Ai Wave Wave Computing Is Expected To Shake Up Its Position.

Wave Computing, the Silicon Valley innovation company of the United States, specializes in the development of machine learning chips, which can be expected to shake NVIDIA's leading position in the field of machine learning chips in accordance with the location and use of the data. The company was founded in 2009 and has raised $more than 100 million.

According to Electronic Design, Wave Computing executive Derek Meyer said the time of their chip to perform the task is hundreds of times faster than a drawing chip (GPU), and can comb millions of images or hours of speech to specialize in the training and inference of AI algorithms.

Wave Computing used to develop general chips before, but in 2013, the business model had a big turn, after all, machine learning software began to grow, and more start-ups such as Graphcore and Cerebras Systems joined the war, and now Wave Computing does not sell generic chips directly.

Meyer says they will deploy different systems and programs according to the location and purpose of the data. At present, most of the new chips on the market belong to the accelerator and contain thousands of cores, which can carry out machine learning tasks faster and efficiently, but the NVIDIA GPU and other accelerators still depend on instructions from individual processors, not only power consumption, but also the problem of delay.

The Wave Computing processor is between the field programmable logic gate array (FPGA) and the special application integrated circuit (ASIC). Meyer says that in the AI world, software changes almost every day, and chips must adapt to continuous changing algorithms.

However, the question is whether customers are willing to transfer technology from NVIDIA GPU to Wave Computing. NVIDIA is dedicated to expanding the use of GPU and supporting every major machine learning environment, including TensorFlow and Caffe, but Wave Computing currently only supports TensorFlow. Wave Computing still needs to work hard to educate customers. In 2017, it introduced the infrastructure of Wave Computing for potential customers.

Wave Computing recently acquired MIPS Technologies, trying to combine MIPS's multiple threads (multithreading) and instant functionality, as well as the Wave Computing's own data stream hardware, but the spokesman still did not confirm what the new scheme would be.

The MIPS infrastructure supports billions of devices around the world, and supports a wide variety of operating systems and tools. Karl Freund of Moor Insights and Strategy says Wave Computing's acquisition of MIPS is a bold move and is expected to accelerate profits and carry out industrial cards.