The Intel Xeon E5 chip is a mainstay of enterprise servers and its latest incarnation is designed make it better suited for use in virtualised environments, while boosting performance and cutting power draw over previous generations.
The third generation processor is available in more than 22 SKUs for both single socket servers, as the 1600, and dual socket, as the 2600.
Diane Bryant, general manager of Intel's data centre group, said the processor is designed to provide a common base for compute, storage and network-oriented workloads in the datacentre.
"There is a clear value proposition to running all compute storage and networking workloads on a common architecture," she said at an event launching the processor today.
"We're moving datacentre infrastructure from static to dynamic, from manual to automated, from proprietary to open standard running on a common architecture."
Two hundreds and fifty machines, servers as well as storage and network appliances, which use the new Haswell-EP-based processor will be available within 60 days. These machines include new releases from large server OEMs, such as Dell and HP. As well as delivering more than 22 standard SKUs, Intel has also released 20 custom versions of the processor tailored to customer workloads.
Overall processor performance is boosted by upping the maximum number of cores to 18, while a range of features have been added to accelerate common datacentre workloads.
One of the new features in the E5 v3 is Intel Advanced Vector Extensions (AVX) 2. AVX 2 improves the chip's ability to handle certain high-performance computing, professional imaging and feature detection tasks, by adding support for fused multiply-add operations on floating point numbers. Bryant said that Linpack benchmarks demonstrated an up to 90 percent performance increase compared to systems based on the previous generation E5 processor. Software handling relevant math, codec, image processing and digital signal processing-related functions should also benefit from a widening of the processor's integer registers to 256-bits. Running certain hashing algorithms delivers a two times performance boost generation over generation, said Bryant.
Another addition is support for DDR4 memory, which offers up to 44 percent increased bandwidth and up to 50 percent power efficiency compared to DDR3 memory supported by the previous generation E5. These benchmarks were generated by Intel using two Intel Xeon E5-2699 v3 processors and 384GB of 2133 MT/s DDR 4 memory arranged in three DIMMs per channel.
Meanwhile the chip's Intel Data Direct I/O feature allows a portion of the processor cache to be allocated to directly write data to HDD or SDD, bypassing the memory, improving I/O performance and reducing demands on memory bandwidth.
Another benchmark found Cloudera's version of the distributed computing framework Hadoop, commonly used for big data storage and analytics, demonstrated 110 percent performance improvement in certain tasks by using the latest version of the E5, Bryant said.
Helping to boost power efficiency is a new feature that allows the chip to throttle the frequency of each core individually. The Per Core P-States (PCPS) feature matches the frequency of each core depending on the load on each core.
The processor also makes information about the processor utilisation, temperature, power consumption and more available through its Node Manager 3.0 firmware. These 22 data points are designed to provide detailed information to the orchestration software used to manage physical and virtual infrastructure.
Other new features are designed to deliver better and more predictable performance to virtualised infrastructures that make cloud platforms possible. The Cache QoS features provides information on the individual virtual machines that are using the cache on processor cores. This information helps infrastructure orchestration engines to optimise where virtual machines should be running, for example shifting VMs taking up large amounts of cache to less heavily utilised processors where they won't cause such a problem for neighbouring VMs.
The processor also introduces support for 40Gb Ethernet controllers, with converged LAN/SAN. To better support network function virtualisation technologies Intel has released the next version of its Data Plane Development Kit, that incorporates support for 40 Gbe and improves throughput for packet processing over previous generation processors.
- 22nm Intel Xeon Processor E5-2600 v3 with new instructions
- DDR4 memory support
- Intel C612 chipset
- Enhanced Sata support
- Enterprise SMbus and MCTP support
- Intel SPS 3.0 Firmware with BMC-assist modules
- Can work alongside Intel Xeon Phi co-processor
- Intel SSD Data Centre Family for PCIe
- Family of PCIe storage cards with up to 2TB capacity
- Intel Ethernet XL710 family
- Intel Flow Director
- Intel QuickAssist Server Adapter
- Hardware-based crypto acceleration and compressions
4 - 18 cores
TDP: 55W to 145W (SVR); 160W(WS
|Scalability ||25 capability|
4 x DDR4 channels
1333, 1600, 1866 (2DPC), 2133 (1DPC)
|QuickPath Interconnect (QPI) |
2 x QPI 1.1 channels
6.4, 8.0, 9.6 GT/s
PCIe* 3.0 (2.5, 5, 8 GT/s)
PCIe Extensions: Dual Cast, Atomics
DMI2 – 4 lanes; Up to 6xUSB3, 8x USB2 ports,
10xSATA3 ports; GbE MAC (+ External PHY)
|LAN ||40GbE - 1GbE|
Servers: Intel Server Platform Services (SPS)
Workstations: ME 9.x
A range of new servers have been unveiled today based on the new Intel processor, including new line-ups from HP and Dell.
