Two times two: AMD Phenom II X2 and Athlon II X2 processors. AMD V, Athlon II, Turion II, Phenom II mobile processors: reference information Comparison of phenom 2 and athlon processors

As the well-known Russian proverb says, "the hand washes the hand." This proverb is very well applicable to various figures in the IT industry. And that's why. Manufacturers of software and hardware actively interact with each other so that the people, that is, you and I, change their pieces of iron for something fresher as much as possible and as often as possible. And each time, as a result of such joint work, marketers from different companies declare the need and revolutionary nature of new technologies. A reasonable question arises: what about those who do not want or cannot pay fabulous money for a new top-end hardware? Do you really have to sit on "Celerons" or "Semprons" all your life? No! You won't have to. AMD presented to the public a couple of its new developments - processors AMD Athlon II X4 630 and A MD Athlon II X4 620. New items compare favorably with their "phenomenal" brothers extremely attractive price. On the example of Athlon II X4, I propose to figure out what we get as a result from AMD.

First of all, let's look not only at the processor itself, but also at its specifications in comparison with other models of modern AMD processors:

So, after analyzing the presented table, let's try to set the vector of this material. Let's set ourselves a series of questions, the answers to which will help you navigate when choosing a new processor:

1. How well do the new processors overclock compared to the older Phenom? The question is perhaps the most relevant, since, in our opinion, a processor with such an attractive price (the recommended retail price of AMD Athlon II X4 620 is around $ 120) is aimed primarily at overclockers, and don’t feed them (that is, you) with bread , let me just disperse.
2. Which is better - 4 inexpensive cores without L3 cache or 3 slightly more expensive cores with 6MB L3 cache? Based on the characteristics of both CPUs, it is impossible to make an unambiguous conclusion in favor of one or another solution. You need to carefully look at the test results.

First of all, let's finally get acquainted with the processor itself and the hardware that will act as a test bench:

The latest version of CPU-Z 1.55.2 at the time of testing identified the processor and accurately showed its technical characteristics. The following components were used as a test platform:

• Processors: AMD Athlon II X4 620, AMD Phenom II X3 720
• System CPU cooling: ProModz Cooled Silence Extreme Pack.
• Motherboard: Gigabyte MA-790GP-DS4
• RAM: Kingston HyperX 2x2048MB DDR-2 1066 @ 5-5-5-15
• Video card: AMD Radeon HD 4870 512MB (PowerColor)
• Hard drive: Western Digital Raptor 36GB (SATA, 10,000 rpm)
• Power supply: IKONIK Vulcan 1000 Watt
  
• operating system: Microsoft Windows Vista x64 SP1 Home Basic Edition

Overclocking

When you start overclocking your system, you must remember that overclocking one of the system components is not enough. The maximum performance gain in both 2D and 3D tests is possible only if all system components are overclocked (wisely :)). AMD Athlon II X4 620, unlike its older counterparts Phenom II X4 Black Edition, has a multiplier blocked for increase, and this automatically introduces a number of additional factors that can prevent successful overclocking of the system. For those who are just learning how to overclock computers, we list the factors on which overclocking depends:

• The operating temperature of the processor and other components of the system (directly depends on the cooling systems and TDP CPU);
  
• Operating voltage of CPU, RAM, NB/SB of motherboard, etc. (determined by the BIOS, software and user's hands)
  
• Sometimes the increase in the frequency of one of the components of the system depends on the overclocking ability of another component (for example, the absence of step-down dividers for memory can limit CPU overclocking in case the "RAM" cannot conquer a higher frequency)
  

Let's take a look at the third item on our list. Successful overclocking and maximum results with the AMD Athlon II X4 620 processor can only be obtained if your motherboard overclocks well on the bus and the RAM scales well in frequency (in this regard, using DDR-3 makes the task much easier).

