Intel Granite Rapids wafer manufactured in Intel 3 leaks

Intel unveiled its fourth-generation Xeon Scalable Processor, codenamed Sapphire Rapids, at the beginning of last year, and later in the year, introduced the Emerald Rapids, also part of the Eagle Stream platform, marking the fifth generation of Xeon Scalable Processors. However, the true transformative leap is anticipated with this year’s imminent Granite Rapids, belonging to the new Mountain Stream (or Birch Stream) platform.

Recently, Andreas Schilling shared wafer images of Granite Rapids, marking Intel’s first commercial chip to utilize the Intel 3 process.

The wafer features square chips, each housing 30 cores. Two of these can be combined to form a “Granite Rapids-XCC” processor, boasting a configuration of 56 cores and 112 threads (with 2 cores disabled per chip). Intel plans to integrate several dedicated function accelerators onto the chip to expedite popular server workloads, thereby narrowing the performance gap with AMD’s EPYC server processors.

Rumored to utilize the Redwood Cove core architecture, Granite Rapids is poised to be Intel’s first to embrace AVX10 and APX technologies, alongside an increase in core and thread counts to enhance performance metrics. It is expected to operate at a base frequency of 2.5 GHz, supporting 96 PCIe 5.0 lanes and 12-channel DDR5-6400 memory. Intel has disclosed that Granite Rapids will incorporate multiple small chips within a single SoC, interconnected via EMIB packaging, and will also feature HBM and Rambo Cache chips.

Previous reports, based on the latest entries in Intel SDE 9.33.0, suggest that Granite Rapids will see a substantial increase in L3 cache capacity compared to Emerald Rapids, rising from 320MB to 480MB, a 1.5-fold increase from its predecessor. This expansion in the L3 cache is significantly beneficial for artificial intelligence inference, data center operations, video encoding, and general computational workloads.