Silicon Valley new company Wave Computing specializes in the development of machine learning chips, according to the location and use of information, customized different systems and programs, can be expected to shake Nvidia in the field of machine learning chip leading position. The company, founded in 2009, has raised more than $100 million trillion in funds. According to electronic design, Wave Computing CEO Derek Meyer says their chips are hundreds of times times faster than the graphics chip (GPU) and can crawl millions of of images or hours of speech, specializing in the training and inference of AI algorithms. . Wave computing used to develop a general-purpose chip, but the 2013 business model has made a big turn, after all, machine learning software began to develop rapidly, and then more new companies such as Graphcore and Cerebras systems join the war, now wave Computing does not directly sell generic chips. Meyer said that they would deploy different systems and programs according to the location and use of the information. Currently on the market most of the new chips are accelerator, with thousands of core, can be faster and more efficient implementation of machine learning tasks, but the Nvidia GPU and other accelerators still rely on individual processor instructions, not only power consumption, but also easy to delay the problem. The Wave computing processor is between a field programmable logic gate Array (FPGA) and a special application integrable circuit (ASIC). Meyer said that in the AI world, software is changing almost every day, and the chip must adapt to the algorithm of continuous change. The question, however, is whether customers are willing to move from Nvidia's GPU to wave computing technology. Nvidia is dedicated to expanding the use of the GPU and supporting each of the major machine learning environments, including TensorFlow and Caffe, but wave computing currently supports only TensorFlow. Wave computing still needs to pay more effort to educate customers, and in 2017 has introduced the wave computing infrastructure to potential customers. Wave computing recently also acquired MIPS Technologies, trying to combine MIPS ' multiple threads (multithreading) and instant functionality, as well as wave computing its own streaming hardware, But the spokesman has yet to confirm what the new plan will look. The MIPS infrastructure supports billions of devices worldwide and also supports a wide range of operating systems and tools.
