Intel, the world’s largest chipmaker, has announced a new data center chip that can handle more than double the amount of computing work per watt of power used, compared to its current generation of data center chip. The chip, codenamed ‘Sierra Forest’, is expected to arrive next year and will be fabricated using Intel’s 3-nanometer process technology.
The new chip is part of Intel’s strategy to regain its leadership in the data center market, where it has faced fierce competition from rivals such as AMD and Ampere Computing. Intel said its ‘Sierra Forest’ chip will have 240% better performance per watt than its current data center chip, the first time the company has disclosed such figures. This means that the chip can deliver more computing power while consuming less energy, which is crucial for data centers that power the internet and online services.
A mini Sun on a chip
Intel’s ‘Sierra Forest’ chip is also remarkable for its high core count and temperature. The chip will have 144 cores, which is more than AMD’s upcoming 128-core EPYC Bergamo chip. The cores will be based on Intel’s E-core architecture, which is designed for high energy efficiency, core density, and total throughput. The E-core architecture is different from Intel’s P-core architecture, which is focused on high performance and AI workloads.
The ‘Sierra Forest’ chip will also achieve temperatures nearly seven times hotter than the core of the Sun, which has a temperature of 15 million degrees kelvins. The chip will run at 100 million degrees Celsius, which is the same temperature as the nuclear fusion reaction that scientists in South Korea have recently sustained for 30 seconds. Intel said that this high temperature will enable the chip to perform complex calculations faster and more efficiently.
A glimpse of the future
Intel also revealed some details about its next-generation data center chip, codenamed ‘Clearwater Forest’, which will debut in 2025. The chip will be fabricated using Intel’s 18-nanometer process node, which is an improvement over its 20-nanometer node that was originally planned for the chip. The chip will also feature a new architecture that will improve its performance and scalability.
Intel said that its data center roadmap is on track and that it has solved the underlying issues in its process node technology and chip design methodology that caused delays to its previous products. The company also demonstrated some of its current and upcoming products, such as Sapphire Rapids and Granite Rapids, which are based on the P-core architecture. Intel showed that its chips can outperform AMD’s chips in AI and memory bandwidth benchmarks.
Intel’s announcements show that the company is not giving up on its ambition to dominate the data center market and to innovate in the field of semiconductor technology. The company said that it is delivering five new nodes in four years, an unprecedented pace that will power its broad data center and AI portfolio that includes CPUs, GPUs, FPGAs, and Gaudi AI accelerators.