(Image courtesy of Fritzchens Fritz).

Apple's Latest Silicon Doubles Down on Machine Learning

Sept. 17, 2018
Apple's Latest Silicon Doubles Down on Machine Learning

Last week, Apple announced that its latest line of smartphones would contain a 6.9 billion-transistor chip based on 7-nanometer manufacturing technology, the most advanced process available. But the performance of the new processor, the A12, highlights the company’s increasing focus on neural-network and graphics-computing as the advantages of scorching smaller and smaller transistors onto silicon wafers wind down.

The 64-bit central processing unit (CPU) includes two high-performance cores that handle immediate, intensive workloads 15 percent faster and with 40 percent lower power than its 10-nanometer A11 predecessor. The four high-efficiency cores consume 50 percent less power for sustained background applications. The 4.3 billion-transistor A11 perked up performance by 25 percent and efficiency by 70 percent. This could signal slowing improvements process node to process node.

Cupertino, California-based Apple is trying to take up the slack with improvements in neural networking and graphics computing. The custom graphics processor (GPU) that Apple developed boosts performance by around 50 percent, which improves the realism of reflected light and other effects in video games. Last year, Apple announced that it would discontinue the use of graphics from Imagination Technologies in its smartphones, tablets, and other gadgets.

Apple's new neural network processing unit can complete five trillion operations per second using Apple's CoreML machine-learning framework, up from 600 billion the A11 is capable of. The silicon also supports variable precision to lower the power consumption of on-device inference workloads without having to consult the cloud. The eight-core processor runs machine learning with a tenth of the energy of the previous two-core model, helping extend battery life.

“The A12 Bionic signals two things, one a bigger investment in neural-net and graphics performance and lower investment in CPU,” Patrick Moorhead, Moor Insights and Strategy’s principal analyst, told Electronic Design . “Apple already leads CPU performance and most of the next-generation applications”—including augmented reality and facial-recognition software that unlocks the device and unseals passwords—"are driven by the neural processing unit and graphics processing unit."

Building chips and outsourcing production to Taiwan Semiconductor Manufacturing Corporation—which also produces the 7-nanometer chips inside Huawei smartphones to be released next month—allows the trillion-dollar company to vastly improve the efficiency of its software and hardware. That also helps cut manufacturing costs. The A12 was produced by Apple’s chip division, led over the last decade by senior vice president of hardware technologies, Johny Srouji.

Sponsored Recommendations

Comments

To join the conversation, and become an exclusive member of Electronic Design, create an account today!