intel Xeon E5-2690 v4 and used the GoogLeNet depth image recognition network to measure deep learning inference throughput (images per second). As shown in Figure 3, the Jetson TX2 with less than 15W power consumption performs better than the nearly 200W CPU, enabling data center-level AI functionality at the edge.

Figure 3: Performance of the GoogLeNet network architecture analyzed on the NVIDIA Jetson TX2 and Intel Xeon E5-2960 v4. Figure 3: Performance of the GoogLeNet network architecture analyzed on the NVIDIA Jetson TX2 and Intel Xeon E5-2960 v4.

This excellent AI performance and efficiency of the Jetson TX2 comes from the new Pascal GPU architecture and dynamic energy profiles (Max-Q and Max-P), the optimized deep learning library included with JetPack 3.0, and large memory bandwidth.

Max-Q and Max-P

Jetson TX2 is designed for peak processing efficiency at 7.5W power. This level of performance (called Max-Q) represents the peak power/throughput curve. Each component on the module, including the power supply, is optimized to provide the highest efficiency at this time. The GPU's Max-Q frequency is 854 MHz and the ARM A57 CPU is 1.2 GHz. The L4T BSP in JetPack 3.0 includes a preset platform configuration for setting the Jetson TX2 to Max-Q mode. JetPack 3.0 also includes a new command line tool called nvpmodel that switches configuration files at runtime.

Although dynamic voltage and frequency scaling (DVFS) allows Jetson TX2's Tegra "Parker" SoC to adjust the clock speed based on user load and power consumption at run-time, the Max-Q configuration sets the upper clock limit to ensure that the application is running only at the most effective In the range. Table 2 shows the performance and energy efficiency of Jetson TX2 and Jetson TX1 when running GoogLeNet and AlexNet deep learning benchmarks. The performance of Jetson TX2 running in Max-Q mode is similar to that of Jetson TX1 running at the maximum clock frequency, but the power consumption is only half, so the energy efficiency is doubled.

Although mostly

Power Wall Solar Battery

Power Wall Solar Battery is designed for home using . It's very convenient for collecting the electronic power from solar or wind and grid. It has slim body , portable for transport. Easy to use the power any time any where.

UFO Power Wall Solar Battery offer new function for customers , bluetooth which is very easy for connecting mobile, shows all detailed specification on mobile. The touchscreen function is visual operation window for visual control.

Power Wall Solar Battery,Wall Mounted Battery,Battery For Smart Home,Bluetooth Lithium Battery

ShenZhen UFO Power Technology Co., Ltd. , https://www.ufobattery.com