Microsoft plans to run parts of Copilot locally on the PC. That’s what Intel executives say. The problem, however, is that much hardware does not have enough computing power to make this possible.
Much of the computing for Copilot is now done in the cloud. If this can be done partly locally, it will reduce latency and the Copilot will provide better performance and privacy benefits, because data no longer needs to be sent to the cloud. Those are exactly Microsoft’s plans, Tom’s Hardware notes based on statements from Intel executives. They talked about this during the company’s AI Summit in Taipei.
However, hardware is still an important limitation. The intention is that there is sufficient computing power to run parts of Copilot locally, without users realizing this. Devices should therefore not become slower and the battery should not drain faster. Certainly the latter can be prevented by: neural processing unit, or npu, can be used instead of a GPU, Intel executives say. Microsoft would even insist on that.
TrendForce reported in January that so-called AI PCs with Copilot integration should have NPUs that have 40 tera operations per second, or tops. Intel executives have now confirmed this, but there are not many such NPUs. Intel indicates that their own Core Ultra chips now offer 10 to 15 tops and AMD offers NPUs with up to 16 tops. The only processor commonly used to run Windows that meets the requirements is the Qualcomm Snapdragon X Elite with 45 tops. Intel’s next generation of Core Ultra chips must also meet the requirements, according to the company.
Tags: Intel Microsoft run Copilot partly locally Computer News
