Emerald Rapids Die-Shot Analysis
Emerald Rapids is made using the Intel 7 process, the forerunner of Intel 3 planned for Granite Rapids. These CPUs host Raptor Cove cores and can offer up to 64 cores per package. A key difference is the tile structure. Emerald Rapids has 2 tiles, as compared to 4 on Sapphire Rapids. As for the CPU layout, we see 2 core and cache arrays connected via a Modular Die Fabric. There are a total of 4 memory controllers, which offer support for 8-channel DDR5-5600 memory. Furthermore, 6 PCIe controllers are split between the two dies alongside 4 UPI channels and 4 accelerator engines. The cache structure sees a dramatic uplift, especially in the L3 department. Intel has gone from a max of 112.5MB to a whopping 448MB of L3 cache. Additionally, there is 2MB of L2 cache reserved for each core.
Performance Metrics
To strike a clear difference between the 4th Gen Xeon and 5th Gen Xeon offerings, Intel has compared the flagships from both lineups. The 8480+ is based on Sapphire Rapids with a total of 56 cores, whereas the 8592+ eclipses that with 64 cores, and much more L3 cache. Across a wide variety of benchmarks, Intel does promise a significant leap in performance. The same was the case from Alder Lake to Raptor Lake, where the extra cache did wonders for Intel. Though on a pure IPC-to-IPC basis, Raptor Cove is barely any faster than Golden Cove. Interested consumers will be looking forward to these processors later this month. It is noteworthy that the successor of Emerald Rapids, Granite Rapids is expected to deliver a huge uplift in efficiency and performance. Granite Rapids is scheduled for 2024 and will be built using the Intel 3 (3nm-class) process, Redwood Cove P-cores, and is expected to pack 88 cores for the SP variants and 132 cores for the AP variants. AMD is also keen on taking Intel’s market share next year with Turin, as both companies will go head-on in the server market. Source: Intel, InstLatX64