Beginning practical tests new processor for overclocking, we did not expect that the test motherboard would actually start to behave inappropriately from scratch. As we noted above, for successful overclocking of the AMD Athlon II X4 620, it is necessary that " M o B o" allowed us to work stably at high bus frequencies (preferably up to 300 MHz). In the process of searching for the ceiling of our motherboard from Gigabyte, after another reboot... The open stand became a source of an unpleasant smell of "scorched iron" and the system did not start! managed to subdue before this incident, amounted to 270 MHz.Thank God that we tested the new processor in nominal value BEFORE overclocking the system.It's unpleasant, but there is no other choice.Today we will NOT show the results of overclocking the new budget quad-core from AMD.Nevertheless, we will try supplement this material with tests of this processor on another motherboard, which, in addition, will work with DDR-3 memory.It is on it that the CPU will be re-overclocked.

Since nothing good came of today with overclocking, let's see what happened in nominal terms. Let's move on to the evaluation of the AMD Athlon II X4 620 test results.

EVEREST Ultimate Version 5.0

Judging by the data received, the Phenom II X3 lost the battle in this test. The additional amount of L3 cache "spoiled" the results in the tests of the memory subsystem, while in other tests AMD Athlon II X4 620 took the lead due to the "extra" core. 1:0 in favor of the newbie :)

Super Pi 1M

Since the Super Pi is a single-threaded test, the presence of a fourth core inside the AMD Athlon II X4 62 does not affect the final score in any way. But the Phenom II X3 720 makes full use of its trump card in the form of a 6 MB L3 cache and a 200 MHz higher nominal frequency.

WinRAR 3.90x64

WinRAR, unlike the synthetic Super Pi, is the most "real" application. IN latest versions This archiver implements not only multithreading support, but also support for the x86-64 command set. Judging by the results, the presence of an additional core did not allow the AMD Athlon II X4 620 to overtake the Phenom II X3 720, which took the lead in this test due to the higher core frequency and L3 cache.

Cinebench R10x64

AMD Athlon II X4 620 renders faster in multi-threaded mode. Higher clock speed and L3 cache don't give Phenom II X3 720 the required boost when the test actively uses 4 threads. But in single-threaded mode, the novelty is slightly inferior to the Phenom processor. To tell the truth, AMD Athlon II X4 620 deserves to win in this test, because now hardly anyone will be doing the final rendering on a single core.

Old discounted vs new cheap

We have repeatedly mentioned the sale of processors of previous generations organized by AMD. So "more than once", that there was a reason to think: why do we not have the exact results of any of the two Phenom II X4, which in the current conditions look almost the best deals in the budget market? Yes, of course, we have already tested the extreme ones in the 910 and 980 families, and it is not difficult to estimate the performance of any intermediate model (including 955 or 965) using approximation, but many readers are simply too lazy to deal with it. And besides: approximation by two points is an extremely unreliable thing. It is desirable to add a third, which we recently did for a couple of Athlon II families, and now we will deal with Phenom II.

But there will be no completely new AMD processors in testing. But from Intel we will take a couple of models that appeared not so long ago, which, however, are also included in long-studied families. In a word, today we have the usual routine testing of five processors on our agenda. Not for the purpose of any scientific discoveries, but to clarify the information already available.

Test stand configuration

So, three AMD Phenom II processors. As for the 955, everything has been said above - its wholesale price since autumn is only $81, so this processor is very competitive until the old stocks are exhausted. More precisely, other models in this price class are not very competitive, except perhaps for the no less "sale" A6-3670K, where a weaker processor part is compensated by good graphics. But the buyer of a discrete video card is not interested in it, which makes Phenom II X4 955 practically uncontested within the AMD assortment. Intel has only dual-core Pentiums for this money - older models, of course, but even the older Pentium is just a Pentium: two computing threads are no longer enough for many modern applications (including gaming ones). But more than four is not necessary.

