Source: gate Arab news technical
Is the process of making things smarter through connecting and integrating sensors and building software artificial intelligence inside which for most organizations the biggest challenge in the next five years, where will everything becomes more intelligent in theory, but restrictions on the chip the current status of the work to slow down the process, the technology today simply is not on a professional level, which included his company Graphcore – ups that take from the Bristol headquarters, responsible for the development of a new chip to help speed up the process of adoption of artificial intelligence.
Some neural networks are good enough for using cloud computing massive data sets, however, the artificial intelligence systems the most strength struggling during development in order to address the outstanding accounts quick fast when you use the processing units the current state that operate sequentially, in other words, to reduce the response time.
Adds Nigel Toon Nigel Toon, co-founder of Graphcore: “70 years ago, we have programmed computers to work according to the instructions step-by-step, except that artificial intelligence involves learning computers and with data that oppose it, is considered modern and simple enough to understand, and can be handled through existing technology At the current time, only to understand the language of the entire context and the words more difficult, and requires systems of data storage and memory to understand the background conversations, relieve the things required to verify the data significantly from the traditional process, it’s a completely different type of workload”.
Are considered interim solutions, including the development of CPU in the cloud to see the amount of work that you should be doing and the use of GPUs, not fast enough for the world of Artificial Intelligence the rapid development, and many companies such as Google, Amazon and Apple TV devices to solve this problem, resulting in a huge torrent of unprecedented capital in emerging technology companies.
It was Nigel Toon has launched in 2012 the company’s Asia semiconductor Icera, in collaboration with the company’s co-founder Simon Knowles Simon Knowles, which was sold in 2011 to a manufacturing company chip NVIDIA Nvidia versus $ 435 million, has helped the project in understanding the constraints related to hardware faced by artificial intelligence.
Continues to Nigel Toon and Simon Knowles through 2016 with researchers to identify the problems they face and their plans for the future, and decided to work from first principles and think less in regards to the code and focus more on the incident itself, which requires a new solution of their own to build a completely new type of processors, think of the workload borne by a computer in a different way.
The CPUs usually solve problems by assembling blocks of data, and then run the algorithms or logical operations on that information in a row, and owns the chip has a quad-core modern four processors available, it also features graphics processing units designed for gaming on the processors in parallel can execute multiple tasks at the same time.
And computers in the presence of artificial intelligence systems to pull massive amounts of data in parallel from different locations, and then processed quickly, and this process is known as dialogue, which focuses on real companies rather than the instructions, focus chip Graphcore New, a Processing Unit intelligent IPU, for computing graphical parallel computing floating point low precision.
Owns this chip more than 1000 processor communicate with each other to share the burdens of knowledge work required for automated, where he says Nigel Toon: “the structure of hardware is simple and straightforward, and you can simply reach the stage of devices and then try to figure out how to write the program from it”, the difference is in how the individual processors on the chip with each other and external memory, which is operated through a programme of Poplar with Graphcore.
The programme of Poplar to move data across the chip more efficiently, which means less processing power wasted, as it does so in a timely manner, using all processors serially, so that is the performance improvements are quite significant, as can buy a Graphcore processing artificial intelligence algorithms developed at up to ten times compared to most processors and GPUs available today.
And my company Graphcore that architecture deal with data and processing will become more effective by 100 times of the most powerful Processing Unit graphics, leading to open opportunities to new applications and growth.
It seems that these chips get a lot of attention, as the investment arm to buy Sequoia capital, Sequoia Capital, which focuses on investment within the technology industry and has previously invested in companies such as Google and Apple, to provide $ 50 million for the project to assist in the growth, and with access to the next generation of silicon technology and the continuation of Moore’s law, we can get more transistors in a smaller space, which makes us see people achieve new breakthroughs thanks to artificial intelligence.
Link to it from the source: the race to manufacture chips of artificial intelligence mounts