NVIDIA brings inference to the edge with Tegra
NVIDIA (NVDA) is making efforts to increase the adoption of AI (artificial intelligence) technology in hyperscale computing, cloud data centers, and enterprise servers with its Tesla GPU (graphics processing unit) and DGX-1 supercomputer. The company is also looking to bring the AI processing platform into the embedded device space with its Jetson embedded computing platform.
Jetson is NVIDIA’s Tegra processor, supporting its Drive PX 2 autonomous car platform, its Shield consumer platform (on which a television, a set-top box, and a tablet were built), and its Spot microphone device, which connects all the above Tegra-powered devices to Google (GOOG) or Samsung’s (SSNLF) smart assistants. With Tegra, NVIDIA is looking to support the connected world at the edge.
In an attempt to bring what it describes as “inference at the edge,” NVIDIA has launched its second-generation Jetson TX hardware, enabling the device to perform parallel computing workloads without connecting to the cloud, reducing latency and reliance on mobile data. This platform is ideal for use in drones and autonomous cars, where a weak data connection and high latency could prove to be fatal.
Specifications of Jetson TX2
NVIDIA’s new Jetson TX2 SoC (system-on-chip) is based on the 14 nm (nanometer) Pascal architecture. The company packaged two 64-bit NVIDIA ARM chips, 8 GB (gigabytes) of DDR4 RAM (random access memory), and a 32 GB eMMC (embedded Multi Media Card) on a single chip.
This SoC is backed by NVIDIA’s JetPack 3.0 AI SDK (software development kit). The SDK offers the following:
- deep neural networks through TensorRT 1.0 and cuDNN 5.1
- computer vision and image processing libraries such as VisionWorks 1.6 and CUDA 8
- support for other graphics drivers and APIs (application program interface) such as OpenGL 4.5, OpenGL ES 3.2, EGL 1.4, and Vulkan 1.0
The above features help Jetson TX2 to perform computing on the edge, supporting the creation of faster, cheaper, and more accurate devices. The SoC also offers connectivity options such as 802.11ac Wi-Fi, Bluetooth, and 1 GB Ethernet for wired connections to connect to the Internet and the cloud.
Next, let’s see how the Jetson platform could drive NVIDIA’s future growth.