Distributed AI part of unlocking 5G value
Some of the primary enhancements associated with 5G are major increases to cellular speed and similarly significant reductions to latency. This combination is seen as the enabler of applications like mobile augmented and virtual reality, as well as autonomous control of vehicles. But, in addition to 5G, those applications will also need a distribution of computing power throughout a network and artificial intelligence (AI) to make decision-making keep pace with network abilities and application needs.
Qualcomm, whose SoCs and modems power virtually all current and coming 5G Android devices, this week announced a new AI accelerator chip, the Cloud AI 100, during an event in San Francisco. The new product, built using a 7-nanometer process and expected to sample in the second-half of the year, is optimized for data center use.
Qualcomm President Cristiano Amon, speaking at the event, described how the company’s research and development work around on-device AI will help it extend reach into new markets. “When we think of the 5G era and we combine it with AI, we’re going to be applying our expertise to a much larger set of opportunities beyond the smartphone.”
Qualcomm called out potential applications, including natural language processing and translation for personal assistants, advanced image searches and more personalized, relevant recommendations. For developers, the new chip supports AI software frameworks PyTorch, Glow, TensorFlow, Keras and ONNX.
During the event, leaders from Microsoft and Facebook discussed the role of AI inferencing in their respective operations.
On the mobile SoC side, Qualcomm’s most recent premium-tier Snapdragon 855 came with its fourth generation AI engine. According to the company, the new AI accelerator represents a more than 50x increase in peak performance, more than 350 TOPS.