Nvidia unveiled a $192 Linux-based “Jetson TK1” SBC based on the 2.3GHz quad-core Cortex-A15 Tegra K1 SoC, and demoed its use in a self-driving Audi.
The Jetson TK1 is a more advanced version of Nvidia’s earlier, Tegra 3-based Jetson automotive development platform. Instead of being limited to automotive duty, however, the Jetson TK1 also is aimed at a variety of high-end embedded applications, including computer vision, robotics, medicine, and security. The 5.0 x 5.0-inch single-board computer is “the world’s first mobile supercomputer for embedded systems,” featuring 326 gigaflops of processing power, or “nearly three times more than any similar embedded platform,” claims Nvidia.
Nvidia Jetson TK1 development board
(click image to enlarge)
The Jetson TK1 was announced today at the GPU Technology conference along with a $3,000 Titan Z graphics card with two Kepler CPUs and 12GB of dedicated RAM. Nvidia also previewed a next-generation “Pascal” GPU due in 2016 that’s said to be smaller, faster, and more efficient. Pascal features “NVLink” technology to speed communications between the GPU and CPU, as well as between GPUs.
In addition, Nvidia announced that its Tegra 4 and Android-based Nvidia Shield mobile game console (shown at right) is now supported with remote streaming of games over the Internet, rather than only from a GeForce GTX-equipped laptop. The Shield, which is $50 off through the end of April, will also debut a new Android version of Valve’s Portal game. These and other Nvidia announcements may be found here.
The Jetson TK1 Developer Kit includes a version of its C/C++ “VisionWorks” toolkit based on Nvidia’s CUDA architecture, which drives the Tegra K1’s 192-core Mobile Kepler GPU. The kit also provides access to the CUDA 6.0 developer tool, featuring debuggers and profilers to develop massively parallel applications. CUDA 6.0 offers accelerated libraries for FFTs, linear algebra, sparse matrix, and image and video processing, says Nvidia.
The Tegra K1 was announced in January, and in February received and “experimental” open source driver that earned a thumbs-up from Linus Torvalds. The fifth-generation Tegra K1 adds an industry leading , 192-core Kepler GPU to what is essentially a Tegra 4-like quad-core Cortex-A15 design.
Tegra K1 block diagram showing interface connections
The Tegra K1 is expected to ship in mobile devices in the second quarter. Most of these are likely to be Android devices. Yet, Nvidia chose a more generic Linux OS for the development stack that ships with the Jetson TK1.
At GPU Technology, Nvidia demonstrated a semi-autonomous self-driving Audi car that processes video using the Jetson TK1. The car is designed to drive by itself when traveling at under 40Mph.
At CES in January, Nvidia tipped an upcoming Tegra K1 VCM (vehicle computing module) reference design, said to support semi-autonomous driving features including pedestrian detection and blind spot monitoring. Nvidia is also the sole semiconductor member of Google’s Open Automotive Alliance, a consortium for Android-powered automotive systems.
Inside the Jetson TK1
Once you get past the amazing graphics prowess of the Tegra K1, the Jetson TK1 is a rather ordinary hacker SBC. The $192 version ships with 2GB of RAM, but you can get up to 4GB. (The Tegra K1 itself supports 8GB.) A 16GB eMMC flash drive is also standard, as is a full-sized SD slot. A SATA interface is also onboard.
Jetson TK1 SBC details, block diagram
(click images to enlarge)
Coastline ports include single helpings of gigabit Ethernet, USB 3.0, micro-USB 2.0, and RS232. You also get a JTAG port and audio line-out and mic-in.
Somewhat unusually for an ARM board, the Jetson TK1 provides a Mini-PCI Express expansion slot. In addition, a pair of expansion ports provide access to DisplayPort and LVDS display options, a CSI-2 camera interface, and an SPI-based touchscreen controller. These proprietary connectors also carry signals for GPIOs, UART, HSIC, and I2C.
In addition to the aforementioned Linux, CUDA, and VisionWorks support, the Jetson TK1 supplies OpenGL 4.4 drivers, and a “complete suite of development and profiling tools, and out-of-the-box support for cameras and other peripherals,” says Nvidia. It also provides “Nvidia’s partner support networking,” including Avionic Design, GE Intelligent Platforms, ICD, SECO, and Toshiba DME.
Despite Nvidia’s growing support for open source, there are no open source promises here, at least for now. Developers must “register as a CUDA developer to get software updates and to receive access to Nvidia libraries and tools,” says the company.
Specifications listed for the Jetson TK1 SBC include:
- Processor — Nvidia Tegra K1 (4x Cortex-A15 cores @ up to 2.3GHz with 192-core Mobile Kepler GPU and power management core)
- 2GB to 4GB x16 RAM (64-bit width)
- 16GB eMMC 4.51 flash
- 4MB SPI boot flash
- SD/MMC slot
- HDMI port
- DP and LVDS signals via expansion ports
- Touch SPI 1×4 via expansion ports
- 1×1 CSI-2 camera I/O via expansion ports
- Networking — Gigabit Ethernet port (Realtek RTL8111GS)
- Other I/O:
- USB 3.0 port
- Micro-USB 2.0 (OTG)
- RS232 port
- SATA port (type A)
- JTAG connector
- Audio line-out, mic-in (Realtek ALC5639)
- GPIOs, UARTs, HSIC, and I2C via expansion ports
- Expansion — half-sized mini-PCIe slot; 2x expansion connectors (2mm pitch, 2×25 female )
- Dimensions — 5.0 x 5.0 inches (127 x 127mm)
- Power — 12V DC
- Operating system — Linux with 3.10.24 kernel image, bootloader, Nvidia drivers, and flashing utilities; requires host PC running Ubuntu 9.04 or higher
“Jetson TK1 fast tracks embedded computing into a future where machines interact and adapt to their environments in real time,” stated Ian Buck, vice president of Accelerated Computing at Nvidia. “This platform enables developers to fully harness computer vision in handheld devices, bringing supercomputing capabilities to low-power devices.”
“Having the level of performance and energy efficiency Jetson TK1 offers can potentially support the development of robots with real-time object recognition and compelling autonomous navigation capabilities,” stated Chris Jones, director of strategic technology development at iRobot Corp.
“Tegra K1 can change what’s possible in the rugged and industrial embedded market,” stated Simon Collins, product manager at GE Intelligent Platforms. “We expect to be able to offer solutions in the sub-10 watt space that previously consumed 100 watts or more.”
The Jetson TK1 is available for pre-order at $192 with 2GB of RAM, with shipments due in April. Newegg.com and MicroCenter will sell the board, says Nvidia. More information and pre-orders may be found at the Jetson TK1 product page.