Xeon Platinum 8580 processor using Emerald Rapids architecture leaked

At this year’s Innovation Conference, Intel announced that the fifth-generation Xeon Scalable processor, codenamed “Emerald Rapids”, is set to launch on December 14th, maintaining compatibility with the current Eagle Stream platform. Crafted using the Intel 7 process and featuring the Raptor Cove architectural core, it is projected to achieve an IPC improvement ranging from 5% to 10%. Its core count has been expanded to 64, supplemented with 320MB of L3 cache and 128MB of L2 cache.

Not long after the conference concluded, @YuuKi_AnS disclosed the specifications and performance test results of the Xeon Platinum 8580 from the Emerald Rapids lineup. This processor consists of two large chiplets, boasting 60 cores and 120 threads. It’s designed with a complete P-Core architecture, utilizing Raptor Cove cores, each endowed with a 2MB L2 cache. This culminates in an aggregate of 120MB L2 cache. Its shared L3 cache capacity has witnessed a substantial increase, reaching an impressive 300MB, with the combined L2 and L3 caches amounting to 420MB.

When juxtaposed with the existing Sapphire Rapids, the cache capacity has surged by 2.6 times. Although it still cannot parallel AMD’s EYPC Genoa, with its 480MB L3, or the 1.5GB cache-equipped Genoa-X, for Intel, this increment marks a significant advancement. Intel is also delving into the design of 3D stacked caches, which promises to narrow the gap with AMD competitors in terms of cache capacity.

The leaked Xeon Platinum 8580 is an A0-stepping ES2 sample. Its processor frequency is notably divergent from the ultimate retail version, operating at only 2.0GHz in a dual-socket server configuration. The testing platform has 120 cores and 240 threads, utilizing the existing Eagle Stream platform. Since both the Sapphire Rapids and Emerald Rapids employ the LGA 4677 socket, they both support 8-channel DDR5 memory. Owing to the engineering sample’s lower frequency and suboptimal BIOS optimization, current test scores offer limited value for benchmarking.

Compared to its predecessor, the fifth-generation Xeon Scalable processor, Emerald Rapids, has witnessed a threefold surge in L3 cache, supports an elevated DDR5 memory frequency, and boasts an increased maximum core count of 64. It embraces the CXL Type 3 high-bandwidth interface, optimizes power modes, and has optimized workload performance. There’s a 17% enhancement in performance per watt. Leveraging the Intel accelerator engine, CPU cores can be deactivated to augment power efficiency. There’s a marked elevation in AI inference and training capabilities, and it incorporates AI-accelerated Intel AMX.

As illustrated in the chart above, the product list for the fifth-generation Xeon Scalable processor is extensive. Emerald Rapids represents the culmination of the processors for the current Eagle Stream platform. The forthcoming generation will transition to the Birch Stream platform, soon introducing the Sierra Forest with a staggering 288 E-Cores, alongside the new generation P-Core Xeon processor, Granite Rapids.