The new entrant in the Xeon processor family, the Xeon processor E5 v4 has now been launched. It is a much more complex processor compared to the Skylake based Core i7 a part of Intel’s latest core technologies used largely in mainstream desktop products, being more mission oriented the Xeon E5 v4 series will surely influence the consumer space. The microarchitecture of the Xeon E5 v4 is based on the Broadwell-EP. The Xeon E5 v4 series is looking forward to target different areas of the market like multi-socket servers to professional workstations. The Xeon E5 v4 series supersede the E5 v3 series of last year, whose microarchitecture was based on Haswell-EP. The Xeon E5 v4 has 22 cores compared to the 18 cores, which were present in the E5 v3.
The E5 v4 series has a number of updates and modifications, but is socket compatible with the v3 series. The memory speed has been enhanced from 2133MHz to 2400MHz and can support memory up to quad-channel DDR4. The chip has a phenomenal CPU performance with a multi-threading now providing the user with up to 44 threads, each core is scattered with a 2.5 MB cache and has a range of 2.2 GHz base clock to 3.6 GHz boost clock. With a TDP of up to 145w for servers, the chip has up to 160w for workstations. The CPU block is divided into various cache blocks with a single core having 2.5 MB L3 cache which into the processor comes to 55 MB L3 in total. The v4 will have an increase in the boost clock and TDP range in comparison to the v3 and an improved 14 nm process node supports the evolution in the v4. Intel aims to provide a large market choice for professionals using durable servers and workstations. In the present scenario Intel rules the server market with little competition from IBM, whose power line up is seen in supercomputers, but Intel has been already well established in such areas.
Modifications
- High speed floating point instruction rendering with the large data sets having a better performance than before.
- Better parallelism in micro-operations scheduling.
- Increase in number of cores per CPU, up to 22.
Due to the modification in the microarchitecture of the E5 v4 series, the processing time has become faster stripping out the difficulties from their predecessor. The Xeon E5 v4 series along with their high level updates also has new virtualization and security related qualities. Support for posted interrupts is one of the new attributes in the series, as interrupt handling for CPU is directly conveyed; latency of data access is reduced by combined working of posted interrupts with APIC virtualization. Posted interrupts also increases the speed of VM enters/exits.
Transition from v3 series
The enhancement to the new Broadwell CPU’s from the Haswell CPUs is simple as they are socket compatible. The platforms which are supported by the v3 CPUs are also accepted by the v4 series, thus helping those who are looking for a new faster system which accepts the existing platforms.
Each CPU core in the v4 series provides efficient performance, a higher number of cores being offered by many models. This powers the v4 series to be on the same platform with the previous CPU models. Few comparisons are as follows:
- 2630v4 provides efficient working and performance equal 2640v3.
- 2640v4 is equivalent to 2650v3 in every aspect.
- 2660v4 is equivalent to 2670v3 every aspect.
- 2680v4 and 2690v4 beats 2600v3 CPU.
Important Changes
A simplification in choice is possible due to the removal of the E5-2670 from the CPU line-up also previously, the 2623 model of the E5 series was a high-clock-speed model. There is, however a decrease in the speed of the base clock by 18%. Due to the modifications in the E5 v4 CPUs memory configuration, a reduction in latency and increase in bandwidth was achieved. A 15% bandwidth increase and reduction in latency is depicted in Intel’s numbers.
Due to many new features in the Broadwell CPUs, they tend to be a more appealing choice for virtualized servers. Cases where low priority work applications slow down high priority applications can be avoided due to finer granular control over the sharing of cache and memory bandwidth by applications. For a smoother running of I/O applications on top, virtualized layer enhancements have been made. These new changes in the v4 series will surely be used up to its potential as 60% of the business today run their applications based on virtualization, which is the main ingredient in the swiftly growing cloud services. Intel also made some tweaks mostly in the simulation and scientific calculation department areas in order to keep the HPC crowd content. But this does not prove that HPC and virtualization are the only two contenders for the new 22 cores Xeons. By widening the network and storage bandwidth bottlenecks the upcoming crop of data analytics frameworks has taken a huge performance step forward. Apache spark is one such example, just with better of use of RAM it was able to crunch its way through massive amounts of data much more coherently than Hadoop. Another example is of how you can use the most advanced statistical and machine learning algorithms to extract results out of huge mountains of text data. By combining machine learning and data mining you can create an application that is highly CPU oriented and in which to keep its CPU busy, NVMe – based SSDs are not required.
Also, there is the new Resource Director Technology, or also called RDT, which is accessible with the Xeon E5 v4 series processors. The monitoring and managing of shared platform resources in great detail, including the cache or memory bandwidth allocation can be done with the help of operating systems and virtual machine managers under RDT. Resources can be segregated according to the application with RDT, which functions by allotting RMID qualities to a specific thread. This processor will soon be released in the market and it is hoped that it will dominate all of its predecessors with various enhancements and changes in its working features.
