Cooper Lake to Boost AI Inference and Training Performance by 60%

2020-01-07

At a press conference during the 2020 Consumer Electronics Show, Intel unveiled its latest advancements in artificial intelligence and machine learning. While specific details are still unclear, it offers a glimpse into the performance of the future Xeon Scalable Processor family, codenamed Cooper Lake. Cooper Lake is set to be released in the first half of 2020 and promises a 60% improvement in AI inference and training performance. One notable feature is DL Boost, which incorporates several x86 technologies to accelerate AI workloads such as vision, speech, language, and generation. It supports the bfloat16 (brain floating-point) computational format on Cooper Lake products. Bfloat16 was originally developed by Google and implemented on its third-generation tensor processing unit, a custom-designed AI accelerator chip. Each socket of Cooper Lake can accommodate up to 56 processor cores, doubling the core count of Intel's second-generation Xeon Scalable processors. Additionally, they offer higher memory bandwidth, improved AI inference capabilities, and training performance within a lower power range. Cooper Lake is compatible with the upcoming 10-nanometer Ice Lake processor platform. Intel claims that its products run more AI workloads in data centers compared to other platforms. Intel's AI chip division at Santa Clara has generated $3.5 billion in revenue this year. The market opportunity for AI chips is projected to grow from $2.5 billion in 2017 to $10 billion by 2022, with an annual growth of $1 billion since 2017. This explosive market growth has garnered significant attention from Intel, evident in its recent actions. Earlier, Intel acquired Habana Labs, an Israeli developer of cloud data center programmable AI and machine learning accelerators, for approximately $2 billion. Prior to that, in September 2016, the company acquired Movidius, a San Mateo-based specialist in low-power processor chips for computer vision designs. In 2015, Intel acquired Altera, a manufacturer of field-programmable gate array, and a year later, Nervana, which enriched its hardware platform products and laid the foundation for a new generation of AI accelerator chipsets. In August 2019, Intel acquired Vertex.ai. These are just a few examples of Intel's preparations for AI, demonstrating why its recent announcements are worth anticipating.

RECENT AI NEWS

RECENT AI TOOLS