We need one more processor, namely Phenom II X6 1075T, primarily for the reason mentioned above (but there are others, which are discussed below) - this is the third approximation point for Phenom II X6. And Phenom II X4 960T is interesting in itself. The processor is based on, in fact, the same Thuban, but two cores in Zosma are initially blocked. As a result, this OEM model was once extremely popular among those who like to take risks: in case of success, a cheaper Phenom II X6 was obtained than if one was purchased initially. True, the probability of success was far from 100%, this processor penetrated retail in small quantities, and inexpensive six-core processors (such as 1035T / 1055T) greatly undermined the idea of ​​\u200b\u200bsaving money - why risk it because of some 50 dollars? In fairness, our copy unlocked without any problems - just changing one item in UEFI Setup was enough. But that there are no problems at all - we still won't say: the processor was not tested in this mode. Yes, this is not too interesting: unlocking a couple of cores turns the 960T into practically complete analogue 1075T - only the frequency in turbo mode is 100 MHz lower. But its performance in normal mode is very interesting to us: a priori, we can assume that when all four cores are loaded, it should be slightly lower than that of 955, and in low-threaded applications - at the level of 965. In any case, this is how the frequencies of these processors correlate. Let's see how practice confirms theory. And AMD's six-core nature in itself is now rarely of practical importance, be it innate or "unlocked": processors based on Thuban in Lately AMD is only nominally present in the assortment, and it is extremely difficult to find them in retail. And the model range has not been updated for a long time, so having the results of three models (previously tested 1035T and 1100T and today's 1075T), you can determine the performance of any other with a fairly high accuracy using approximation by clock frequencies.

Initially, we did not plan to include previously tested processors in the list of today's participants, but we decided to make an exception for the Pentium G2120. For two reasons. Firstly, two other Intel processors are not direct competitors of Phenom II X4 955 in today's conditions, but Pentium somehow can. Secondly, on this moment this is the youngest Ivy Bridge "by far", so it's interesting to compare it with the younger Core i3 and the younger Core i5 on the same architecture. As for the i3-3220, there is nothing special about it - we have already tested its older brother (3240), and these processors differ only clock frequency, and only at 100 MHz.

The release of the Core i5-3330 was somewhat unexpected. It would seem that the lower price bar was clearly fixed at $184 in bulk in the summer - when the Core i5-3470 replaced the older i5-3450 on it. And then suddenly Intel releases as many as three cheaper Core i5! Model 3350P does not raise any special questions - as you can see from the index, the video core is blocked here. Most likely, this is just the utilization of a “complete marriage” in the field of the video part. But for just $177 wholesale in both OEM and retail packages, plus a TDP of 69W, it's a great deal for those looking to go with discrete graphics. That is, first of all, of course, for small assemblers of finished systems, but also for individual buyers $ 18 (the difference between the “boxed” versions of the 3350P and 3470) will not be superfluous. Everything is clear with the 3330S - it is supplied only through OEM channels and costs $ 7 cheaper than the 3470S: just a little, but for a large batch of monoblocks or compact desktops (where processors with a TDP of 65 W are used), the savings can be significant . But the Core i5-3330 ... It's not clear - for whom? The "boxed" version is only $8 cheaper than the 3470, the OEM version is $2 (two!) cheaper. At the same time, the processors differ only in frequency, but the "floor" for the 3470 (3.2 GHz without a turbo, which in practice will be a rare occurrence, since the processor can overclock to 3.4 GHz even with a load on all four cores) is the "ceiling" for 3330 (there, this frequency is only achieved in turbo mode, and not more than at half load). Yes, and the maximum frequency of the video core is reduced by 50 MHz - to the level of Core i3 / Pentium.

In a word, an incomprehensible processor. The only explanation is a retail (fortunately, “box” prices coincide) replacement Core line i5-23xx, which it was decided to "shoot" entirely. We wouldn't buy it for ourselves :) But for testing, of course, the processor is interesting. First, because it is the youngest quad-core Ivy Bridge. Secondly, this is another processor with a nominal frequency of 3.0 GHz and a turbo mode, i.e. formally the same as the Phenom II X4 960T and X6 1075T. Its maximum frequency, however, is the minimum (sorry for the pun) in this trio, but the architecture is the most modern. C Pentium G2120 and Core i3-3220, again, it is interesting to compare it.

