When the artificial intelligence program AlphaGo defeated champion Go player Lee Sedol earlier this year, everyone praised its advanced software brain. But the program, developed by Google's DeepMind research team, also had some serious hardware brawn standing behind it. The program was running on custom accelerators that Google's hardware engineers had spent years building in secret, the company said. With the new accelerators plugged into the AlphaGo servers, the program could recognize patterns in its vast library of game data faster than it could with standard processors. The increased speed helped AlphaGo make the kind of quick, intuitive judgments that define how humans approach the game.
ARM Ltd. joined a growing roster of processor specialists zeroing in on artificial intelligence and machine learning applications with the introduction of two new processor cores, one emphasizing performance, and the other efficiency. The chip intellectual property vendor unveiled its high-end Cortex-A75 paired with its "high-efficiency" Cortex A-55 processors during this week's Computex 2017 event in Taipei, Taiwan. Along with greater efficiency and processing horsepower, the chipmaker is positioning its latest processors as filling the gap in cloud computing by boosting data processing and storage on connected devices. Along with accelerating AI development, ARM also is advancing its flexible processing approach that incorporates a so-called "big" and "LITTLE" processor core configuration into a single computing cluster. That architecture is based on the assumption that the highest CPU performance is required only about 10 percent of the time.
Anytime the new cool thing comes around, all the cool engineers have to figure out the best ways to make it work. And we embark on a natural progression – or perhaps succession is a better word – sort of the way a depression along a river changes from pond to wetland to meadow. In general, you can get either performance or one or both of the other two. The thing is, though, that, if the best way to do the new thing isn't known yet, then you're likely to settle for something suboptimal for the sake of flexibility, figuring it out as you go, and perhaps fixing it after it's deployed in the field. After that, you might move to hardware of the flexible sort: FPGAs.
Today, Intel announced that its Xeon Phi processors are finally available to customers. This comes nearly a year after the company's originally-quoted launch date, and seven months after Intel announced that pre-production chips were already in use by select partners. The Xeon Phi processors feature double-precision performance in excess of 3 teraflops along with 8 teraflops of single-precision performance. All Xeon Phi processors incorporate 16GB of on-package MCDRAM memory, which Intel says is five times more power efficient as GDDR5 and offers 500GB/s of sustained memory bandwidth. The MCDRAM can effectively be used as a high-speed cache or as a complimentary addition to the system DDR4 memory.
The news was the big surprise saved for the end of a two-hour keynote at the search giant's annual Google IO event in the heart of Silicon Valley. "We have started building tensor processing units…TPUs are an order of magnitude higher performance per Watt than commercial FPGAs and GPUs, they powered the AlphaGo system," said Sundar Pichai, Google's chief executive, citing the Google computer that beat a human Go champion. The accelerators have been running in Google's data centers for more than a year, according to a blog by Norm Jouppi, a distinguished hardware engineer at Google. "TPUs already power many applications at Google, including RankBrain, used to improve the relevancy of search results and Street View, to improve the accuracy and quality of our maps and navigation," he said. The chips ride a module that plugs into a hard drive slot on server racks.