This article is part of an Essential Guide, our editor-selected collection of our best articles, videos and other content on this topic. Explore more in this guide:
2. - Hardware: Server refresh options: Read more in this section
Explore other sections in this guide:
- 1. - Is it time for those servers to go?
- 3. - Server software upgrades: The other side of the refresh
Server processors are shifting away from speeds-and-feeds to deliver efficiency, management, security and reliability. Intel's latest generation chips are a testament to this future for processors.
A processor is at the heart of any server or endpoint, so IT professionals need to understand advances made in processors and processor technologies when planning for technology refresh and upgrade projects.
In modern data centers, speed is no longer enough. For decades, processors emphasized clock speeds as a measure of performance. Each new chip release did little if anything new -- it did the same things faster. But as clock speeds approached 3 GHz, designers realized that the enhancements would not continue. Major x86 chip vendors can't even cram more cores into existing processor packages.
Processor designs now highlight efficiency over raw speed; an efficiently designed server processor handles workloads faster and with less power than a monolithic, do-everything processor -- a work-per-watt measurement rather than pure speed.
Server processors will diversify in their purpose and functionality, with more emphasis on system-on-a-chip (SoC) designs. By tailoring processors for specific jobs, rather than putting every function into every model, chip makers achieve better performance even at slower clock speeds and lower power draw. For example, Intel is releasing Atom processors in several variations: the Atom S12x9 with 40 lanes of PCIe 2.0 for storage subsystems; the Atom Avoton 64-bit SoC for microservers; and the Atom Rangeley for network and communication devices. Systems designers choose the right chip for the job.
Power. The thermal design power (TDP) of server-class processors is dropping, easing a server's thermal and power requirements. Processors have always demanded the majority of a computer's power, as high as 130 W TDP for a typical processor: AMD's Athlon 64 FX-74 carries a 125 W TDP; Intel's Xeon E7460 hits a 130 W TDP. Lowering TDPs mean lower power consumption and cooling demands at the server. Smaller transistors born of advanced fabrication technologies have different thermal characteristics than earlier generations. Thanks to specialization, processors reduce heat output by operating at lower voltages and slower clock speeds: Intel's Xeon E3 1200 v3 processor offers a TDP of just 13 W.
Graphics support. Software interactions also affect processors' power consumption. Intel added graphics capabilities to the new E3 and improved the graphics transcode performance, enabling the same graphical processing as earlier versions with lower power draw.
Memory. Server processor memory support hasn't kept up with the addressable space allowed by 64-bit addressing. Memory support is moving from 1 to 2 TB of memory per server, to up to 12 TB on large, high-end, multi-CPU motherboards, as seen in Intel's latest-generation Xeon E7 models. This degree of memory support targets high-performance databases and data analytics performed in-memory. These servers could also handle massive workload consolidation for the data center.
Management. Native management support will allow server management tools to communicate directly with the processor, offering IT professionals more granular insights. Although the software tools may be vendor-specific, this server processor improvement will eventually allow other software to leverage its monitoring and reporting capabilities.
Intel's Xeon X5 family of processors will support management tools such as Intel Node Manager and Intel Data Center Manager (DCM) software. Node Manager helps optimize power and cooling in the data center by reporting actual processor and memory power consumption, with policy options that limit power use. DCM provides a software development kit for power and thermal monitoring and management of servers, racks and supporting equipment.
Security. Data centers protect against viruses and other malware and security threats in myriad ways, including at the system-hardware level. Servers face targeted attacks directed at vulnerabilities on the processor and its instruction execution.
Server security advances include motherboard technologies -- consider the Trusted Platform Module -- and processor execution and instruction hardening. One potential weakness is random number generation, leading to initiatives to improve randomization for superior cryptographic keys. Intel's Xeon E5 processors will incorporate an additional instruction (RDRAND) and random number generator for cryptographic uses.
New processors also strengthen protection against malware. For example, instructions should not execute in memory space that is not set aside for a workload, but in a privilege attack, malware can gain privilege in the OS and run malware outside of the application space. Better privilege security prevents malicious software from running in memory located outside of the space allocated for the particular application.
Self-healing. The push is on for more processor reliability or, rather, recoverability. When a processor error occurs, the results are usually catastrophic and require a system restart, which can disrupt data center operations. With better processor, firmware, hypervisor and OS integration, servers could recover from errors that previously proved fatal. Intel has promised this kind of technology, which they dub Run Sure, in future Xeon E7 processors. Any push toward better processing reliability will also need to address memory reliability techniques, such as memory sparing or mirroring, which preserve data integrity and reduce the urgency of service calls.