As we have already warned more than once, we have not yet used the ability of Ivy Bridge to work with DDR3-1600 in the main line of tests. However, increasing the memory frequency does almost nothing for the top Core i7-3770K (when using a discrete graphics card, of course), so it would be difficult to expect record yields in relation to the Core i5, i3, or, especially, the Pentium (recently, we received for representatives this class of processors is only 2% on average from replacing DDR3-1066 with DDR3-1333, but a further transition to DDR3-1600 will not give that much). However, in tests according to the next version of the test method (the transition to which is not far off), we will no longer "level" the environment for processors under LGA1155, but for now we will keep today's practice unchanged (otherwise we would have to re-test a considerable number of already studied processors of the Ivy Bridge family).

Testing

Traditionally, we divide all tests into a number of groups and show the average result for a group of tests/applications on the diagrams (for details on the testing methodology, see a separate article). The results in the diagrams are given in points, for 100 points the performance of the reference test system, the site of the sample of 2011, is taken. It is based on the AMD Athlon II X4 620 processor, but the amount of memory (8 GB) and video card () are standard for all tests of the “main line” and can only be changed as part of special studies. Those who are interested in more detailed information are again traditionally invited to download a table in Microsoft Excel format, in which all the results are shown both in converted points and in "natural" form.

Interactive work in 3D packages

As expected, the 960T was slightly faster than the 955, but slower than the 1075T, a low-threaded group of tests where Turbo Core technology can unleash its full potential. However, as we can see, this “power” itself is not enough - Intel processors with such or even slightly lower frequencies are much faster. And what is also kept in a dense group is understandable - as we have already installed Hyper-Threading in this group, it only interferes, and additional "honest" cores are simply not needed.

Final rendering of 3D scenes

These subtests are already able to load any reasonable number of computational threads, so the Phenom II X6 1075T almost caught up with the Core i5-3330. Achievement? Not very - average six-core processor almost caught up with the junior quad. Well, with such initial data, quad-core models are naturally capable of acting on equal terms only against two cores with Hyper-Threading. And the only thing that saves the situation here is that the second one is more expensive. And for the same money, Intel offers only two conventional cores, which are significantly slower.

From a less global one - as expected, with such a load, the 955 is slightly faster than the 960T: Turbo Core does not work when the cores are fully loaded.

Packing and unpacking

Multithreading support is only present in one of the four subtests, so the 960T is slightly faster than the 955 and both fall behind the Pentium G2120. But the 1075T is able to compete with the Core i3-3220 - in general, this is also a rather ridiculous comparison :)

Audio encoding

By the type of load, this group of tests is similar to rendering, so the results are the same. Not too joyful for Phenom II - X4 is capable, of course, of overtaking ordinary dual-core processors, but such processors are found only among budget products. But “two cores, four threads” at comparable clock speeds is no worse in performance than four “real” old-style cores. Well, six of those, of course, are hardly able to argue with four more modern ones. Yes, we remember that the 1075T is not the oldest Phenom II X6, but there were two models faster than it. And the Core i5-3330 is the slowest of the desktop quad-core Ivy Bridges.

Compilation

Compiler tests have always been a strong point of Phenom, but at the moment their victory here is starting to turn into a purely nominal one: yes, somewhat faster, but whom faster? A couple of years ago, the same 1075T easily overtook the fastest Core i5, and the Phenom II X4 kept at a comparable level to the last. Compare this with the current state of affairs.

Mathematical and engineering calculations

You can do without detailed comments - as you can see, these types of loads have a bad effect on Intel processors(since Pentium, Core i3 and Core i5 "hang out" at the same level despite different prices), and for Phenom II they are like death in general (since the comparison with Pentium will be politically incorrect here).

Raster graphics

There is some multi-threaded optimization in terms of programs, but it only allows you to line up Intel processors in the correct sequence and allows Phenom II X6 to overtake X4. That's all - two practically non-intersecting worlds.

