Recently, at a wonderful press conference, Intel officially announced the code name as"emeraldrapids"of the 5th generation Xeon Scalable processors. Designed for servers and data centers, this processor delivers significant improvements over previous generation Xeon processors in terms of average performance, power efficiency, and AI inference capabilities. The newly released fifth-generation Xeon Scalable processors are configured with up to 64 cores, improving overall performance by 21%, AI inference performance by 42%, and reducing total cost of ownership by 77%.
The new processor continues the manufacturing process, architecture, and package interfaces of the previous generation, with upgrades and enhancements in layout design and core specifications. First of all, in terms of layout design, the fifth-generation Xeon uses 2 chips instead of the 4-chip design of the previous generation. The benefit of this is that the number of modules is reduced, which reduces power consumption. Although the area of the chip has become larger, Intel's manufacturing process has matured and can cope with this challenge well. Each chip is connected by a modular die bond located between the core and cache array. Each chip has 35 cores, 2 memory controllers, 3 PCIe controllers, 2 UPIs, and 2 accelerator engines. For processors with a smaller number of cores, Intel uses a single-chip structure. Compared to the previous generation of processors, the number of cores of the 5th generation Xeon has been increased to 64 cores and a larger cache pool has been provided, increasing the cache capacity to 5MB per core. In addition, the new processor supports faster DDR5-5600 memory speeds and PCIe 50 interface.
According to Intel, fifth-generation Xeon processors deliver a 21% increase in average performance, a 36% increase in energy efficiency, and a 77% reduction in total cost of ownership over the previous generation. In terms of AI capabilities, the new processor has a built-in AMXAI accelerator and further boosts the Turbo frequency, enabling performance in certain AI inference workloads to reach 1 of the previous generation42 times. It is also the only general-purpose CPU that has passed MLPERF training and inference benchmark performance tests. According to the test data, the fifth-generation Xeon has achieved significant performance improvements in workloads such as AI speech recognition, threat detection, and transcoding. In addition, it is capable of running large language models with good memory bandwidth and low latency performance. The processor also supports features such as the QAT Quick Assist, DLB Dynamic Load Balancer, DSA Data Flow Accelerator, and IAA In-Memory Analysis Accelerator, which can be enabled on an on-demand basis.
For customer application examples, the 5th generation Xeon processors are available at IBM Watson XThe network query throughput of the data platform has been increased by 27x, 2x better threat detection performance on PaloAltoNetworks' deep learning Xi model, and up to 6x better inference performance on GalliumStudios' NumentaAI platform than GPU cloud instances5 times.
The 5th Gen Xeon processors are available in several models, including the XeonPlatinum 8592 series, the XeonGold series, and the XeonSilver series. The flagship XEONPLATINUM8592 series is available in three versions,** ranging from approximately $11,000 to $12,000. The processor's core count, cache capacity, and power consumption range have all increased, as have the base and boost frequencies.
In addition, Intel also revealed the next two processors, which are SierraForest based on the pure E-core architecture and GraniteRapids based on the pure P-core architecture. Sierraforest will launch in the first half of next year with 288 cores and threads. Graniterapids will be launched in the second half of 2024, and the specific specifications have not yet been disclosed.
In conclusion, Intel's fifth-generation Xeon Scalable processors stood out at the conference for their superior performance, power efficiency, and AI inference capabilities. It will provide servers and data centers with more computing and processing power, driving the development of AI technology.