Comparison of i7 processor performance. Comparison of Intel and AMD platforms: What is the difference between processors

Part 1: 53 configurations with integrated graphics

Changing the year on the calendar usually leads to updating testing methods computer systems, and therefore - to summing up the results of testing central processors (which is a special case of testing systems) carried out in the past year. In principle, we received the bulk of the results long before the end of the year, but we wanted to add the “seventh generation” Core to the results (at least in limited quantities). Unfortunately, this was not possible: the “original” version of Windows 10 used in tests using the 2016 method is incompatible with graphics Intel drivers, suitable for HD Graphics 630. More precisely, of course, the opposite: this driver requires at least the Anniversary Update. In principle, there is nothing new in this, the latest versions of graphic Nvidia drivers, for example, behave similarly, but changing the set of test bench software violates the concept of tests “in the closest possible conditions.” However, tests of new processors using the 2017 method have already shown that there is nothing truly “new” in them - as expected. Therefore, it is possible to do without the results of “Skylake Refresh” for now, which is what we will do.

The second point that should also be taken into account is the number of subjects. Last year's results presented the results of 62 processors, 14 of which were tested with two “video cards” - an integrated GPU (different for everyone) and a discrete Radeon R7 260X, and four with different types of memory. In total there were 80 configurations. It’s not that difficult to “shove” them all into one article (after all, not so long ago we had 149 test configurations in one article ), but the diagrams were, to put it mildly, not very convenient to view. In addition, there is no great need for a direct comparison of the “atomic” Celeron N3150 and the extreme ten-core Core i7-6950X: these are still fundamentally different platforms. The “vastness” of the final articles using the “old” methods was mainly due to the fact that in the main line of tests all participants worked with the same discrete video card, but this approach was not always applicable before - as a result, some computer systems had to be removed into a separate line of tests, and then summarize individual testing results.

This year we decided to do the same. Today's article will present results from 53 different configurations: 47 processors, five of which were tested with two different types of memory, and one with different TDP levels. But everything is done exclusively using the integrated GPU (also different for everyone). To some extent, this is a return to the results of 2014 - only there are more results. And in the near future, those who wish will be able to familiarize themselves with summary material based on testing of 21 processors with the same Radeon R9 380. Some of the participants overlap, and in general the test results are “compatible” with each other, but to improve their perception, it seems to us, better two separate materials. Those readers who are only interested in dry numbers can (and for quite a long time) compare them in any set using the traditional one, which, by the way, also includes information on several “specialized” tests, adding which to the final materials is somewhat difficult.

Test bench configuration

Since there are many subjects, it is not possible to describe their characteristics in detail. After thinking a little, we decided to abandon the usual short table: anyway, it is becoming too vast, and at the request of the workers, we still put some parameters directly on the diagrams, just like last year. In particular, since some people are asking to indicate right there the number of cores/modules and computational threads running simultaneously, as well as operating clock frequency ranges, we tried to do just that, adding information about the thermal package at the same time. The format is simple: “cores (or modules)/threads; minimum-maximum core clock frequency in GHz; TDP in Watts.”

Well, all other characteristics will have to be looked at in other places - the easiest way is from manufacturers, and prices - in stores. Moreover, prices for some devices are still not determined, since these processors themselves are not available in retail (all BGA models, for example). However, all this information is, of course, in our review articles devoted to these models, and today we are engaged in a slightly different task than the actual study of processors: we collect the data obtained together and look at the resulting patterns. Including paying attention to the relative position not of processors, but of entire platforms that include them. Because of this, the data in the diagrams is grouped precisely by platform.

Therefore, all that remains is to say a few words about the environment. As for memory, the fastest one supported by the specification was always used, with the exception of the case that we called “Intel LGA1151 (DDR3)” - processors for LGA1151, but paired with DDR3-1600, and not the faster (and “main” according to specifications) DDR4-2133. The amount of memory has always been the same - 8 GB. System storage () is the same for all subjects. As for the video part, everything has already been said above: this article used exclusively data obtained with the built-in video core. Accordingly, those processors that do not have it are automatically sent to the next part of the results.

Testing methodology

The technique is described in detail. Here we will briefly inform you that the main ones for the results are two “modules” out of four standard ones: and . As for gaming performance, it, as has been demonstrated more than once, is mainly determined by the video card used, so, first of all, these applications are relevant specifically for GPU tests, and discrete ones at that. For serious gaming application Discrete video cards are still needed, and if for some reason you have to limit yourself to IGP, then you will have to take a responsible approach to choosing and configuring the game for a specific system. On the other hand, our “Integral Game Result” is quite suitable for quickly assessing the capabilities of integrated graphics (first of all, this is quality, not quantification), so we will give it too.

