Intel has unveiled additional details about its upcoming 5th Gen Xeon processors. Codenamed Emerald Rapids, these CPUs are architecturally identical to Sapphire Rapids but pack more cores and cache on the same platform. More specifically, the new Xeon Scalable chips will offer up to 448MB of L3 cache, thrice as much as their predecessors. The core counts are a smidge higher, four more than Sapphire on the flagship. The official slides claim notable gains in performance and performance-per-watt.
There are two areas where Emerald Rapids will shine: AI and HPC workloads. Intel is substituting AI tasks with Speech Recognition and HPC with LAMMPS. The next-gen Xeon processors are expected to be up to 1.4x faster than the Sapphire Rapids flagship in these two workloads. Media transcoding is also set to get a 20% generational uplift.
In addition to the CPU cores and cache, Emerald Rapids will upgrade its memory support to DDR5-5600 while increasing the Gen 5 PCIe lane count to 80.
Physically, Emerald Rapids is quite different from Sapphire. Unlike the latter, it only features two (denser) chiplets, using the extra space to squeeze in more SRAM. It leverages the same 7nm-class process technology but integrates several workload-specific optimizations to improve overall efficiency.