HP's enterprise-targeted line-up includes the tower format ML350 and the rack-mount DL360 and DL380 systems, plus the BL460c blade server. Two new models, the DL160 and DL180, replace the current DL360e and DL380e as the entry-level rack systems.
Improvements in Proliant Gen 9 systems over the predecessors include a PCI Express workload accelerator, HP Smartcache and Flexfabric adapters for improved performance, and converged management tools across servers, storage and networking.
HP will also launch a new server tray module option for its Apollo 6000 and 8000 high performance computing (HPC) systems based on the new Intel Xeon E5-2600v3 chips.
HP is hoping that its budget DL160 and DL180 models will help it compete at the low-end of the server market, where "the competition has been eating us up from below", according to Paul Schrady, HP VP and general manager for rack and tower for HP servers global business unit.
As well as the improvements brought by the Xeon E5 v3, other enhancements over Gen 8 servers include:
- Double the data throughput to storage – with 12GB/s connections to Smart Array Controllers, Smart HBA and SAS HDD and SSDs
- Compliance with Ashrae A3/A4 environmental guidelines
- HP OneView management software for both blades and rack servers
- New versions of HP SUM, iLO 4 and SPP management software
- Introduction of UEFI Secure Boot
- 20Gb FlexFabric adapters for blades
- 40Gb Ethernet FDR Infiniband
- RDMA over Converged Ethernet (RoCE) - an important feature in distributed systems as it reduces latency between servers in the cluster
- UEFI SecureBoot support
- Embedded 1GbE lights out management on DL servers
- RestFUL interface tool
Dell's new range of PowerEdge servers introduce improvements targeted at web, enterprise and hyperscale applications.
The new Dell PowerEdge portfolio span blade, rack and tower servers form factors. The range includes the PowerEdge R730xd, R730, and R630 rack servers, the M630 blade server, and the T630 tower server.
Data throughput and storage density is enhanced. The R630 uses 1.8-inch SATA flash drives to provide up to 2.4x the input/output operations per second (IOPS) in the same amount of space as 2.5-inch SSDs. Support for smaller drives allows up to 24 SSDs to be packed into 1U chassis, for 19.2TB of storage. For a SQL database the enhancements should allow up to 10.5x faster online transaction processing.
The R730xd provides up to 100TB of storage in a 2U chassis, with Dell citing it making 50 percent larger mailboxes possible.
Dell's Fluid Cache for SAN and DAS allows servers to cache the most important data on Flash storage.
Dell’s updated systems management capabilities also allows administrators to monitor and manage datacentres using a hand held device.
Additional server management tasks are automated, such as storing, accessing, and deploying a server configuration profile or firmware update from a central repository.
Dell also has introduced direct-attached storage offerings developed for the new PowerEdge servers. The new Dell Storage MD1400 and MD1420 offer double the bandwidth and improved performance compared to the previous generation offerings, as well as self-encrypting-drives (SEDs), and data management between the server and storage enclosure.
The new Dell PowerEdge servers will begin shipping September 8.
A mainstream two socket (2S)/2U rack server – targeted at providing scale out storage demanded by today’s cloud service providers, Hadoop/Big Data users and co-location hosting - optionally offers in-box storage tiering.
A mainstream 2S/2U rack server aimed at supporting virtualised platforms, virtualization, web serving and high performance computing. The two GPU option also enables medical imaging.
A mainstream 2S/1U rack server – designed for high density datacenter scaling for database, and virtualisation environments, web serving and high performance computing.
A mainstream 2S/5U rack-mount tower server – designed to deliver performance and scalability to mid-size offices, remote and branch offices and datacentres. Internal storage can support large databases and CRM systems, while and four GPUs enable desktop virtualisation and medical imaging.
A 2S blade server used with the 10U M1000e and VRTX shared infrastructure chassis. Suited to large data centers - hosting virtualisation, or running business intelligence applications and large databases. A building block for private clouds.