On March 15, I attended the Intel press conference, where the company introduced the new Xeon E5 series 2600 server processors. The official announcement of these processors took place in early March, and the processors themselves showed a week before during the Cebit exhibition
about which my colleague
vilianov wrote . The event featured workstations, servers, and even a whole cluster with new processors. In addition to the speeches of the company representatives and partners, the performance was compared to the performance of old and new systems.
Let's see how the new processors will surprise us.

New processors contain 8 physical cores (2 more than the previous models), which with support for HyperThreading gives 16 logical processors.
Screenshot of the task manager on a two-socket system:
')

8 cores and 20MB L3 cache (compared to 12MB) explain the huge number of transistors - about 2.263 billion. Even using 32nm technology, the crystal size is 416 mm
2 .

A little bit about microarchitecture.
The new Xeon model is distinguished not only by an increased number of cores, but also by changes in the micro-architecture:
- Advanced Vector Extensions (Intel AVX)
AVX is an extension of the SSE instruction set that doubles to 256 bits and increases the size of the register for processing vector instructions (so far only for working with a float point, but with a planned extension of functionality) and providing support for instructions with three operands. - Intel Turbo Boost Technology 2.05. This technology allows to increase the clock frequency in the case when not all computational cores are involved.
In the previous generation of Xeon 5690 processors, one active core could increase its frequency by 266 MHz, and in the E5-2690 processor - up to 900 MHz. - Intel Integrated I / O Technology (Intel IIO).
Intel IIO allows you to move the I / O subsystem controller from a separate chip on the motherboard directly to the processor. - Intel Data Direct I / O Technology (Intel DDIO)
This technology allows peripheral devices, such as Ethernet controllers, to route I / O subsystem traffic directly to the processor’s cache. As a result, the amount of data transferred to the system memory, power consumption and latency of the I / O subsystem is reduced. - Intel Advanced Encryption Standard - New Instructions (Intel AES-NI) instruction set.
It is an updated technology of hardware acceleration of encryption according to the Advanced Encryption Standard (AES).
The Xeon E5 chip contains an integrated four-channel DDR3-1600 memory controller (the previous generation Intel Xeon supports three-channel DDR3-1333). The amount of supported memory is 768 GB.
Test results.
All of the above resulted in a noticeable advantage over the previous generation, in confirmation of which I will bring independent tests from such authoritative publications as
Tom's Hardware and
AnandTech .
The first highlighted the scenarios characteristic of workstations:

AnandTech - mainly server:
Test | The advantage of Xeon E5-2660 over Xeon X5650 |
---|
Virtualization (ESXi + Linux) | + 40% |
DB (MS SQL Server 2008 R2) | + 30% |
HPC (LS Dyna) | + 77% |
Encryption / Decryption (AES) | + 42/41% |
Encryption / Decryption (Twofish / Serpent) | + 37/49% |
Compression / decompression | + 35/37% |
Power usage.
It is very important that the comparison was made between processors with almost identical TDP. Thus, the performance indicator for 1W increased significantly. In the server market, this is very important, since performance per square meter and per 1W determine the main part of the so-called "cost of ownership".
Manufacturers of final solutions.
Among the manufacturers of final solutions, Intel has noted a number of major global manufacturers. At the Russian presentation itself, servers and workstations from Aqarius, Arbyte, Depo, Etegro, iru, kraftway, R-Style, Svet Computers and RSK were presented.
I was most impressed by the cluster system of the RSC group of companies.
It is a “super-dense solution” based on the RSC Tornado architecture, in simple terms, a rack-cabinet for blade-servers with liquid cooling. The standard rack height of 42U accommodates 128 dual-processor servers.
Thanks to liquid cooling, a record low percentage of power consumed for cooling components is achieved. Only about 6%. In traditional systems, this value often reaches 30-40%.
And the system looks more than impressive:


Well, finally, a little tasty.
In this post, I told the most general points, but the respected Habrapolz users have the opportunity to contact Intel experts directly with questions and not only get answers to them, but also participate in the prize draw. In order to take part in the contest, you need to leave your question on the
corresponding page of the
IT Galaxy community.
The authors of the 3 most interesting, according to experts, questions will receive a solid-state drive Intel 520 series with a capacity of 60 GB. Competition questions are accepted until April 21.