Geekly Articles each Day
📜 ⬆️ ⬇️

Twice as powerful for the same money! Intel Xeon E5 vs. E3-series

Not so long ago, we published the article Comparison of the performance of Intel processors of different generations , which presents the results of testing 8-core Intel Xeon E5 processors. Since then, the promotion has sold several hundred servers based on V1 / V2 processors, which have become a worthy analog of the E5-2620V4 / 2630V3 in such tasks as: a database server, a terminal server, video surveillance, etc. But, unfortunately, a model the series was limited to only 2-processor platforms, which imposed an additional financial burden when using 1-processor configurations. A large number of requests prompted us to develop a budget model of a 1-processor server / workstation, which should significantly benefit in performance from platforms based on Xeon E3 / Core i7, and at the same time compete with them in price.

For this line of servers used a motherboard on the C602 chipset, which is not discontinued. Three types of performance were chosen as the basis: MidiTower, Tower / 4U and 2U.
In order to properly position the line of servers on the market,
Comparative testing of Intel Xeon E5-2650V2, E5-2680V2 and E5-1650V2 processors against E3-1230V6.

Stand configurations


Stand number 1:
1. Motherboard: MBD-X9SRL-F
2. Processor: Intel Xeon E5-2650V2
3. Memory: 4 x 8GB DDR3L-1333 ECC Reg
4. Drives: 2 x 150Gb SSD

Stand number 2:
1. Motherboard: MBD-X9SRL-F
2. Processor: Intel Xeon E5-2680V2
3. Memory: 4 x 8GB DDR3L-1333 ECC Reg
4. Drives: 2 x 150Gb SSD
')
Stand number 3:
1. Motherboard: MBD-X9SRL-F
2. Processor: Intel Xeon E5-1650V2
3. Memory: 4 x 8GB DDR3L-1333 ECC Reg
4. Drives: 2 x 150Gb SSD

Stand number 4:
1. Motherboard: MBD-X11SSH-F
2. Processor: Intel Xeon E3-1230V6
3. Memory: 4 x 8GB DDR4-2133 ECC
4. Drives: 2 x 150Gb SSD

Test results


PerformanceTest 9.0




A general test showed that E5 processors confidently outperform E3 processors due to a larger number of cores. And this is despite the lower frequency of memory and processor. In some computational problems, the advantage of a 10-core processor was more than 2 times.
Of course, the V2 processors have an early version of SSE and AVX, so the testing stage of the extended instructions was lost, as was the memory test - as different generations were compared: DDR3 and DDR4.

But all this did not prevent the E5-V2 processors from getting the best result in terms of the total coefficient.

CINEBENCH R15




Rendering is a task that is perfectly parallelized using all the computing power. Obviously, in this test, multi-core E5 processors also bypassed the 4-core Xeon E3 model. A similar result was observed in other software products, such as 3DsMAX. Also, the E5 line showed an almost 2-fold increase in video coding tasks.

7-zip




The performance test built into the 7-ZIP archiver showed a result similar to the previous test - parallel calculations are faster performed on a larger number of cores, albeit with a lower frequency.

3dmark


Of particular interest were the results of the game test - how the server processor will behave in the popular 3DMark benchmark.

For these tasks, a separate stand was assembled with a mid-level gaming graphics card.
Compared desktop processor i7-7700 and server E5-2650V2



The result did not reveal serious performance differences, but showed that even in games, the E5-2650V2 is not inferior to the desktop i7 with a frequency of 4.2 GHz in the Turbo Boost mode.

A brief analysis of the characteristics


In order to understand in part why we see such results, let's consider the main parameters of the tested processors.



In addition to the standard characteristics, I added one line - "The difference between the base frequency and the maximum frequency during acceleration." It is in this parameter, in my opinion, that the success rate of the multi-core Xeon E5 is hidden. After all, how does Turbo Boost work? If the load on all the cores is not high, or some of the cores are completely idle, the processor can overclock the active cores to a frequency higher than the base one. Moreover, acceleration can occur in steps - depending on the load.