Let's make detailed results of all tests available in the form. Directly in the articles, we use relative results, divided into groups and normalized relative to the reference system (as last year, a laptop based on Core i5-3317U with 4 GB of memory and a 128 GB SSD). The same approach is used when testing laptops and other ready-made systems, so that all results in different articles (of course, using the same version of the technique) can be compared, despite the different environment.

Working with video content

This group of applications traditionally gravitates towards multi-core processors. But when comparing formally identical models from different years of production, it is clearly visible that the quality of the cores is no less important here than their quantity, and the functionality (primarily) of the integrated GPU is also important here. However, fans maximum performance“there’s still nothing to please us with: AMD has never played in this market (even the company’s plans include the most fast processors IGP will be deprived), and for Intel these are solutions for LGA115x, where the performance per thread and clock frequency gradually increases with the platform number, but while maintaining the formula “four cores - eight threads”, and the frequencies cannot be said to be increasing very actively. In the end Core comparison The i7-3770 and Core i7-6700K give us a 25% increase in performance over five years: the same notorious “5% per year” that people tend to complain about. On the other hand, in the Pentium G4520/G2130 pair the difference is already quite significant 40%, and the new models of these processors for LGA1151 have acquired support for Hyper-Threading, so they behave like the Core i3-6100 with all that it implies. In the field of nettop-tablet solutions, there is still room for intensive methods of increasing performance, which is brilliantly demonstrated by the Celeron J3455, which is already outperforming some fully desktop processors. In general, progress in different market segments comes with at different speeds, but the reasons for this have long been and repeatedly voiced: desktop computers have ceased to be the main purpose, and the times when it was necessary to increase productivity at any cost, since it was, in principle, not enough to solve the problems of mass users, also ended in the last decade. There are, of course, server platforms, but (again, unlike the situation at the end of the last century), this has long been a separate area, where considerable attention is also paid to efficiency, and not just performance.

Digital Photo Processing

We continue to observe similar trends, adjusted for the fact that Photoshop, for example, has only partial multi-threaded optimization. But some of the filters used actively use new sets of commands, so to some extent one compensates for the other in the case of budget desktop processors, but not “atomic” » platforms. In general, there is an increase in performance over a long time interval, and with a certain devaluation of old processor families (Core i7 for LGA1155 is approximately Core i5 for LGA1151), but the global “breakthroughs” that some “potential buyers” have been dreaming of have been around for a long time not anymore. Perhaps they are not there because changes generally occur only in the Intel assortment, and even those are planned :)

Vector graphics

We abandoned the use of Adobe Illustrator in the new version of the methodology, and the final diagram clearly shows the reason for this decision: the last thing this program was seriously optimized for was Core 2 Duo, so for work (note: this is not a household application, and it is very expensive) A modern Celeron or a five-year-old Pentium is quite enough, but even if you pay seven times more, you can only get one and a half times the speedup. In general, at least in in this case performance is interesting to many, there is no point in testing it - in such a narrow range it is easier to assume that all colas are the same:) The only “in-flight” solutions are “atomic” solutions - it’s not for nothing that it was said about them for 10 years in a row that they are intended for consuming content, and not for producing it.

Audio processing

Adobe Audition is another program that, starting this year, is leaving the list of those we use in testing. The main complaint against it is the same: the “required level of performance” is achieved too quickly, and the “maximum” differs too little from it. Although the difference between Celeron and Core i7 in each iteration of LGA115x is approximately twofold, it is easy to see that most of it is still “made up” within, if not budget, then inexpensive processor lines. Moreover, the above is true only for Intel processors- to today's AMD platforms The application is generally somewhat biased.

Text recognition

The times of rapid progress in character recognition technologies are long gone, so the corresponding applications are developed without changing the basic algorithms: they, as a rule, are integer and do not use new instruction sets, but they scale well in terms of the number of computational threads. The second provides a good spread of values ​​within the platform - up to three times, which is close to the maximum possible (after all, the effect of code parallelization is usually not linear). The first does not allow us to notice a significant difference between processors of different generations of the same architecture - a maximum of 20 percent over five years, which is even less than the “average”. But processors of different architectures behave differently, so this application continues to be an interesting tool.

