Home > News content

Intel demonstrated the Lake Crest depth learning chip: 32G HBM2, 20 times PCIe

via:博客园     time:2017/2/5 15:30:32     readed:1690

Although the consumer-level chip is Intel, NVIDIA and other companies, one of the core business, but in fact the real "cash cow" but also a number of enterprise-class, industrial products, such as accelerator cards, server computing programs, Strength of the Hyun technology opportunities. Last year, Intel and NVIDIA also because of the "deep calculation" of some indicators in the end who is strong on the pinch, is eye-catching. Intel has demonstrated the progress of its new core, the Lake Crest family, for deep neural network computing (DNN), based on the Nervana platform, which claims to deliver much more performance than the GPU at the same transistor density.


It is reported that, Intel from August 2016, invested $ 350 million research and development services in the DNN Nervana hardware and software integration platform, now includes the Lake Crest family and the next generation Knights Crest, the first is scheduled for the first half of this year is the release of independent accelerator card , They should all be used in the form of ARRIA FPGA, matching Xeon processor.


Unlike the current Xeon Phi, Lake Crest is designed for AI loads. Intel said its own use Flexpoint architecture, MCM multi-chip package, equipped with 32GB HBM2 storage, internal interconnect speed is 20 times PCIe, floating-point performance 8TFLOPS.

The roadmap shows that Intel should announce Lake Crest when it launches this year's Knights Mill (Xeon Phi) and Xeon E5 V5 Skylake processors.

To say regret, may have to mention because of the lock-up order, Intel's accelerated product can not enter China.

12 operating cores

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments