Intel today took the wraps off its Xeon 6 data center processor family, aimed at powering the next wave of artificial intelligence (AI), cloud and high performance computing (HPC) workloads.
The new lineup introduces two distinct microarchitectures: Xeon 6 with P-cores for HPC and Xeon 6 with E-cores for high-density, efficient cloud deployments.
The P-core variants, led by the flagship Xeon 6900 series launching in Q3 2024, are squarely targeted at AI training and HPC (HPC) workloads. Intel claims up to 2.3x higher performance versus prior generations in mainstream HPC applications. While the P-cores are about pure power, the E-core-based Xeon 6 7000 series launching first on June 4 is about helping to improve data center economics through efficiency for cloud-native workloads.
“The data center isn't one size fits all, there is diversity across the data center in terms of workloads and capabilities,” Matt Langman, GM and VP of Intel Xeon 6 processors said during a media briefing. “Workloads like AI and HPC will lean towards these P-core deployments, while other workloads like cloud-native, web, microservices and digital services will lean more towards E-core, taking advantage of the higher density and improved performance per watt.”
Inside the Intel Xeon 6900 series P-coreThe Xeon 6900 P core, also known as Granite Rapids is noteworthy because it will provide higher levels of performance improvements across workloads compared to previous generations of Xeon processors.
Some key things about the Xeon 6900 P core highlighted by Intel include the following:
- It’s optimized for high performance per core and will provide the highest performance of Intel's Xeon 6 lineup.
- It’s projected to provide up to 2.3 times higher performance than prior generations for HPC workloads.
- For AI workloads, it is expected to deliver up to two times better performance for AI inference compared to previous Xeon platforms.
- With improvements in core architecture, memory bandwidth and accelerators like AMX, it aims to be the leader for running AI head nodes.
While the P-cores flex their muscle, the E-core-based Xeon 6 7000 series launching first on June 4th aims to redefine data center economics with higher density and efficiency for cloud-native workloads.
As the first product based on Intel's new Efficient core microarchitecture, it provides significant improvements in performance per watt and performance per core compared to previous-gen Xeon platforms. It offers up to 2.6x higher performance per watt for workloads like web microservices and up to 3.4x improvement for networking workloads like firewalls.
“To give you an example, Xeon 6 E-core provides a three to one consolidation rate ratio at a rack level versus our second generation Xeon platforms,” Justin Hotard, EVP and GM of the datacenter and AI group at Intel said in the launch presentation. “That means you can deploy the same amount of performance in a third of the rack capacity.”
Granite Rapids has a chip-on-chip designThe new Xeon has a host of enhanced capabilities across both 6900 and 6700 series variants.
Among the architectural improvements are support for DDR5 memory and Intel's new MCR DIMM technology for bandwidth and capacity, more PCIe 5.0 lanes for accelerator attachments and enhanced security capabilities.
Under the hood, Xeon 6 leverages a multi-chip design, partitioning the compute cores and I/O logic onto separate silicon chiplets optimized for different manufacturing processes.
“We have a partitioning where a lot of the I/O capabilities are aggregated in a silicon construct and that is optimized on one technology node,” Sailesh Kottapalli, Intel senior fellow explained during the media briefing. “Then the rest of the compute along with the optimized memory support is on a technology node that brings in the most performance as well as power efficiency.”
Comments