In April 2018, Jim Keller joined the Intel team to lead the development of processor architectures. In a recent interview, Jim admitted that he would like Intel to develop a completely new architecture every five years, although at present this occurs once a decade.
Intel has already released Ice Lake processors with the Sunny Cove microarchitecture. By the end of the year, Tiger Lake processors featuring the Willow Cove microarchitecture and significantly overhauled cache structure will be available, but it's unlikely that Jim Keller's influence will be felt before the release of the Golden Cove microarchitecture, scheduled for 2021 or later. Industry experts believe that a radically new architecture will not be created by Keller and his team until at least 2023 — essentially, five years after Jim joined Intel.
Last week, the host of the channel managed to conduct a one-and-a-half-hour interview with Jim Keller, and somewhere around the twenty-ninth minute, the discussion turned to the necessary conditions for Intel's success in creating new architectures. Keller stated that in today's reality, a new architecture needs to be presented every three years, but ideally, it should be built from scratch every five years. Currently, such radical updates occur once every ten years, and Jim is ready to put in the effort to change the existing state of affairs.
These changes will come at an enormous cost. There will be resistance from marketing structures that focus on achieving short-term results, but are not very willing to sacrifice them for the sake of a long-term goal. According to Keller, in many cases it is better to make a 'long-term bet,' sacrificing something in the short term; the history of many companies confirms that this is justified. Developing sequentially released architectures by two parallel teams of specialists in a 'checkerboard' pattern allows for smoothing out the 'gaps' between the periods of radically updated products.
Source: 3dnews.ru