Archiving and unarchiving data

Archivers have also, in principle, reached such a level of productivity that in practice you can no longer pay attention to their speed. On the other hand, they are good because they quickly respond to changes in performance characteristics within the same processor family. But comparing different ones is a dangerous task: the fastest among those we tested (of those included in today’s article, of course) was the Core i7-4970K for an already formally “outdated” platform. And not everything is going smoothly in the “atomic” family either.

File operations

The diagram clearly shows why, from 2017, these tests will no longer be taken into account in the overall score and will “go” into their own: with the same fast drive, the results are too even. In principle, this could have been assumed a priori, but it didn’t hurt to check. Moreover, as we see, the results are smooth, but not perfectly smooth: “surrogate” solutions, low-end mobile processors and old AMD APUs do not squeeze the maximum out of the SSD used. In their case, SATA600 is supported, so no one seems to be stopping you from copying data at least at the same speed as “adult” platforms, but there is a decrease in performance. More precisely, it was until recently, but now it ceases to matter.

Scientific calculations

Questions regularly arose in the forum regarding the use of SolidWorks Flow Simulation for testing low-cost systems, but in general the results of this program are quite interesting: as we see, it scales well across cores, but only across “physical” ones - various SMT implementations are contraindicated for it. From a methodological point of view, the case is interesting, but not unique; while most of the programs in our set are, if they are multi-threaded, then fully multi-threaded. But overall, the results of this scenario fit into the overall picture.

iXBT Application Benchmark 2016

So, what do we have in the bottom line? Mobile processors are still a thing in themselves: they have the same performance as desktop processors, but of lower classes. There is nothing unexpected in this - but their energy consumption is significantly lower. The performance increase between similarly positioned desktop Intel processors over five years is 20-30%, and the more “top-end” the family, the slower it grew. This, however, does not in any way interfere with “social justice”: it is precisely in the budget segment that higher performance is needed, as well as more powerful graphics (there may simply not be enough money for discrete ones). In general, thrifty buyers are lucky - one might say that the primary focus on laptop computers has also contributed to budget desktops. And not only in performance and purchase price, but also in cost of ownership.

In any case, this is true for Intel solutions - the second remaining manufacturer of x86 processors on the market has been doing worse in recent years, to put it mildly. FM1 is a five-year-old solution, FM2+ until the end of 2016 remained the company’s most modern and powerful integrated platform, but they differ... literally by the same 20% as different generations of Core i7. However, it cannot be said that nothing has changed at all over the past years: the graphics have become more powerful, and energy efficiency has increased, but gaming has remained the main niche of these processors. Moreover, for graphics performance at the level of low-end discrete video cards, you have to pay with both low performance of the processor part and high energy consumption - which is what we are just moving on to.

Energy consumption and energy efficiency

In principle, the diagram clearly explains why budget processors “grow” in speed faster than “non-budget” ones: power consumption is more limited than, generally speaking, necessary for desktop computers(although this is better than the horrors of the 90s and zeros), the relative share of “full-size desktops” has also decreased greatly over the past years and continues to fall. And for laptops or tablets, even older “atomic” models are no longer very comfortable - not to mention quad-core Core ones. Which, in a good way, is long overdue to be made a main mass product - you see, the software industry will find useful use for such power.

Let us note that not only the efficiency increased, but first of all, energy efficiency increased, since more modern processors spend less energy to solve any problem in the same or even less time. Moreover, working quickly is useful: in energy saving mode you will be able to stay longer. Let us recall that these technologies began to be actively used in mobile processors- when did such a division even exist, because now all processors are like this to a certain extent. AMD has the same trend, but in this case the company failed to repeat the success of at least Sandy Bridge, as a result of which the most “tasty” market segments were lost. Let's hope that the release of processors and APUs based on a new microarchitecture and a new technical process will solve this problem.

iXBT Game Benchmark 2016

As stated in the description of the methodology, we will limit ourselves to a qualitative assessment. At the same time, let's recall its essence: if the system demonstrates a result above 30 FPS at a resolution of 1366x768, it receives one point, and for the same thing at a resolution of 1920x1080 - two more points. Thus, given that we have 13 games, the maximum score could be 39 points - it does not mean that the system is gaming, but such a system at least copes with 100% of our gaming tests. It is by the maximum result that we will standardize all the others: we calculated the points, multiplied by 100, divided by 39 - this will be the “Integral game result”. For really gaming systems, it is not needed, since everyone there is more interested in the nuances, but for assessing “universal” ones it will do just fine. It turned out to be more than 50 - which means that sometimes you can play something more or less comfortably; about 30 - even lowering the resolution will not help; Well, if 10-20 points (not to mention zero), then it’s better not to even mention games with more or less 3D graphics.