Vector graphics

Two threads is enough, which leads to a certain chaos in the product range under LGA1155, but Phenom does not help much. The difference between the three models taken today is completely determined by the Turbo Core (or the lack of this technology in the 955) and does not allow any of them to fully compete with the older Pentiums. However, we note once again - the younger Core i5 also has difficulty in doing this, which is why Intel has to artificially restrain the frequencies of dual-core budget models: there is a lot of software similar to these two programs on the market.

Video encoding

On the one hand, there is room for multi-core processors to develop, on the other hand, as we have said more than once (including most recently), the number of cores for video codecs is an important, but not the only, parameter of processors. Accordingly, all that Phenom II X4 955 and 960T managed to do was to overtake "simple" dual-core processors, and Phenom II X6 1075T was enough to compete with dual-core, but also four-threaded processors. Again, we recall that a couple of years ago everything looked completely different: in video encoding, only Core i7 could cope with X6, and X4 performed on an equal footing with older Core i5. Now - everything is different. Because AMD has all the same processors as then, while Intel has the old family names only :)

Office software

And again the same! Nothing unexpected, of course - most of the tests in this group are generally single-threaded. Just another illustration of the fact that you need to be very careful when choosing processors by the number of cores - not all of them will necessarily be used by the software. And selecting software “for multi-core” is a simple task only for testers: there are a lot of “inconvenient” applications among the popular ones. As if not even the majority - if by "popular" we mean massively used.

Java

But in some specific niches, the oldies, of course, perform well. Relatively good - in comparison with other applications, and not at all in absolute results. From their point of view, as we said above, the victories of the average six-core processor over the junior quad-core one or the once good quad-core ones over the Core i3 at best do not cause much optimism.

Games

As we have said more than once, modern games require four threads of computation in all cases where the bottleneck is not the video card. However, as we see, in the "general and whole" fast dual core processor(Pentium type) is quite capable of keeping up with slow quad-core ones (Phenom II type). If you look at the detailed results, it is noticeable that some applications still "like" the latter a little more. But there is no talk of any unambiguous superiority. Here, with the same architecture, we can say for sure that four cores are better than two in games (and any - even “flavored” Hyper-Threading, not to mention “ordinary”), but with different ones, anything can happen.

Multitasking environment

As we have said more than once, there is no exclusivity in the results of the test with the simultaneous launch of several programs - we just simulated another multi-threaded application. And the result is appropriate: the junior quad-core Phenom II X4 is 25% faster than the dual-core Pentium, but is approximately equal to the Core i3, and the average six-core Phenom II X6 1075T is just a little ahead of the junior third-generation Core i5. Such effective cores in the Ivy Bridge family turn out to win not by numbers, but by skill.

Total

Here, in fact, is the answer to the question why the Phenom II X4 955 is at the Pentium level. Yes, because its performance is on average at the same level! No miracles, which so many economical buyers hope for - the price of each item is determined by how much it can be sold for. And for processors, the latter depends on performance and power consumption. Can the 955 now cost more than $100, as it did in the summer? Of course not - for that kind of money there are already more attractive offers. But for "about 100" - already a very good processor, capable (with a multi-threaded load) to compete with the Core i3. But, note, not with Core i5, where the same four cores - quantity does not always translate into quality. So it is precisely this (and not at all concern for the low-income segments of the population) that explains the price reductions. And the disappearance of Thuban from retail chains with the formal continuation of supplies is also the same: for market success, all six-core AMD models (including top ones) should cost no more than $150, and the company has neither the desire nor the opportunity to produce them with such initial data ( if you remember the crystal size of 346 mm² - more than two (!) times more than the quad-core Ivy Bridge). Of course, somewhere in specific areas of application, multi-core Phenom II still look very good, but no less often (and just in widely demanded mass-purpose applications) they "dry" lose to budget Intel processors. Here are the developments on the new microarchitecture (both APUs and updated ones) - a much less sad sight, while the "classic" Athlon and Phenom have definitely reached a dead end.

