Intel 5th Gen Emerald Rapids 64-Core & Granite Rapids Xeon CPUs Performance Unveiled

Intel Xeon W9-3595X "Emerald Rapids" HEDT CPU Spotted: 60 Cores, 120 Threads, 4.6 GHz & 112 MB L3 Cache 1

Intel has unveiled the latest performance and internal projections of its upcoming 5th Gen Emerald Rapids & next-gen Granite Rapids Xeon CPUs.

Talking about its Xeon CPUs during SC23, Intel unveiled new performance metrics which include actual results for its upcoming 5th Gen Emerald Rapids and projections for its next-gen Granite Rapids Xeon CPUs. The chipmaker compared the results against its own Sapphire Rapids 4th Gen chips while also showcasing HPC metrics for the Xeon Max CPUs against AMD's EPYC Genoa 96-core chips.

Starting with the details, the Intel 5th Gen Emerald Rapids CPUs will be pin-compatible with the Sapphire Rapids CPUs on the Eagle Stream platform. They utilize the same LGA-4677 socket and offer mainly an optimized design with few additional cores/threads, larger pools of cache, & focus on higher performance per watt.

The company highlights that Emerald Rapids Xeon CPUs will offer workload-optimized performance and energy-efficient compute. At the same time, the addition of faster DDR5-5600 memory should result in higher bandwidth and transfer speedups. The 5th Gen Xeons will also offer CXL 1.0 and 2.0 support along with up to 80 PCIe Gen 5.0 lanes. Intel touts a 40% speedup in performance. Details are mentioned below:

Intel Emerald Rapids (Xeon 8592+ 64-Core) vs Sapphire Rapids (Xeon 8480+ 56 Cores):

AI Speech Recognition: 1.4x Boost

HPC LAMMPS (Copper): 1.4x Boost

Media Transcode (FFMPEG): 1.2x Boost

Some of the features to expect within the 5th Gen Xeon "Emerald Rapids" CPUs include:

Up To 3x larger LLC cache

Increased Memory Speed

Higher CPU Core Count (Up To 64 Cores)

CXL Type 3 high bandwidth interface

Workload optimizations (Intel AE)

Optimized Power Mode

Up To 17% General Purpose Perf/Wat Increase

Offload CPU cores with Intel Accelerator Engines for Power Efficiency

Huge performance boost across inference and training

Intel AMX for built-in AI Acceleration

Out-of-box deployed with Optimized SW stacks

Intel is also stating the internal performance projections of its upcoming Granite Rapids Xeon CPUs which will come in P-Core-only flavors while Sierra Forest will adopt the E-Core architecture & offer up to 288 cores. These chips will offer compatibility with the next-gen Birch Stream platform which comes in two distinct sockets, LGA 4710 and LGA 7529. The company is confirming that the Granite Rapids Xeon CPUs will offer increased core counts, frequencies, and the latest Intel AMX (Advanced Matrix Extensions). Granite Rapids will also add FP16 to broaden the precision support for AI-based developers. The platform will offer 12-channel MCRDIMM support to tackle larger LLM models that are memory-bound.

In terms of performance, the internal estimates against existing 4th Gen Sapphire Rapids CPUs show that Granite Rapids Xeon chips will offer a 2.9x boost in AI Inferencing (DeepMD+LAAMPS), 2.8x boost in memory bandwidth and up to 3x better performance in AI workloads. These are just rough estimates and final performance may differ.

Lastly, Intel also showcases some HPC performance benchmarks of its Xeon Max 9480 CPU which packs 56 cores and 64 GB HBM memory against the AMD EPYC 9654 with 96 cores. The Xeon Max CPU is shown to offer up to 30% better performance. The EPYC 9645 retails for $11,805 while the Xeon Max 9480 costs $12,980 US.

Both CPUs are rated at 360/350W but the Sapphire Rapids CPUs do consume higher wattage so that might lead to lower TCO than the AMD solution. Furthermore, AMD also offers its 3D V-Cache boosted offerings in Genoa-X flavors which can offer competitive performance in these bandwidth-bound workload scenarios.

With that said, Intel will be launching its 5th Gen Xeon Emerald Rapids CPU family on the 14th of December while the Granite Rapids CPUs will debut soon after Sierra Forest by 2nd half of 2024.