At Ignite Developers’ Conference, Microsoft revealed its chips that are designed to execute AI computing tasks specifically. Other than Microsoft other companies like Qualcomm and MediaTek have developed devices with generative AI capabilities with their upcoming chipsets for flagship phones and mid-range phones.
So they have also entered the market of manufacturing AI chipsets. This shows a departure from traditional CPUs and a shift towards specialized processing units that are optimized to work with AI models.
Recent development by MediaTek and Qualcomm have shown that the market is evolving towards developing chipsets that integrate on-device generative AI functions. So specialized processing units are being developed that can work along with generative AI. This shows the broader industry progression towards integrating AI with conventional processing units.
But how are the AI chips different from traditional processing units?
AI chips- AI chips are specialized category of semiconductors developed to work easily with on-device AI capabilities that can execute Large Language Models or LLMs. It mostly adopt a SoC ‘system-on-chip’ configuration that has different functions beyond CPU’s responsibility for computing and general processing.
CPUs and AI chips both offers heightened computations but AI chips incorporates design features that are optimized for specific processing methodologies. While CPUs apply sequential computing method by issuing one instruction in succession after another after waiting for the completion of the command.
AI chips in contrast work simultaneously with computing processes to complete several commands at a time. By functioning parallel with computer processing, it can do both training and inference tasks. This advantage from traditional CPUs helps AI chips gain speed and efficiency in its function. It also enhances inference and training of AI algorithms. This the reason for the trend towards developing on-device generative AI capabilities in major companies like Microsoft, MediaTeK and Qualcomm.