Xeon Scalable Makes up a Full Slate of Server Components
Intel just released a new generation of server processors, branded as Xeon® Scalable. These processors dispense with older Xeon processor naming conventions, such as E5 and E7, and instead embrace a simple Bronze, Silver, Gold, and Platinum lineup.
For many, Intel is a company that “only” designs and manufactures processors, but in fact Intel designs a wide range of components for both servers and client computers. If you think about it, this approach makes sense. Many types of innovations in a CPU require complementary upgrades in other server components—such as chipsets, bus, networking, storage, I/O, software, and firmware—to show the benefits from those innovations.
To better reflect the reality of Intel’s broader focus and product line, Intel is touting Xeon Scalable not only as a new lineup of processors, but as an entire server platform. Xeon Scalable really isn’t just a new CPU; Xeon Scalable also refers to a broad lineup of server components that accompany these new processors. In some cases, the processor includes support for a related new server platform component, such as a storage or networking feature. In other cases, it’s the complementary component that supports some related new feature or performance capability in the CPU.
There’s a lot to this new server platform, and a full list of its components and features is hard to find. In the interest of documentation, we’ve compiled the following list as an attempt to capture most of the new components and technologies that make up the Xeon Scalable platform.
New CPU Features
- Up to 28 cores (compared to 22 in the previous generation)
- Greater memory bandwidth
- Six DDR4 2600MHz memory channels per CPU (up from 4 per CPU)
- Higher I/O
- Support for 48 lanes of PCIe 3.0 (up from 40)
- Uniform shape for all Xeon Scalable CPUs that fit into the same socket type, Socket-P.
- Intel Mesh internal architecture that improves performance relative to the earlier “ring” architecture
- Requires no more cooling (and sometimes less) than previous generation despite the increase in cores
- Thermal design power (TDP) = 45-165 watts, as opposed to 115-165 watts for the previous generation
- AVX-512 instructions
- This is an updated version of the vector operations instruction set that was known as AVX2 in the previous generation. (Here’s a helpful animation that explains the basics of AVX-512.) The AVX-512 instruction set is intended to accelerate performance for the most processor-intensive workloads, such as scientific simulations, financial analytics, AI, 3D modeling, audio and video processing, cryptography, and compression.
- Intel VMD
- New PCIe lane control points in the root complex of the CPU support hot plug and LED management for NVMe drives.
New On-Board Features and Technologies
- Support for up to 8 sockets
- All sockets are the same size (unified from previous generation)
- Intel Ultra Path Interconnect (Intel UPI)
- This is a faster inter-processor channel among the (up to) 8 sockets, replacing QuickPath Interconnect (QPI)
- On-board support complementing the CPU features of 28 cores, 6 channels of DDR4 memory, and 48 PCIe lanes per processor
- Product Family: Intel Server Board S2600WF
- Support for Intel Optane SSDs and Intel 3D NAND SSDs
- These new types of drives require server board support
New Chipset Features
- Faster chipset (C620 Series)
- Support for 1.5x memory bandwidth compared to the previous generation
- Intel QuickAssist Technology (Intel QAT)
- Chipset-based hardware acceleration for compression and cryptography
- New, faster DMI3 chipset bus
- 4 lanes
- 3.93GB/s and 8.0 GT/s (up from 2GB/s and 5.0 GT/s)
- Support for Internet Wide Area Remote Direct Memory Access Protocol (iWARP RDMA)
- Provides support for high-throughput, low-latency communication between memory in user space across a network, via TCP/IP and 4x10Gbps Ethernet
- Intel Platform Trust Technology (Intel PTT)
- Integrated into the chipset
- Eliminates the needs for a separate Trusted Platform Module
- Innovation Engine
- Embedded core in the Peripheral Component Hub (PCH) designed to be a dedicated subsystem for system vendor custom firmware
- Improves input/output (I/O), compression, and encryption of data
- Improves error containment and helps ensure system survival for mission-critical data
New Networking Features and Components
- Intel Ethernet 700 series
- Common architecture framework that supports 10, 25, and 40 GbE
- Ethernet Converged Network Adapter X710
- Supports 4x10GbE
- Intel Ethernet Network Adapter XXV710
- Supports 2x25GbE or 2x10GbE
- Intel Omni-Path Architecture (Intel OPA)
- New high-speed interconnect for clusters
- Replaces InfiniBand
- Relies on fiber optic cable
- Eliminates need for a separate interface card (cards are “connectionless” and integrated into the server board)
- Flagship: Intel Omni-Path Host Fabric Adapter 100 Series
New Storage Features and SSD Drives
- Intel Optane SSDs
- A new type of SSD based on a technology, called 3D XPoint, that is much faster than traditional NAND devices.
- Flagship: Intel SSD DC P4800X
- New NVMe SSDs based on 3D NAND technology
- Flagship: Intel SSD DC P4500 and P4600 series
- New SATA SSDs based on 3D NAND technology
- Flagship: Intel SSD DC S4500 and S4600 series
- New internal architecture supports improved performance
- New RAID features
- Virtual RAID on CPU (VROC)
- Bootable, high-performance RAID built into the CPU for NVMe drives.
- More resistant to power outages than traditional RAID.
- Flagship RAID module: RAID Module RMS3AC160
- Tri-mode: Supports SATA, SAS, and NVMe drives
- Virtual RAID on CPU (VROC)
Backplane / Expansion Cards
- Various hot-swap backplanes supporting SATA, 12Gbps SAS, and NVMe
- 5” and 3.5” versions