HPC (high-performance computing) servers, which have notably embraced the GPGPU (general-purpose computing on graphics processing units) concept in recent years, are increasingly being employed for computer vision and other deep learning-based applications. Beginning in late 2014, NVIDIA supplemented its general-purpose CUDA toolset for GPU-accelerated heterogeneous computing with its proprietary CuDNN software library, which codifies the basic mathematical and data operations at the core of deep learning training.

At this year’s Consumer Electronics Show, I walked many miles and saw countless demos. Several of these demos were memorable, but one in particular really got my mental gears turning: Microsoft’s HoloLens.

Qualcomm formally unveiled the Snapdragon 835, its latest premium-tier mobile processor, at the Consumer Electronics Show earlier this month after a last-fall "tease".

 “…software is changing at head-spinning rates, driven by advances in how computers process giant amounts of data using artificial-intelligence techniques,” said the Wall Street Journal in a recent article, highlighting the challenge faced by smartphone software developers.

In March, InsideDSP covered the CEVA-X4, the first member of the company's cellular DSP core family based on its CEVA-X architecture framework, and focused on smartphones and tablets that support gigabit-per-second LTE-Advanced and LTE-A Pro protocols.

NVIDIA was an early and aggressive advocate of leveraging graphics processors for other massively parallel processing tasks (often referred to as general-purpose computing on graphics processing units, or GPGPU). The company's CUDA software toolset for GPU computing has to date secured only modest success in mobile and desktop PCs; with game physics processing acceleration, for example, along with still and video image processing acceleration. However, GPGPU has been embraced in the HPC (high-performance computing) server space, and NVIDIA is the dominant supplier of GPUs for HPC.

Earlier this week, Google announced the spin-off of its self-driving car project into a stand-alone business. Will Google become a major player in the automotive industry? Today, that idea seems far-fetched. On the other hand, 15 years ago Apple was a personal computer company, and few would have guessed that it would eventually become a dominant player in consumer electronics and photography.

There is now little question that deep learning is an effective means for a wide range of detection and recognition applications. It is increasingly used in computer vision, where it has vastly improved accuracy rates for object recognition. In some cases, the effectiveness of deep learning has also resulted in unrealistic—or, as-yet unrealizable—expectations. For example, one customer came to BDTI with a long list of detection and recognition functions that it wanted implemented on a low-cost embedded processor using a neural network.

Look back over the history of processors, and you'll see many examples of tasks initially restricted to running on high-end processors that, once they became popular and standardized, eventually attracted specialized co-processor or processor support (Figure 1).

Programmable logic IP cores, intended for integration within a broader-function ASIC, are a long-discussed and –explored product option that has yet to achieve more than niche adoption. Small FPGA IP players such as Flex Logic Technologies and Menta linger in the market, but picking such a supplier for your next SoC can be a gamble; what happens if they get acquired or otherwise disappear?