As we can see, with this approach everything is simple: only AMD APUs for FM2+ (most likely FM2) or any Intel processors with fourth-level cache (with eDRAM) can be considered “conditionally gaming” solutions. The latter are faster, but quite specific: firstly, they are quite expensive (it’s easier to buy an inexpensive processor and a discrete video card, which will provide greater comfort in games), secondly, most of them have a BGA design, so they are sold only in components of ready-made systems. AMD, on the other hand, plays on a different field - its desktop A8/A10 are practically no alternative if you need to build a computer that is more or less suitable for games, but has a minimal cost.

Others Intel solutions, as well as younger (A4/A6) and/or outdated AMD APUs, it is better not to consider them as gaming solutions at all. This does not mean that their owner will have absolutely nothing to play - but the entire range of available games will also include either old or applications that are undemanding in terms of graphics performance. Or both at once. For other things, they will have to purchase at least an inexpensive discrete video card - but not the cheapest, since “low-end” solutions (as has been shown more than once in relevant reviews) are comparable to the best integrated solutions, that is, money will be wasted.

Total

In principle, we made the main conclusions about processor families directly in their reviews, so they are not required in this article - this is primarily a generalization of all previously obtained information, nothing more. More precisely, almost all of them - as mentioned above, we have postponed some systems for a separate article, but there will be fewer of them there, and the systems will be less widespread. The main segment is here. In any case, if we talk about desktop systems, which now come in different designs.

Generally speaking, last year, of course, the processor events were quite poor: both Intel and AMD in the mass market continued to sell what debuted in 2015, or even earlier. As a result, many participants in these and last year’s results turned out to be the same - especially since we tested the “historical” platforms once again (we hope that for the last time :)) But the slowest last year was the Celeron N3150: 54.6 points, and the fastest - Core i7-6700K: 258.4 points. In this regard, the positions did not change, and the results actually remained the same - 53.5 and 251.2 points. The top-end system had it even worse :) Note: this is despite a significant reworking of the software used, and precisely in the direction of the most demanding tasks on the computer’s performance. The budget “old man” represented by the Pentium G2130, on the contrary, grew from 109 to 115 points over the year, just as the “non-budget old man” Core i7-3770 began to look even a little more attractive than before after a software update. On this, in fact, the idea of ​​acquiring “productivity for the future” can be closed - if someone has not already done this;)

Almost every year a new generation of Intel Xeon E5 central processors enters the market. Each generation alternates the socket and process. There are more and more nuclei, and heat generation is gradually decreasing. But a natural question arises: “What does the new architecture give to the end user?”

To do this, I decided to test the performance of similar processors of different generations. I decided to compare models from the mass segment: 8-core processors 2660, 2670, 2640V2, 2650V2, 2630V3 and 2620V4. Testing with such a generational spread is not entirely fair, because Between V2 and V3 there is a different chipset, a new generation of memory with a higher frequency, and most importantly, there are no direct peers in frequency among the models of all 4 generations. But, in any case, this study will help to understand to what extent the performance of new processors has increased in real applications and synthetic tests.

The selected line of processors has many similar parameters: the same number of cores and threads, 20 MB SmartCache, 8 GT/s QPI (except 2640V2) and the number of PCI-E lanes is 40.

To assess the feasibility of testing all processors, I turned to the results of the PassMark tests.

Below is a summary graph of the results:

Since the frequency is significantly different, it is not entirely correct to compare the results. But despite this, conclusions immediately arise:

1. 2660 is equivalent in performance to 2620V4
2. 2670 is superior in performance to 2620V4 (obviously due to frequency)
3. 2640V2 sags, and 2650V2 beats everyone (also due to frequency)

I divided the result by frequency and got a certain performance value at 1 GHz:

Here the results are more interesting and clear:

1. 2660 and 2670 - an unexpected turnaround for me within one generation, 2670 is justified only by the fact that its overall performance is very high
2. 2640V2 and 2650V2 - a very strange low result, which is worse than the 2660
3. 2630V3 and 2620V4 are the only logical growth (apparently due to new architecture...)

After analyzing the result, I decided to weed out some of the uninteresting models that are of no value for further testing:

1. 2640V2 and 2650V2 - an intermediate generation, and not very successful, in my opinion - I’m removing them from the candidates
2. 2630V3 is an excellent result, but it costs unreasonably more than 2620V4, given the similar performance and, moreover, this is the outgoing generation of processors
3. 2620V4 - reasonable price (compared to 2630V3), high performance and, most importantly, this is the only model of the latest generation 8-core processor with Hyper-threading on our list, so we definitely leave it for further tests
4. 2660 and 2670 - an excellent result in comparison with 2620V4. In my opinion, it is the comparison of the first and last (at the moment) generations in the Intel Xeon E5 line that is of particular interest. In addition, we still have sufficient stocks of first-generation processors in our warehouse, so this comparison is very relevant for us.

