Intel may equip Granite Rapids with 480MB L3 cache

Intel unveiled its fourth-generation Xeon Scalable processor, codenamed Sapphire Rapids, at the beginning of last year, with plans to launch the Emerald Rapids, the fifth-generation Xeon Scalable processor and a member of the Eagle Stream platform, by the end of the year. However, the real transformative leap is anticipated with the imminent arrival of Granite Rapids and the new Mountain Stream (or Birch Stream) platform, which will also accommodate Sierra Forest, featuring all-efficient cores on the same platform.

As reported by Tom’s Hardware, the latest entries in Intel SDE 9.33.0 reveal a significant increase in the L3 cache capacity of Granite Rapids, surging from 320MB to 480MB, a 1.5-fold augmentation over its predecessor. This substantial enhancement in the L3 cache is poised to considerably benefit artificial intelligence inference, data center operations, video encoding, and general computational workloads.

Over the past few years, AMD’s EPYC server processors have maintained a stronghold in the data center market. Intel, aspiring to remain competitive within this domain, believes that augmenting the L3 cache is a straightforward strategy for directly boosting performance. It remains challenging to ascertain the extent of Intel’s success with this approach, especially given AMD’s effective strategy of combining multicore quantities with large caches, a highly favored tactic in the data center market. Intel’s endeavor to catch up swiftly appears daunting.

Rumors suggest Granite Rapids will be fabricated using Intel’s 3-process technology, incorporate Redwood Cove architecture cores, and feature increased core and thread counts. Its base frequency is expected to be 2.5 GHz, supporting 96 PCIe 5.0 lanes and 12-channel DDR5-6400 memory. Intel has hinted that Granite Rapids will comprise multiple chiplets within a single SoC, employing EMIB packaging, and will include both HBM and Rambo cache chips.