Page 13

EETE APR 2014

Five new graphics technologies at Nvidia’s GPU Technology Conference By Nick Flaherty Nvidia has revealed five new technologies at its GPU Technology Conference (GTC). The conference offers a vision of the future with high speed interconnect and multiple GPUs in a wide range of embedded platforms from aerospace and defence through to vision systems. Like Imagination Technologies, it is seeing ray tracing maturing and is using its chips to push ray tracing into the data centre. If you want a supercomputer of your own, then the latest graphics card will provide 8TFLOPS for a mere $3000. GPU interconnect for exascale computing Nvidia and IBM have developed a new high-speed interconnect, called NVLink, that Nvidia plans to use into its future GPUs. This will provide data rates five to 12 times faster than today and pave the way for a new generation of exascale supercomputers that are 50-100 times faster than today’s most powerful systems. NVLink technology is being used in the Pascal GPU architecture expected to be introduced in 2016 and it will also be used in future versions of IBMs POWER CPUs. With NVLink technology tightly coupling IBM POWER CPUs with Tesla GPUs, the POWER data centre ecosystem will be able to fully leverage GPU acceleration for a diverse set of applications, such as high performance computing, data analytics and machine learning. Faster data movement coupled with Unified Memory will simplify GPU accelerator programming, allowing the programmer to treat the CPU and GPU memories as one block of memory. The programmer can operate on the data without worrying about whether it resides in the CPUs or GPUs memory. Supercomputer graphics in aerospace and defence GE Intelligent Platforms is to develop and manufacture rugged high performance embedded computing (HPEC) and graphics systems using Nvidia’s Tegra K1 mobile processor. GE will also be Nvidia’s preferred provider of the new technology to serve applications in harsh environments, most notably to customers in the military/aerospace market. The Tegra K1 has 192 fully programmable GPU computing cores capable of executing CUDA algorithms in addition to four ARM Cortex-A15 CPU cores with a fifth battery-saver core, to deliver record levels of performance and battery life for a 10W power envelope. The reductions in heat dissipation and weight are key for highly SWaP (size, weight and power) constrained environments that increasingly characterize military/aerospace deployments such as man-wearable and portable systems. “We are working with customers for whom GPU technology is an ideal fit for their sophisticated but SWaP-constrained applications but who have been precluded from using it because of its power consumption and heat dissipation”, said Simon Collins, Product Manager, GE Intelligent Platforms. Tegra K1 changes all that, and will allow those customers to implement the optimum solution. Tegra K1 allows GE to uniquely address a much broader spectrum of potential applications. The Tegra K1 also enhances GEs HPEC capability, said Collins. Previously, our HPEC offering ranged in power consumption from around 2,000 watts at the high end to 100 watts at the entry level. In future, we will be able to offer HPEC solutions that consume as little as 10 watts, allowing CUDA algorithms to be easily ported across an even broader range of platforms. Embedded systems Computer vision, image processing and real-time data processing applications can get supercomputer performance with the launch of a developer platform based on the world’s first mobile supercomputer for embedded systems. The Jetson TK1 Developer Kit provides developers with the tools to create systems and applications that can enable robots to seamlessly navigate, physicians to perform mobile ultrasound scans, drones to avoid moving objects and cars to detect pedestrians. With 326 gigaflops performance, the Jetson TK1 Developer Kit includes a full C/C++ toolkit based on Nvidias CUDA architecture, making it easier to program than FPGAs, custom ASICs and DSP processors. Jetson TK1 fast tracks embedded computing into a future where machines interact and adapt to their environments in real time, said 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 lowpower devices. The Jetson TK1 Developer Kit comes with the full support of the CUDA 6.0 developer tool suite, including debuggers and profilers to develop massively parallel applications. CUDA 6 also brings to the ARM platform Nvidia’s accelerated libraries for FFTs, linear algebra, sparse matrix, plus image and video processing. It can be pre-ordered starting today for $192, from Avionic Design, SECO and Zotac in Europe. www.electronics-eetimes.com Electronic Engineering Times Europe April 2014 13


EETE APR 2014
To see the actual publication please follow the link above