The workhorses of most data center servers are x86 processors. Intel regularly tunes its processors to address...
new usage scenarios, system architectures and competitive threats. Currently, this means optimizing processors for hyperscale clouds and scale-up, mission-critical enterprise workloads.
Intel recently completed the fourth-generation refresh of both Xeon processor families -- Xeon E5 and E7. While the new Xeon E5 v4 and Xeon E7 v4 target different infrastructure designs and applications, the architectural and feature differences remain blurred.
For most workloads, the biggest performance improvement in this release comes from using a smaller, more power-efficient process that results in more cores and cache in the same package and power envelope.
Here's what data center managers should know about the Xeon E5 v4 and Xeon E7 v4, including their effect on enterprise workloads and upgrade considerations.
Xeon E5 v4 designed for the cloud
The E5 family is built for scale-out, distributed or clustered workloads, such as those used for cloud services or big data technologies like Hadoop. The 2 series targets one- and two-socket servers, while the 4 series scales up to four-socket systems. As organizations increasingly deploy data center infrastructure as virtualized software instead of dedicated hardware, Intel now builds versions of E5 designed for network and storage workloads, such as firewalls, load balancers or network-attached storage and storage area network servers.
For compute servers, Xeon E5 v4 comes in three size ranges:
- Basic: 6 or 8 cores, 1.7 GHz with no turbo boost or hyperthreading
- Standard: 8 or 10 cores, 2.1-2.4 GHz with turbo boost and hyperthreading (two threads per core)
- Advanced: 12 or 14 cores, 2.2-2.6 GHz with turbo boost and hyperthreading
Every Xeon v4 processor, whether E5 or E7, has 2.5 MB of shared, on-chip cache per core. In Intel's Broadwell architecture, the cache isn't equally allocated, so a particular core may use much more than the 2.5 MB average with cache allocation automatically balanced according to the needs of each thread currently running.
Although the Xeon E5 v4 family is socket-compatible with the Haswell-based v3, it includes new features and microarchitecture changes that yield an aggregate 20% performance improvement per thread, measured as the number of instructions executed per clock cycle. These include:
- Cache allocation technology: The ability to partition on-chip, last-level cache per thread to enable workload prioritization, consolidation and resource partitioning, which improves the performance of virtualized workloads with noisy neighbors.
- Memory bandwidth monitoring: Per-thread checks of memory bandwidth utilization to identify conflicts and bottlenecks and enable thread migration to other processors.
- Cryptographic function acceleration that can as much as double performance of cryptographic algorithms.
Xeon E7 v4 for memory-hungry databases, analytics
Although Xeon E5 systems can support up to 1.5 terabytes (TB) of
Currently, most scale-out 1U or 2U servers are outfitted with 128-256 GB. In contrast, the Xeon E7 v4 is designed for 4- to 8-socket systems with as much memory as you can afford to throw at them -- 3 TB per socket or up to 24 TB in an 8-socket server. Like the Xeon E5 v4, E7 comes in small, medium and large sizes, ranging from an 8-core, 2.1 GHz, non-turbo boost basic configuration to a 24-core, 2.2 GHz frequency-boostable behemoth.
The Xeon E7 v4 includes all the microarchitecture and feature enhancements listed above for the Xeon E5 v4, plus it improves cache and memory access performance over the E7 v3. The E7 v4 also inherits dozens of reliability, availability and security features from the E7 v3 product.
Where to use Xeon E5 v4 vs. Xeon E7 v4
The E5 family is the mainstream Xeon for most workloads. It's suitable for virtual server farms, hyperconverged appliances, such as those from Nutanix, VCE and SimpliVity, and cloud stacks, whether with public services like AWS and Azure or on-premises installations using OpenStack or Azure Stack. Given the capacity of each system -- typically 2S, 20-32 cores and 128-256 GB RAM -- the Xeon E5 v4 is for organizations building rack-scale clusters of distributed infrastructure that are operated via the cloud or via software-defined automation software.
In contrast, the Xeon E7 v4 is designed to consume and analyze a lot of data on a single system. It works best with software such as an enterprise relational database management system that can use dozens of threads, terabytes of memory and needs near 100% uptime. Typical workloads include business intelligence, transaction processing and real-time analytics or vertical applications, such as those used in the finance, logistics and oil/gas industries.
The E7 is also good for organizations that prefer to operate a mainframe-style infrastructure and consolidate virtual workloads on as few servers as possible.
Quick tips for the Xeon E5 and E7 families:
- For distributed and scale-out cloud deployments, consider the E5 family.
- For monolithic, scale-up, in-memory databases and analytics, consider the E7 family.
- An upgrade is most critical when your infrastructure is three generations old.
- Don't upgrade last-generation servers unless you have money to burn or CPU-bound workloads with systems where you can do a CPU socket replacement.
E5 and E7 upgrade considerations
Since Xeon E5 v4 and E7 v4 are both socket-compatible with systems using version 3 processors, it's possible to do a single generation upgrade. However, given the incremental 20-40% performance improvement on most workloads, it's hard to justify the expense of replacing relatively new CPUs. Knowing the typical server lifetime is three to five years, Intel benchmarking reports often include performance comparisons going back three generations, and here the improvements are dramatic.
The Xeon E5 v4 provides about 2.7x the performance across a variety of workloads as a comparable v1 Sandy Bridge part. The results are similar when comparing inter-generational E7 performance. For example, going back three generations to the Nehalem architecture Westmere-EX (E7-4870) processor, Intel benchmarks show a comparable Xeon E7 v4 yields a 2.8x speed boost on OLTP workloads as measured by TPC-E and 3.7x on a suite of virtualized enterprise applications.
Intel aims toward the cloud with NVMe SSD
Xeon E7 processors beef up for big data
Increasing server density with E5, E7 processors