Intel announced a 4-20 core Xeon D-2700 and 2-10-core D-1700 with up to 3.0GHz/3.5GHz clock, Intel DL Boost for AI, Intel TCC and TSN, 100GbE LAN, and up to 32 PCIe Gen4 lanes.
On February 23, Intel announced its low-power 12th Gen Alder Lake P- and U-series Core CPUs, as well as an updated OpenVINO, an Alder Lake based Intel NUC 12 Extreme system, and a next-gen Sapphire Rapids version of its Xeon Scalable platform for high-end cloud servers. The chipmaker also unveiled a pair of Xeon D processors using a new “Ice Lake-D” architecture, which we cover here. Congatec announced some COM-HPC and Type 7 modules based on the SoCs, which we will get to soon in a separate report.
Intel’s Xeon D-2700 and Xeon D-1700 are designed for the “software-defined network and edge” applications in “in space- and power-constrained ruggedized environments,” says Intel. These include security appliances, enterprise routers and switches, cloud storage, wireless networks, AI inferencing systems, and edge servers.
Xeon D-1700 and D-2700 (left) and Intel chart showing range of deployment strategies with Intel Atom, Xeon D, and Xeon Scalable
(click images to enlarge)
The latest Xeon D cores are based on the same 10nm Sunny Cove core technology that debuted on the recently discontinued, low-power 10th Gen Ice Lake U-series processors. Despite the shared Ice Lake architecture, the platforms couldn’t be more different. Whereas Ice Lake-U operates at 15W, the headless Ice Lake-D Xeon processors range from 25W to 125W.
Compared to the last major Xeon D generation, the 14nm “Broadwell” Xeon D-1500, the 36 Ice Lake-D Xeons have more cores and higher clock rates plus an improved front end and branch predictor. Since the D-1500, Intel has introduced a 6th Gen Skylake based Xeon D-2100 family and a later, Broadwell based Xeon D-1600, but only the Xeon D-1500 has shown up on the lower end edge server products covered by LinuxGizmos.
The Xeon D-2700 is the successor to the 4- to 18-core, up to 3.0GHz D-2100 while the D-1700 replaces the 4- to 16-core, up to 1.6GHz/2.2GHz D-1500. (For more background on recent Xeon D SoCs and insights on the new ones, see this NextPlatform report.)
Xeon D-2700 (left) and Xeon D-1700 block diagrams
(click image to enlarge)
The new Xeon D-2700 is equipped with 4- to 20-cores at up to 2.5GH/3.5GHz on the octa-core Xeon D-2738. The Xeon D-1700 models range from 2- and 10-cores at up to 3.0GHz/3.5GHz on the octa-core D-1739. Cache ranges from 15MB to 30MB on the D-2700 models and 5MB to 15MB on the D-1700 SKUs.
Intel’s Xeon Ice Lake-D SoCs integrate the AVX512 Vector Network Instructions (VNNI) instructions for accelerating AI/inference workloads, as well as the related Intel DL Boost technology, which “accelerates AI workloads by eliminating unneeded precision in calculations so they can be completed more quickly,” says Intel. The technologies were introduced with Ice Lake-U and are available on 11th Gen and 12th Gen Core processors. Intel claims the D-2700 offers up to 2.4x times higher visual processing inference compared to a D-1600 and up to 7.4x better AI inferencing for IoT compared to a D-1500, among other claims.
The Xeon D-2700 offers up to 32x PCIe Gen4 lanes while the D-1700 supplies up to 16x lanes. Both models provide 24 lanes of PCIe Gen3, SATA, or USB 3.0.
Xeon D-1700 (left) and D-2700 parts list
(click images to enlarge)
The Ice Lake-D Xeons have built-in Ethernet controllers for up to 8x ports with support for up to 100GbE. The SoCs add support for Intel Time-Coordinated Computing (TCC) and Time-Sensitive Networking (TCC).
The Xeon D-2700 and D-1700 offer Intel Software Guard Extensions along with Intel QuickAssist Technology (QAT) for accelerating encryption and compression and enabling up to 100Gbps crypto and 70Gbps compression functionality simultaneously. Memory encryption is enabled via AES XTS extensions. The D-1700 offers a packet processor and switch with QAT Gen3 while the D-1700 has QAT Gen2.
The Ice Lake-D Xeons support 4-channel (D-2700) or two-channel (D-1700) DDR4. The D-2700 can use up to 512GB 3200 MT/s RAM and the D-1700 supports up to 384GB 2933 MT/s RAM. The 52.5 x 45mm D-2700 has a 64-125W TDP while the 45 x 45mm D-1700, which is larger than the earlier D-1500, operates at 25-85W.
Xeon D vs. Atom C3000 and P5900
In the deployment chart above, “Intel Atom” on the low end refers to the up to 16-core Atom C3000, which has been widely used in what we would consider high-end embedded gear but is on the low end from the vantage point of the 5G server world. Two years ago, Intel announced a 24-core, 10nm successor, the Atom P5900, but the SoC has yet to show up on LinuxGizmos since then.
According to a Light Reading report from last June, the P5900 is sold primarily to manufacturers of 5G gear for mobile networks, such as Ericsson, Nokia, and ZTE. The story noted that in a traditional radio access network (RAN), the Atom P5900 would look after the control plane and leave signal processing to a custom ASIC. By comparison in the new world of virtual RANs using technologies such as FlexRAN and OpenRAN, manufacturers have begun to develop base station servers that control everything from a Xeon without the help of an Atom. At the time, the story suggested that equipment vendors were skeptical the Xeon-D was up to the job. They would likely be more amenable with new Ice Lake-D models.
The Xeon D-2700 and Xeon D-1700 are available now. More than 70 companies are working with Intel on Xeon Ice Lake-D products, including Cisco, Juniper Networks, and Rakuten Symphony. More information may be found in Intel’s announcement, as well as its blog announcement and product page, which links to a more detailed product brief.
Intel | www.intel.com