So for assembly new system Phenom II, despite the price reduction, is not of much interest (except in the case of a "crazy programmer" who compiles something 24 hours a day, producing electricity using a personal windmill). However, there are users who can win thanks to the ongoing "sale": Phenom II X4 955 and 965 are great for upgrading the system on some Athlon II, not to mention older AMD processors (the latter, of course, only if you have technical feasibility). Especially the "hundred-bucks upgrade" will be of interest to owners of large amounts of DDR2 memory: so what if the performance is far from the maximum on the market - but this is the only way not to change the memory along with the processor, and system board. AMD is also aware of this. And do not mind (despite the established reputation of Robin Hood - the defender of the poor and oppressed) to earn extra money on it: only 955 and 965 have fallen in price, but for slightly faster models they ask for 140-160 dollars.

However, since all currently sold Phenom II X4 belong to the Black Edition family, ways to deal with this injustice have long been known. Yes, yes: the cobblestone dispersal is the tool of the proletariat. AMD's unwillingness to cut prices for the Phenom II X6 can be defeated in the same way: Phenom II X4 960T can still be found on sale, and (if you have a suitable motherboard) you can also unlock a couple of cores for it. There is, of course, a risk that it will not work out, but the end result, in our opinion, is worth the risk. Moreover, in case of failure, you will get a processor with a performance similar to Phenom II X4 955, which, considering the minimal difference in the price of these processors, is quite normal. But if everything goes well, you will get an almost complete analogue of the Phenom II X6 1075T. Not only much more expensive, but also in a different performance class.

In any case, one should not forget that all the advantages of multi-core Phenom II can be experienced in practice only if there are a large number of programs optimized for multi-threaded processors among the constantly used applications. If there is no confidence in this, then there is not much point in four or six cores either. One or two computing threads - the realm of Pentium, in which these processors are able to easily compete on an equal footing with Core i3 / i5, not to mention Phenom II. And the video part in them is noticeably better than in old (technologically; it doesn't matter what is still sold) integrated AMD chipsets, and the power consumption of such models is noticeably lower.

However, a sale is always a good thing, as there are ways to take advantage of it. As well as the phased transition of LGA1155 processors to Ivy Bridge is also good: they are better than their predecessors, which, in general, will be noticeable to all their customers. Although this transition sometimes goes in strange ways, sometimes giving rise to very strange models, such as the Core i5-3330. Until recently, the 2320 of the previous generation remained nominally the cheapest Core i5, and now Intel has apparently decided to make a replacement for it (and, by the way, a little faster than the i5-2400). But here practical implementation let us down: compared to the 3470, the processor is too slow, and the real retail prices of these models in Moscow often differ by only 100 rubles, or even less. 2320 or older 2310 allow (if you look well) to save 300 rubles that way, which is much more interesting when money is in the first place. In general, why he was born like this - we absolutely do not know. On the other hand, its presence on sale, in general, does not interfere with anyone, and it can be useful for assemblers of ready-made systems. The main thing is not to buy inadvertently. Why, in fact, we did not spare the time to test it: forewarned means forearmed.

AMD removed the X2, X3, and X4 core count suffixes from the logo, instead changing the item number: the 9000 models have four cores, while the upcoming tri-core models will have the number 7000.

It's been a tough year for AMD. Not only the Phenom processor, which everyone has been waiting for so long, came out at significantly lower clock speeds (2.3 GHz instead of 3 GHz), but an unpleasant error was revealed in the current stepping of the Barcelona core. It is possible to bypass it, but only the updated stepping will allow AMD to continue producing quad-core processors for the server segment. And the fact that AMD's quad-core processor doesn't have enough performance to compete with Intel in the high-end segment doesn't help either. As a result of all these problems, AMD had to change its product strategy and position the processor, along with the new Spider platform, to the mass market. However, despite all the problems, Phenom is not as bad as many believe that you will see on this comparison between Phenom and Athlon 64 X2.