The cost of servers based on 2660 and 2620V4 processors can differ by almost 2 times, not in favor of the latter, so by comparing their performance and choosing a server on V1 processors, you can significantly reduce the budget for purchasing a new server. But I will tell you about this proposal after the test results.

For testing, 3 stands were assembled:

1. 2 x Xeon E5-2660, 8 x 8Gb DDR3 ECC REG 1333, Intel SSD Enterprise 150Gb
2. 2 x Xeon E5-2670, 8 x 8Gb DDR3 ECC REG 1333, SSD Intel Enterprise 150Gb
3. 2 x Xeon E5-2620V4, 8 x 8Gb DDR4 ECC REG 2133, SSD Intel Enterprise 150Gb

PassMark PerformanceTest 9.0

A more detailed testing report allows us to draw some conclusions:

1. Mathematics, incl. and floating point, mainly depends on frequency. The difference of 100 MHz allowed the 2660 to outpace the 2620V4 in computational operations, encryption and compression (and this despite the significant difference in memory frequency)
2. Physics and calculations using extended instructions are performed better on the new architecture, despite the low frequency
3. And, of course, the test using memory was in favor of V4 processors, since in this case different generations of memory were competing - DDR4 and DDR3.

It was synthetic. Let's see what specialized benchmarks and real applications show.

Archiver 7ZIP

Here the results have something in common with the previous test - a direct link to the processor frequency. It doesn’t matter that slower memory is installed - V1 processors confidently take the lead in frequency.

CINEBENCH R15

The Xeon E5-2670 pulled up the frequency and beat the 2620V4. But the E5-2660, which has a not so visible advantage in frequency, lost to the 4th generation processor. Hence the conclusion - this software uses useful additions of the new architecture (although perhaps it’s all a matter of memory...), but not so much that this is a decisive factor.

3DS MAX + V-Ray

The results were similar to CINEBENCH: the Xeon E5-2670 showed the lowest rendering time, and the 2660 could not beat the 2620V4.

1C: SQL/File

When testing a database with file access, the E5-2620V4 processor confidently leads. The table shows the average values ​​of 20 runs of the same test. The difference between the results of each stand in the case of a file database was no more than 2%.

Single thread test SQL databases showed very strange results. The difference turned out to be insignificant, given the different frequencies of the 2660 and 2670, and the different frequencies of DDR3 and DDR4. There was an attempt to optimize the SQL settings, but the results turned out to be worse than they were, so I decided to test all the stands on the basic settings.

The results of the multi-threaded SQL test turned out to be even more strange and contradictory. The maximum speed of 1 thread in MB/s was equivalent to the performance index in the previous single-threaded test.

The next parameter was the maximum speed (of all streams) - the result was almost identical for all stands. Since the results of different runs fluctuated greatly (+-5%) - sometimes they were at different stands with a significant gap in one direction or the other. The same average multi-threaded SQL test results lead me to 3 thoughts:

1. This situation is caused by an unoptimized SQL configuration
2. The SSD became a system bottleneck and did not allow the processors to overclock
3. There is almost no difference between the frequency of memory and processors for these tasks (which is extremely unlikely)

The result for the “Recommended number of users” parameter also turned out to be inexplicable. The average result of 2660 turned out to be the highest - and this despite the low results of all tests.
I will also be glad to see your comments on this issue.

Conclusions

Of course, I’m not saying that there is no difference at all between the processors, I just want to note that for certain applications there is no point in a “planned” transition to a new generation.

If you do not agree with me or you have suggestions for testing, the stands have not yet been dismantled, and I will be happy to test your tasks.

Economic benefit

Below is an approximate calculation.

Pros

Cons

Under Intel is better programs and games optimized

Lower power consumption

Performance tends to be slightly better

Higher cache frequencies

Work effectively with no more than two resource-intensive tasks

Higher cost

When the line of processors changes, the socket also changes, which means the upgrade is more complicated

Lower cost

Better price/performance ratio

Work better with 3-4 resource-intensive tasks (better multitasking)

Most processors overclock well

Higher power consumption and temperatures (not entirely true regarding latest processors Ryzen)

Worse program optimization