Intel Restructures Data Center Portfolio With Xeon 6 Plus 18A Processors Plus 800 Series Ethernet Connectivity and Crescent Island AI Accelerators
The Intel Corporation has announced a comprehensive restructuring of its data center portfolio to align with the rapidly changing landscape of artificial intelligence. Key elements of this restructuring include the unveiling of the new Intel Xeon 6 Plus processor family, enhancements to the 800 Series Ethernet portfolio with the Intel Ethernet E835 controller, and developmental updates regarding the future Crescent Island accelerator. The announcement highlights an emerging trend within the industry where the CPU is once again becoming the dominant control plane for managing orchestration, data movement and simultaneous execution within massive AI configurations.
Kevork Kechichian, the executive vice president and general manager of the Intel Data Center Group, noted that:
"AI scales as a system rather than a group of parts."
He added that, "The shift in workloads toward more agent like and autonomous performance means the bottlenecks have now shifted to the network, data movement and concurrent processing." To underscore his philosophy, Kechichian stated:
"AI does not scale as a collection of parts, it scales as a coordinated system."
The new Intel Xeon 6 Plus processor family is the first Data Center CPU that utilizes the 18A process technology from Intel. The system has been designed with optimal power efficiency to maintain the best possible density. The flag ship models now come with up to 288 Efficient cores. Each of these core sets delivers up to 2.5 times greater performance than the earlier generations of processors and up to 45% better performance per thread per watt. This high density is intended for orchestration intensive workloads typically seen in cloud native and telecommunications environments.
Data bottlenecks are avoided thanks to the use of 12 channels of DDR5 memory to maximize memory throughput. This system offers 96 lanes of both PCIe Gen 5 and CXL to facilitate accelerated data transfer between different components within a server. To provide even greater control, the CPU is also equipped with Application Energy Telemetry, a tool to monitor real time CPU power usage at a per workload level. Such efficiency allows for up to a 9 1 server consolidation ratio to be achieved when replacing an older generation 2nd generation Intel Xeon. Silicon security such as SGX and TDX are available to ensure the Confidentiality of Multi tenant Cloud deployments.
Data Center throughput has increasingly become dictated by the efficiency of network components in the increasingly distributed environment. The new 800 Series Intel Ethernet E835 controllers and adapters offer efficient connectivity to enterprise, cloud and edge deployments. The adapter can deliver up to 200 GbE throughput with options of 2x25 GbE, 4x25 GbE, 2x100 GbE and 1x200 GbE via the use of the Intel Ethernet Port Configuration tool.
Benchmark results have demonstrated that the Intel Ethernet E835 CQDA2 adapter can offer up to 1.9x more performance watt than a NVIDIA ConnectX 6 DX and up to 1.4x more performance watt than the Broadcom BCM957508 P2100G. Such efficiency can contribute greatly to the reduction of Opex in the virtualized environment. RoCEv2, iWARP and other RDMA protocols are available to help offload the CPU, along with Dynamic Device Personalization to aid with packet processing. Built in security include Hardware Root of Trust and signed SPDM with DMTF standard interfaces, and a 10 year lifecycle commitment to all product offerings.
Intel has also released new small to medium business SMB hardware options. The 12 core variant has been added to the Intel Xeon 6300 processor family. This means that a full 4 more cores than before are available with the addition of these CPUs, making the system ceiling for the product family rise to 12 cores. Importantly, these processors can be swapped into any existing motherboards which are designed to use an 8 core processor within that product family. Key manufacturing partners such as ASUS, Dell, Ericsson, GIGABYTE, HPE, Lenovo and Supermicro will provide entry server systems supporting these processors.
Intel has also provided new details on its Crescent Island accelerator, built using Xe 3P technology. This architecture has been specifically designed to provide significant bandwidth and high memory capacities that are required for large language model training. It includes the use of LPDDR5x to reach capacities up to 480 GB, in a power efficient 350W air cooled PCIe form factor. All datatypes ranging from native FP4, through to MXFP4, all the way up to FP64 are supported, along with the use of Intel's open programmable stack to easily facilitate model creation.
