IBM debuts power-efficient NorthPole machine learning processor
IBM Corp. today detailed NorthPole, an internally designed artificial intelligence chip that it says is significantly more power-efficient than competing processors.
NorthPole was developed over the course of eight years by staffers at the company’s research lab in Almaden, California. The chip builds on an earlier IBM research project that produced an AI-optimized processor called TrueNorth. According to IBM, NorthPole is about 4,000 times faster.
The newly debuted chip consists of 22 billion transistors made using a 12-nanometer process. According to IBM, those transistors are organized into 256 cores that can each perform 2,048 calculations per clock cycle with eight-bit precision. This means the calculations are carried out on relatively small numbers that take up to eight bits of space.
Each of NorthPole’s 256 cores includes an integrated memory pool that stores the information used in computations. Because the chip’s memory circuits and cores are near each other, data can travel between in a short amount of time. That reduces the processing delays associated with data movement and thereby improves performance.
“Architecturally, NorthPole blurs the boundary between compute and memory,” said Dharmendra Modha, IBM’s chief scientist for brain-inspired computing. “At the level of individual cores, NorthPole appears as memory-near-compute and from outside the chip, at the level of input-output, it appears as an active memory.”
Another benefit of the memory circuits built into NorthPole’s cores is that they increase energy efficiency. Moving data between different parts of a chip not only delays processing but also consumes electricity. By minimizing the distance that data must travel between memory circuits and cores, NorthPole’s architecture saves energy.
IBM measured the chip’s power consumption by having it run an open-source image recognition model called ResNet-50. The model is commonly used to compare different AI processors’ capabilities. According to IBM, NorthPole ran ResNet-50 with 25 times higher power-efficiency than graphics cards and central processing units based on 12- and 14-nanometer nodes, respectively.
NorthPole’s power-efficiency is constrained by the fact that it’s also made using a 12-nanometer node, which is several generations behind the latest three-nanometer process available to chipmakers. However, the processor can be upgraded to newer manufacturing technologies. IBM researchers reportedly estimate that such an upgrade could allow NorthPole to provide 25 times higher energy efficiency than rival chips based on “current designs.”
A processor’s power consumption directly influences the amount of heat it produces. IBM says that NorthPole’s energy efficiency limits its maximum temperature, which means the chip requires only simple external cooling equipment to run. As a result, it could potentially be deployed in space-constrained systems such as autonomous cars that have limited room for cooling gear.
A message from John Furrier, co-founder of SiliconANGLE:
Your vote of support is important to us and it helps us keep the content FREE.
One-click below supports our mission to provide free, deep and relevant content.
Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger and many more luminaries and experts.