And China will be more and more a place. "On the basis of 10 nanometers to 7 nanometers, the core competencies of the AI chip will come from clouds, large data, servers, and so on," said Wu Gengyuan, director of vertical integration and strategy marketing at SMIC. In addition to the British Weida, the "Big Four", Apple, Microsoft, Amazon, Google will be involved. "
No one wants to miss the next technological breakthrough. Silicon Valley famous investor MarcAndreessen said the software is devouring the world. While artificial intelligence is devouring software. NVIDIA's data show that over the past year, the number of GPU-driven AI services has increased more than four times to 1,300.
NVIDIA has launched a new generation of GPU, named Volta. The Volta is the largest technology leap in this era since the invention of the CUDA, a parallel parallel computing architecture. It integrates 21 billion transistors, using 12-nanometer optimized TSMC process and Samsung's fastest HBM memory, you can perform high-speed matrix operations, the depth of element-level learning.
According to reports, each VoltaGPU support 120 trillion floating-point operations. NVIDIA's DGX-1AI supercomputer connects eight TeslaV100GPUs to achieve near-thousands of floating-point operations.
Not long ago, Google also introduced its second-generation TPU chip at its I / O Developer Conference, which provides 128 trillion floating-point operations. In contrast, iPhone6 can provide 100 trillion floating-point operations. TPU is Google's own research and development for a AI computing services for high-performance processors.
Last year, Google has released the first generation of TPU chips, mainly in the Google data center and including AlphaGo such artificial intelligence which use, the main operational performance. Google said that the second generation of TPU machine learning training speed than the market now on the graphics chip (GPU) save half the time.
So far, Google is still buying Intel and NVIDIA processors. But as Google is increasingly dependent Silver Wing on the independent development of the chip architecture, the future or will save billions of dollars of their own chip spending scale. This also makes Intel, NVIDIA and other industry giants are worried.
According to the latest news from Bloomberg, Apple is also developing an AI chip named "AppleNeuralEngine" (AppleNeuralEngine), which may be used in smart cars, Siri voice systems and future enhancements in real-world AR, and one day Is used in the iPhone, iPad, to help improve the performance of the software. Apple will announce more AI-related plans at the WWDC conference.
