Intel Debuts Midrange Xeon 6 CPUs To Fight AMD In Enterprise Data Centers

In launching its new midrange Xeon 6700P and 6500P server CPUs for enterprise data centers, Intel says its expanding Xeon 6 family can best AMD in performance and enable lower total cost of ownership for data centers across several workloads.

Intel said its new midrange Xeon 6 processors will help customers consolidate data centers for a broad range of enterprise applications and that they enable superior performance and lower total cost of ownership than AMD’s latest server CPUs.

Using the same performance cores of the high-end Xeon 6900P series that debuted with up to 128 cores using 500 watts last year, the Xeon 6700P series scales to 86 cores on 350 watts while the Xeon 6500P series reaches up to 32 cores on 225 watts, opening the Xeon 6 family for a broader swath of the data center market.

The semiconductor giant marked the launch of the Xeon 6700P and 6500P processors alongside the low-end Xeon 6300P series, the Xeon 6 system-on-chips for network and edge applications, and two new Ethernet product lines Monday.

The announcement came more than three weeks after interim Intel co-CEO and Intel Products CEO Michelle Johnston Holthaus vowed to “stabilize” Intel’s dominant but waning server CPU market share, saying that “this year is all about improving Xeon’s competitive position as we fight harder to close the gap to competition.”

Intel said its Xeon 6 processors “have already seen broad adoption across the data center ecosystem, with more than 500 designs available now or in progress” from major vendors such as Dell Technologies, Nvidia, Hewlett Packard Enterprise, Lenovo, Microsoft, VMware, Supermicro, Oracle, Red Hat and Nutanix, “among many others.”

Intel’s Xeon 6 Performance Claims Against AMD EPYC CPUs

In a briefing with journalists and analysts last week, Intel shared several examples of where the Xeon 6900P and 6700P series performed better and enabled a lower total cost of ownership than those from AMD’s fifth-generation EPYC CPU lineup, which debuted last year.

This marked the first time Intel provided several competitive performance comparisons for the flagship Xeon 6900P series since it launched in the third quarter of last year.

For instance, the company said internal testing showed that the Xeon 6900P series is faster than AMD’s EPYC 9005 series across four applications for data and web services when comparing similar chips at different performance levels. The Xeon chips were 62 percent faster for NGINX TLS 1.3, 17 percent faster for Mongo DB, 14 percent faster for Redis Memtier and 10 percent faster for Redis vector similarity search, according to Intel.

In high-performance computing, Intel showed higher gains for the Xeon 6900P series against AMD’s EPYC 9005 series: 52 percent faster for the HPCG benchmark test, 43 percent faster for the OpenFOAM computational fluid dynamics model, 23 percent faster for the LAMMPS molecular dynamics workload and 15 percent faster for the WRF weather prediction system.

But it was across four AI workloads where Intel said its Xeon 6900P series had the biggest advantage. Compared with AMD’s 128-core EPYC 9755, Intel’s flagship,128-core Xeon 6980P ran 2.17 times faster for the ResNet-50 image classification model, 87 percent faster for the DLRM recommender system model, 85 percent faster for the BERT-large language processing model and 76 percent faster for a transformer-based object detection vision model.

The company also said its 86-core Xeon 6787P has a performance advantage over AMD’s 128-core EPYC 9755 across multiple AI workloads: 3 percent faster for the LCM image construction model, 27 percent faster for BERT-large, 53 percent faster for ResNet-50, 25 percent faster for the transformer-based vision model and 21 percent faster for DLRM.

In a single-socket server, Intel said that its 64-core Xeon 6767P has a performance advantage over AMD’s 64-core EPYC 9575F for vector databases. When it comes to vector database indexing, Intel said the workload can take advantage of the CPU’s Advanced Matrix Extensions (AMX) for a 30 percent speed-up. As for graph-based similarity search, the CPU can run 2.6 times faster by using its Scalable Vector Search feature.

Intel also provided four examples of how the Xeon 6900P and 6700P series can not only provide better performance than AMD’s EPYC 9005 series but also lower data center costs.

