In 1965, Gordon Moore predicted an annual doubling of component density on an integrated circuit at minimum manufacturing costs. That prediction of ever-increasing computational ability held for an amazing 50 years. But clock speeds are no longer doubling every 18 months, chips do not have room for more cores, and it is becoming obvious that assumptions about computing performance always increasing no longer hold true, says a new report from Hewlett Packard.
Kirk Bresniker, Hewlett Packard Labs chief architect and HPE fellow, says “It was great while it lasted.” To some computer scientists, it encourages a new era of creativity. Bresniker and other panelists suggested that the end of Moore’s Law might be the best thing to happen in computing since the beginning of Moore’s Law.
Computational needs continue to increase, pointed out Natalia Vassilieva, HPE senior research manager of the Software and AI groups at Labs. We need to take a step back, and while we have some glide path, we need to think about what comes next.
Clock speeds aren’t really expanding much, and we can’t get more cores onto chips. Yet hardware designers need to avoid communication and memory bottlenecks. “Those hardware limitations are pushing computer scientists forward because innovation isn’t slowing down,” said HPE research scientist Cat Graves,
Bresniker continues, noting “Today, ‘just in time’ means real time. We don’t have the luxury of waiting for information to trickle into a data center. We have microseconds to analyze petabytes of data. And it all has to happen at that edge, close to the sensors and actuators.”
“It is obvious that AI is booming,” Vassilieva added. “Artificial intelligence is becoming a factor in our daily lives, from Facebook tagging our photos to using Siri. And it’s not exactly a news flash that the trend is to apply AI technologies to vertical markets and industries. But we can’t accomplish those AI goals using general-purpose computers,” Vassilieva said.
These computer scientists see the transition as an opportunity to respond to “the tyranny of the general-purpose computer and the monoculture of technology,” as Bresniker described it. The need to move past Moore’s Law assumptions may encourage computer scientists to invent biologically inspired devices, circuits, and architectures, and new programming models.