In fact, AMD has quite a few significant advantages over Intel when it comes to upgrading current systems to a quad-core processor. If Intel is very quick to release new platforms for each new generation of processors due to changing requirements, then AMD has not changed the Socket AM2 specifications at all. Therefore, it is technically possible to install a quad-core Phenom processor on a motherboard. socket board AM2, replacing the Athlon 64 or Athlon 64 X2, all you need is a BIOS update. However, this is also not always true - some motherboards can't handle Phenom power consumption (95W or 125W), but most enthusiast motherboards can be upgraded to a quad-core processor. At least in the future, because at the moment we have only been able to install Phenom on two "old" motherboards out of ten .

The upgrade situation does require some attention as AMD and Intel are planning their next major technology upgrade in about six months. AMD will introduce Socket AM3, which will support DDR3 memory, and Intel's next generation processors, codenamed Nehalem, will finally bring the memory controller to the processor. Given all this, even the upcoming Core 2 Duo E8000 or Core 2 Quad Q9000 lines can only be considered as intermediate products on the way to the next generation, even if they overtake the existing Core 2 products by about 10%.

November 17 AMD launched two Phenom models on the market A: Phenom 9500 and 9600, at 2.2 and 2.3 GHz, respectively. They both have a TDP of 95W, which is close to the 105W reported by Intel for the Core 2 Quad Q6600 (2.4GHz) and Q6700 (2.66GHz). All faster models scheduled for release in the first quarter of 2008 will run on a 125W TDP. Toward the end of 2008, a Black Edition may appear, which is friendly to overclockers, but not higher than the top frequency of 2.3 GHz. But AMD has unlocked the multiplier to provide ideal overclocking conditions, and this version should not be more expensive than usual.

You will be able to install a Phenom processor in almost any Socket AM2 motherboard on the market when all problems will be solved. Even cheap motherboards support the standard 95W TDP, but for the 125W versions you need to use an enthusiast platform, which is true if you plan to significantly overclock the Phenom. The BIOS update situation is far from ideal, so installing Phenom on existing Athlon boards is not as easy as AMD promised. Technically, this is the same socket with a 1000 MHz HyperTransport link, but there are problems.

The Phenom microarchitecture is known under the code name K10, but then it was renamed Stars. The most significant difference, which mainly affected the number of transistors, is the L3 cache, which is an extension to AMD64's two-level cache design. While each compute core has its own L1 cache for data and instructions (64 KB each), as well as 512 KB of L2 cache, L3 provides an additional 2 MB of fast storage for all Phenom cores.

This is not the first desktop processor to come with L3 cache: 3.2-GHz, 3.4-GHz and 3.46-GHz models Intel Pentium The 4 Extreme Editions, all of which were built on the 130 nm Gallatin core, also included 2 MB of L3 cache (along with 512 KB of L2 cache). But, unlike the Pentium 4 EE L3 cache, the Phenom L3 cache works as a buffer for writing data to RAM.

AMD has also made some improvements to the branch prediction process, since the so-called sideband stack optimizer updates the ESP (enhanced stack pointer) without consuming CPU time. And the memory prefetcher is able to load data exclusively into the L1 cache, bypassing the L2 cache (that is, without unloading data from there). Note also the 128-bit SSE computation width, as well as the 32-byte instruction fetch block. AMD's virtualization technology has been around for months, and it's included with every Phenom processor.

Support for 1.8GHz HyperTransport 3.0 protocol is the latest performance enhancement feature that has been added to the Phenom. While HT 2.0 at 1.0 GHz supports 8.0 GB/s in both directions, HT 3.0 delivers up to 20.8 GB/s. This will be especially important in the future, when four or more cores need to be able to access other cores, for example, to get data from memory or to work with a PCI Express device such as a video card.

