Covering Scientific & Technical AI | Friday, February 28, 2025

Intel Moving Forward with New Xeon Lineup 

(Source: Intel)

In September of last year, Intel launched its Xeon 6900P Granite Ridge processors and got themselves back in the game against AMD. This week, Intel continued to expand its processor line with the launch of the Xeon 6500P and Xeon 6700P server processors.

The new processors are available in up to 86 cores per socket and can scale to four and eight-socket platforms. While the new processors are a step up in terms of sockets, they are pretty much a step down in terms of their features; rather than the 12 memory channels of DDR5 offered by the 6900P, the 6500P and 6700P provide up to 8 DDR5 channels with some slightly slower memory options. In terms of PCI lanes, the 6500P and 6700P offer up to 88 lanes, compared to 96 lanes for the 6900P. Like their big brother, the 6900P, both processor lines support MRDIMM memory. They have up to 64 lanes of CLX 2.0, Intel AVX 512, and AMX (Advanced Matrix Extensions). Other differences can be seen in the figure below (also including the new Xeon 6 SoC edge processor line).

(Source: Intel)

As part of the release, Intel also touted the AI capabilities of the new processors optimized to capture a significant share of the growing market, delivering leadership performance in CPU machine learning, smaller GenAI models, and GPU-accelerated workloads as a host CPU. The ability to run the recently released DeepSeek model on x86 servers with good memory bandwidth and no GPU does point a place for traditional CPU in using models for inference.

In terms of HPC performance, Intel released a suite of benchmarks that indicated an uptick of 1.2 to 1.5 times faster than the Xeon 8592+ 64 core processor (with equal and maximum cores. The improvements are notable because they operate in the power envelope of 350. The graphic below depicts the results. Further independent tests are expected in the near future.

(Source: Intel)

Michelle Johnston Holthaus, interim co-CEO of Intel and CEO of Intel Products, commented on the processors, “The Xeon 6 family delivers the industry’s best CPU for AI and groundbreaking features for networking while simultaneously driving efficiency and bringing down the total cost of ownership.”

Edge processors

Intel also announced the Xeon 6 SoC, or Granite Rapids D. With built-in 200G Ethernet, the Xeon 6 SoC is designed as an edge processor. The flagship version, Xeon 6726P-B, has 42 cores, four channels of DDR5-6400 memory, 2.3GHz base frequency (2.9GHz turbo), a 235-watt TDP, and 200G Ethernet. The full line starts at 12-cores with the 6503P-B Granite Rapids D SoC.

The Xeon 6 SoC line ranges from 12-42 cores (price range $934-$3,795), with some of the lower core count models dropping the Ethernet to 100G. All models support Intel AMX (Advanced Matrix Extensions)

Speaking of Ethernet

Intel also announced two new Ethernet controller and network adapter product lines to address the growing demands of enterprise, telecommunications, cloud, high-performance computing (HPC), edge, and AI applications. Initial availability includes dual-port 25GbE PCIe and OCP 3.0-compliant adapters.

The E830 line of controllers and network adapters delivers up to 200GbE bandwidth with support for PCIe 5.0, flexible port configurations (x200GbE, 2×100/50/25/10GbE and 8×25/10GbE), and advanced precision time capabilities, including Precision Time Measurement (PTM). These adapters are optimized for high-density virtualized workloads, offering robust security features and performance.

The E610 series controllers and network adapters provide 10GBASE-T, 5G, 2.5G, and 1000BASE-T connectivity optimized for control plane operations. The 610 series offers outstanding power efficiency, 50% less power than previous generations, advanced manageability, and robust security features that simplify network administration.

Process forward

As recently reported at a conference in San Jose, California, senior principal engineer Steve Carson said Intel has produced 30,000 wafers, yielding thousands of computing chips, in a single quarter with its new ASML’s high numerical aperture (NA) lithography machines.


This article first appeared on HPCwire.

AIwire