发帖
雷达卡
[size=1.1em]
Moore’s law describes a phenomenon that has arguably influenced our modern world more than any other. The exponential growth of computing performance over the past 40 years has reshaped our lives in myriad ways, empowering all manner of wonders from the world wide web to smartphones to the Internet of Things.
But Moore’s law has been fading fast, if it has not already expired. The big, as yet unanswered, question is to what extent this matters.
In an article in 1965, Gordon Moore, the future founder of Intel, explained how the number of components that could be crammed on to an integrated circuit doubled every year. In 1975, he revised this to two years and his forecast, subsequently adopted as a “law”, has pretty much held true ever since.
The impact of exponential change is often hard to grasp. In practical terms, the increase in computing performance has enabled chips to become smaller, faster and cheaper to an astonishing degree. Whereas the Apollo guidance computer that steered Neil Armstrong to the moon contained 12,300 transistors, Apple’s iPhone 7 boasts 3.3bn.
But physics is exacting its revenge, as it always tends to do. Now that manufacturers are fiddling around at sizes of seven nanometres we would appear to have reached the physical limits of how many can be reliably crammed on to an integrated circuit. The easy ride that chip designers have enjoyed over the past decades has come to an end.
That would appear to be a problem for the computing industry — and the rest of us — as future performance improvements will involve far greater cost and complexity. But chip designers seem to be relishing the challenge. Some are even forecasting that it will open an exciting new era of innovation.
Mike Muller, chief technology officer at Arm, the chip designers, says that for the foreseeable future, computer users are unlikely to notice much difference as their products continue to improve. “Some people are saying that Moore’s law is slowing, Moore’s law is dead. But from Arm’s perspective, we do not care,” he says. “The end products are still going to get better and better even if our jobs are going to get harder and harder.”
By necessity, chip designers are having to become ever more creative. They can “cheat” by improving infrastructure architecture or stacking chips on top of each other to create a 3D effect. They are also manufacturing chips for specialist purposes, such as artificial intelligence.
For example, John Hennessy, chairman of Alphabet, says that Google’s Tensor processing unit, a hardware chip designed for machine learning, can conduct 92tn operations a second, about 100 times the throughput of a general purpose processor. “That is the way we are going to make advances at a time when Moore’s law will not deliver as easily as it has in the past,” he told the Zeitgeist Minds conference last month.
Broader changes in the industry are also expanding the ways in which users can exploit powerful computing. Jason Zander, executive vice-president of Microsoft Azure, says that computing is increasingly moving up to the intelligent cloud and out to the intelligent edge. It is now easy to access vast computing power in the cloud at reasonable cost. But it is also increasingly possible to process data locally at the edge on all types of connected devices. “We are at a magical pivot point in computing,” he says.
Still, the race is on to develop radical new ways of computing that will take us beyond the limitations of the silicon age. The most intriguing is quantum computing which promises a big increase in performance, even if it remains diabolically difficult to develop.
提升卡
置顶卡
沉默卡
变色卡
抢沙发
千斤顶
显身卡
京公网安备 11010802022788号
