At the International Super Computing Conference held in Frankfurt, many exciting new technologies have emerged, driving extensive applications in the development of artificial intelligence and deep learning technology in all walks of life. Intel provides a wide range of comprehensive portfolios, including the upcoming Intel Strong Extensible Processors and Intel Live Programmable Door Array (FPGA), as well as the upcoming code for Knights Mill Intel® to strong melting nuclear processors, enhanced depth learning technology to a new height.
This new member of this Intel to Strong Nuclear Series is specifically optimized for deep learning training, which is expected to be put into production in the fourth quarter of 2017. The processor is designed to meet data scientists, engineers, and all users who are dedicated to machine learning technology applications. Knights Mill can greatly shorten the time of training depth learning model by making full use of low-precision calculation advantages.
Why is it so important about low accuracy?
Simply put, data scientists require hardware to accelerate in training models. In the past, the depth learning model may take a few days or even a few weeks to complete an iteration, which makes it difficult to study within a limited time. Today's hardware can shorten the training time to several hours by low precision calculation - this is equivalent to speeding up the calculation speed. As long as the hardware meets the accuracy requirements of the depth learning framework, then the most important thing is how fast the hardware training model is. Therefore, low-precision calculations can be used to solve deep learning load problems, and compared with high performance calculations, the latter typically requires single or double precision operation performance.
So the Knights Mill and the BoNights Landing are the intellikikikin. What is the difference between the processor?
We often listen to this issue that focuses on high-performance computing, artificial intelligence and machine learning.
Knights Mill uses the same overall architecture and dispensing of KNights Landing, both CPUs are second-generation Intel's Strong Relief Processor, and use the same platform. The difference is that Knights Mill uses different instruction sets to improve low-precision performance, but sacrifices a two-precision performance that is very important for many traditional high-performance computing loads. This means that Knights Mill is suitable for processing depth learning loads, while Knights Landing is more suitable for high-performance computing loads and other high-precision operations.
These different instructions are called "quad-fold fusion acquisition instructions" (QFMA: Quad Fused Multiply Add) and "Quad Virtual Neural Network Directive" (QVNNI: Quad Virtual Neural Network Instruction). QFMA can double the single precision performance of Knights Mill, while the QVNNI instruction can further reduce accuracy while meeting the accuracy requirements of the depth learning framework. The result of improving single precision performance is doubled and further reducing accuracy will enable Knights Mill to provide higher computing performance for deep learning loads than Knights Landing. In addition, improvements in frequency, power and efficiency have also promoted performance, but the command set changes are the largest factors in performance.
Take another step, the Knights Mill processor is not just to accelerate depth learning loads, and is a new processing function in an existing Intel technology-based environment. Intel is compatible with Intel's Strong® Processor. Almost all loads running on Intel to strong processors can run in Intel's to strong megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload megadownload
On the other hand, Intel is a unified depth learning practitioner uses a deep learning framework on the entire hardware platform. These are benefited from Intel Nervana Graph to bring advanced features to the depth learning framework. This calculation and execution map for neural networks allows developers to automatically optimize on multiple hardware objects, allowing users to share their software investments on different Intel platforms.
In recent years, Intel has continuously extended artificial intelligent technology layout, acquisition of deep learning and neural network chip and software field leadership manufacturers Nervana, through a series of investment and Intel Strong, combined with strong nuclear products, FPGA, provide full stack strength processing Go to the end data. The upcoming Intel? Strong Reluki Processor Knights Mill, which has significantly accelerated deep learning, driving the development of artificial intelligence.
Original link: https://www.eeboard.com/news/intel-56/
Search for the panel network, pay attention, daily update development board, intelligent hardware, open source hardware, activity and other information can make you master. Recommended attention!
[WeChat scanning picture can be directly paid]
Our other product:
