Currently, Intel controls 95 percent of the data center chipset market. Accordingly, the company continually develops new server semiconductors for its clients that are faster, more powerful, and more affordable. However, the firm is also interested in expanding its offerings to parallel markets. For instance, the corporation has developed a new artificial intelligence (AI) chip that offers a staggering amount of processing power.
Intel introduced its new AI-embedded chipset, code-named Keem Bay, at its November 12 AI Summit.
The processor can perform 10 trillion operations per second (TOPS) using a 6.5-watt accelerator. With it, electronics manufacturers can build high-performance, low latency smart cameras, drones, and virtual-reality headsets. Set for release in the first half of 2020, the vision processing unit (VPU) looks to give other leading AI component makers a run for their money.
A Better Mouse Trap
Intel has pursued the edge enabled, AI-empowered Internet of Things (IoT) market since the middle of this decade. In 2016, the Santa Clara, California-based corporation acquired an Irish firm called Movidius. The startup specialized in creating low-power, high-performance processors for computer vision programs.
Before being bought by Intel, the company made components for Google’s Clips camera and DJI’s consumer recreation drones. Post-acquisition, the company developed a groundbreaking new processor called Myriad X in 2018. The chipset could process 1 TOPS using a custom accelerator called a neural compute engine.
Keem Bay is Movidius’ third-generation VPU and boasts ten times the processing power of the Myriad X.
During its Keem Bay unveiling, Intel revealed that its new chipset will utilize its OpenVINO toolkit and Edge AI DevCloud. Consequently, the firm’s clients can prototype and test their AI systems in a host of different hardware configurations. With those features, it’s clear that Intel is taking aim at Nvidia’s EGX Edge Super Computing platform.
Furthermore, the firm is working to lure in customers by offering a comparable interface with superior hardware. At its AI event, Intel claimed that Keem Bay is capable of four times the TOPS as Nvidia’s Xavier chipset. Moreover, the chipmaker said that its product provides a superior level of performance with one-fifth of its competitor’s power consumption.
Intel also used the summit to assert hardware superiority over Chinese conglomerate Huawei. The American company boasted that its forthcoming chipset has 1.25 times the processing power of the Sino firm’s HiSilicon Ascend 310. Plus, the manufacturer said that Keem Bay outperforms its overseas counterpart in energy efficiency by a factor of three.
This year, Intel’s AI revenue will exceed $3.5 billion, a threefold increase from 2017. Even though the firm estimates that the deep learning chipset market will hit $40 billion by 2022, its AI component segment has a lot of room to grow.
For instance, Google and UPS both have drone delivery in their near-term roadmaps. Those firms will need chipsets that can perform vast amounts of computations at the edge to complete their deliveries. With Keem Bay, Intel has positioned itself as the potential best-in-class supplier of those components.