For example, the company said the 128-core Xeon 6980P can enable 87 percent faster server performance than AMD’s 128-core EPYC 9755 for the DLRM recommendation system model, resulting in a total cost of ownership (TCO) reduction of 46 percent. For the OpenFOAM computational fluid dynamics workload, Intel said the CPU can enable a 28 percent TCO reduction with a 43 percent performance improvement in servers.

As for the Xeon 6700P series, Intel said the 64-core Xeon 6760P CPU can enable a 41 percent TCO savings for the NGIX TLS web services application with a 55 percent server performance boost over AMD’s 64-core EPYC 9535. The processors can also enable a 52 percent TCO reduction for a transformer-based image construction vision model with a 2.09 percent performance improvement for servers.

How New Xeon 6 CPUs Can Help Consolidate Data Centers

While Intel said the Xeon 6700P and 6500P series deliver “significant gains” over their fifth-generation counterparts across a broad range of workloads, the company took time to point out how the new chips can offer a substantial upgrade over its second-gen Xeon CPUs, not just for performance but also space, energy and money.

This is because the semiconductor giant believes there are ample opportunities to upgrade aging data center infrastructure, even as customers try to keep them running for longer.

“When we’re talking to our customers, our customers are living longer on the infrastructure that they have, but they’re also saying, ‘I have issues with my infrastructure,’” said Ronak Singhal, senior fellow and chief architect of Xeon products, in last week’s briefing.

For example, Singhal said, the ongoing demand for AI applications is pushing customers to figure out how to free up extra power within fixed infrastructure. He also pointed to customers looking to lower their energy consumption to meet sustainability goals or improve the manageability or security aspects of their data centers.

With Intel’s first- and second-gen Xeon processors representing a “significant volume of what’s deployed across all data centers today,” the company’s Xeon 6 processors can help customers significantly consolidate that infrastructure, according to Singhal.

One example from Intel pointed to a data center for MySQL data services consisting of 780 single-socket servers running on 28-core Xeon 8280 processors from the second generation. With 86-core Xeon 6787P chips improving performance by nearly 4.3 times, the company said the CPUs could enable the data center to reduce its server fleet to 180.

Across a four-year timeline, this would result in the data center reducing energy consumption by 14,735 megawatt-hours and CO2 emissions by 6,247 metric tons, according to Intel. It would also result in lower costs for cooling and hardware maintenance. All of this would result in a TCO savings of $3.3 million over four years, which would amount to a 29 percent savings.

Intel said the TCO savings can be greater for some workloads, such as MongoDB for data services and NGINX TLS for web services. For those workloads, the company said the Xeon 6787P processors can enable a TCO reduction of 44 percent and 67 percent, respectively, representing millions of dollars in savings across four years.

While customers can use Xeon 6 to reduce their data center footprint, they can also use the chips to crank up application performance and still use less energy, according to Singhal.

“It's a significant footprint reduction that they can have at the same performance or significantly higher performance at lower power, and so they have different opportunities on how they want to play this,” he said.

Xeon 6700P, 6500P Specifications, Features

With the Xeon 6700P series scaling up to 86 cores and the Xeon 6500P series scaling up to 32 cores, the lineup includes processors for servers with one, two, four or eight sockets.

These processors support up to eight channels of DDR5 memory with speeds of up to 6,400 mega transfers per second (MT/s). Several Xeon 6700P models can reach 8,000 MT/s by using the new, high-performance DDR5 MRDIMM modules.

While the Xeon 6700P and 6500P models for dual-, four- and eight-socket servers support 88 PCIe lanes, the single-socket models go up to 136 lanes for higher PCIe device density.

The processors come with Intel Software Guard Extensions (SGX) for application isolation and Intel Trust Domain Extensions (TDX) for virtual machine isolation. They also support Intel TDX Connect, which provides a “high-performance encrypted connection between the CPU and PCIe devices,” such as a GPU, according to Intel.

Other features include up to four Ultra Path Interconnect 2.0 links for CPU-to-CPU transfer speeds of up to 24 giga transfers per second, a maximum of 64 CXL 2.0 lanes, the CXL-adjacent flat memory mode for reducing memory costs as well as Intel AMX and Intel Advanced Vector Extensions 512 (AVX-512) for AI acceleration.