Intel highlights the Sierra Forest's expected performance metrics, projecting a 2.4 times improvement in performance per watt and a 2.7 times increase in performance per rack compared to its second-generation predecessors. This advancement suggests that the equivalent of 200 racks of second-gen Xeon CPUs could be matched by just 72 racks of Sierra Forest servers, marking a significant step forward in data center efficiency and sustainability.
Additionally, the forthcoming Granite Ridge processors will feature support for the MXFP4 data format. This new format, endorsed by the Open Compute Project and leading technology firms, is set to dramatically enhance computational efficiency. Specifically, Granite Ridge processors are expected to facilitate a reduction in next-token latency by up to 6.5 times relative to fourth-gen Xeon CPUs using FP16 precision. Furthermore, these processors will be capable of running complex 70 billion parameter Llama-2 models in just 86 milliseconds, showcasing their potential for advanced data processing applications.
While the release date for Granite Ridge remains unspecified, Intel assures that it will follow closely after Sierra Forest's debut. The company's emphasis on E-core architecture in Sierra Forest chips underlines a strategic focus on scalable, high-performance solutions for cloud-native and containerized environments, leveraging Intel's advanced semiconductor technologies to meet the growing demands of modern data centers.