So, the delta of these frequencies in the 2650V2 and 2680V2 is 0.8 GHz, and in the 1230V6 - only 0.4 GHz.
This means that if the process supports parallel calculations, but is limited (for example, in 3-4 streams), then the 1230V6 processor will be able to process it at a frequency of 3.5 GHz, since Turbo Boost will not be active due to the load on all the available processor cores. And the 2680V2 processor will be loaded only by a third, therefore, it will be able to process a similar task at a frequency close to the maximum - 3.6 GHz. This situation is likely if the necessary conditions are met - the latest version of Turbo Boost evaluates many parameters during overclocking, including the thermal mode and the power consumption of each core separately and the processor as a whole. So, if there is no overheating and the power consumption is normal, theoretically, nothing prevents the processor from overclocking not only one (as in the earlier version of Turbo Boost), but several cores at once.
In my opinion, it is this effect that we saw in the 3DMARK test results.
Well, if the process is parallel ideally, then 10 cores against 4 show the obvious result, despite the difference in frequency.

The ability to significantly increase the frequency in single-threaded computations and use all cores with multi-threaded ones makes the E5 processors a universal solution for most typical and specialized tasks, including both professional and home ones.

Price analysis of models


We have three 1-processor models based on multi-core Intel Xeon E5-V2 processors:


Let's compare the cost of classic 2U servers based on E3 and E5 in a test configuration:


Let's compare the cost of 4U / Tower-servers based on E3 and E5 in a test configuration:


Let's compare the cost of Miditower servers / workstations based on i7 and E5 in a test configuration:


Total - in the case of rack-mount design, the model based on the E5 has the same or lower cost than the model based on the E3.

In the case of a workstation, the situation is different due to a not entirely direct comparison - in the i7-based workstation, the desktop motherboard and memory without error correction are counted. The workstation on the base of E5, on the contrary, has a server MP and RAM. If you compare not with a PC on i7, but with a server on E3, then models based on E5 will also be slightly cheaper.

Why do we sell "outdated equipment"? Our model range includes different generations for two reasons. The first reason is reflected in the last article about performance comparison - the new generation does not always provide performance advantages. The second reason is the lifespan of the model. The low lifespan of the model is detrimental to the project business. There are projects that are being implemented for a very long time. From the moment of design to the moment of delivery it may take more than one year. If the lifespan of the model is lower than the implementation time, then there are obvious difficulties in reconciling the differences between the specifications laid down in the project and delivered in fact.

Such problems are faced by fans of HP and other vendors. The 8th generation of servers was laid everywhere, and after the release of the 9th it quickly disappeared from the market. The same is waiting for Gen9 in the near future - some models are either already out of production, or will be removed in the near future and will be difficult to access, because formed a new model line on Gen10.

Why is V2 cheaper than V4 or Scalable? Initially they were purchased, in due time, at the same prices. But, recently, the market has been filled with offers of second-hand servers based on processors of this generation at prices substantially lower than those of our new servers. Therefore, it was decided to mark down the components to create a competitive environment. This action showed a positive effect, which prompted us to extend it until the end of the production of motherboards and the full sale of existing stocks. At the moment, new servers are available with a 3-year warranty at prices significantly lower than the latest generation.

Or, in the framework of this article, comparing with lower-end Intel Xeon processors - much more powerful, but for the same money.

findings


Comparison of performance and price analysis shows that the line of servers / workstations based on Xeon E5-V2 processors can be not only a more budgetary alternative to the latest generation servers with equal power, but also significantly outperform 1-processor systems in the same price range.

The single-processor STSS Flagman servers of the EX series are actively used by our customers for various tasks, starting with database servers and terminal servers, ending with video surveillance and virtualization. As I already mentioned, there is certainly a certain percentage of software that uses the latest technology of the latest generation of processors in the calculations. Under such a specialized range of tasks, we recommend new models based on Intel Xeon Scalable.

If we consider the STSS Flagman EX workstation as a personal workstation, then we get a high-performance multi-core server configuration for the same money as a PC based on the latest generation i7.

From personal experience


When I replaced my working sistemnik with the configuration of E5-2650V2 + 32 GB of RAM - I forgot what brakes are in principle. On duty, I constantly have something spinning on my computer, video is being encoded and played back, software is being tested. And this is in addition to the usual office tasks. Constantly open 3 browsers and 100,500 tabs, databases, email, tables. Everything flies, the processor confidently works in Turbo Boost at 3 GHz (3.4 GHz at the peak), the load rarely exceeds 65%. Memory - from 32 GB maximum eaten off 24-26 GB. The reason for replacing the PC was just the inability of the desktop 4-core processor to cope with my everyday needs.

Habr's readers who apply for the STSS Flagman EX line with the HABR code word receive a 5% discount .

Make a request through the configurator or write to 1@stss.ru

Thanks for attention!

Source: https://habr.com/ru/post/351268/

More articles:


All Articles