We were quite intrigued by AMD's claim that the Phenom is 25% faster per clock than the current Athlon processors 64x2. Given that there are no architectural revolutions like the one that Intel made when they switched from NerBurst to Core, a 25% increase in performance per clock is very significant. It is sometimes even hard to believe in it, which is why it was interesting for us to take a closer look at the new processor. We compared the Athlon 64 X2 and the Phenom 9900 at a base clock speed of 2.6 GHz using just one core.

All Phenoms look similar: here is our engineered sample with an unlocked multiplier.

Closing the circle of "historical testing", today we will deal with a platform that formally remains alive and well, although ideologically even older than the previously reviewed AMD FM1 and Intel LGA1156. How does she do it? We have already dealt with this issue: Socket AM3 + 2011 is practically no different from the “just” AM3 2009, which was obtained by switching from DDR2 to DDR3 from AM2 / AM2 + from 2006, and these, in turn, are practically nothing more than Socket 939 from summer 2004, but with DDR2 rather than "plain" DDR. However, it would be more correct to speak even about 2003, when Socket 940 appeared: Socket 939 is its simplification, without support for multiprocessor configurations. During this time, not only memory standards, of course, but also some other interfaces have changed, but conceptually, in the form of AM3 +, we have a classic platform of the zero years - three-chip and a relatively low degree of integration. It is also worth noting that the latest microarchitectural updates of processors produced for it date back to the end of 2012, i.e. from this point of view, even the latest modification of AM3 + is already history (to the same extent as LGA1155, for example). However, within other platforms, AMD ships no more than two-module processors (supporting, respectively, only four computing threads) with a significant bias towards integrated graphics, so that the most productive AMD processors are still AM3 + devices. They have not been updated for a long time, but their final obsolescence is planned only for the second half of this year - in connection with the transition to a single (finally!) AM4 socket, for which both high-performance processors without integrated graphics and relatively low-end ones with such will be produced. It is easy to see that this is not yet an analogue of LGA1155 and subsequent Intel platforms- rather, a repetition of LGA1156, because when choosing fast processor"In the load" will have to use a discrete graphics card. But it's still much better than what happened to the company's range of the last five years, when the various FMx and the same long-outdated AM3+ were simply incompatible with each other.

How did the company manage to keep the AM3+ afloat without upgrading the processors? Yes, very simply: due to the price. We had to forget about competition for lovers of high performance anyway, but for about the same money a buyer can buy either an eight-core FX-8350/8370 or a four-core Core i5-6400. Yes, of course, the comparison of prices in this case is not entirely correct, since it does not take into account other features of the platforms and, first of all, the ability to save on a video card in the case of an Intel platform. However, if you still need to purchase a video accelerator (for example, when you are interested in games - we have and continue to adhere to the opinion that a full-fledged gaming computer without a discrete video card is still impossible), this problem disappears. And at first glance, it doesn’t matter that the same FX-8350 appeared back in 2012: advertising in its case generally talks about eight cores (forgetting to clarify that these are somewhat different cores than in other processor architectures, even by AMD itself), that . e. gives the impression of a processor, which in the performance of Intel costs piecebucks. This is the right approach, the wrong one, but it works. And how - it is useful to check. In the end, as mentioned above, this year we will finally be able to get acquainted with new AMD processors - so in any case they will have to be compared with the old ones. So today we will create an "information reserve" on old and even very old processors, since such an opportunity presented itself.

Test stand configuration

There will be two main characters. The FX-8370 processor is relatively new - it appeared at the end of 2014, but differs from the FX-8350 (the firstborn of the Vishera family) only in the turbo mode clock speed. Note that formally the top representatives of the family are the FX-9370 and FX-9590, but the latter exist only formally: a TDP of 220 W not only scares many people off by itself, it also leads to compatibility problems with many motherboards, and also a thoughtful approach to the choice of cooling system. Well, if all this does not scare you, then you should not forget that any processors of the FX family have fully unlocked multipliers, allowing for arbitrarily fine tuning - including frequency. This, by the way, is another reason that the platform still has a certain popularity among those users who do not care about the result - the main thing is the process itself. Which in this case is also facilitated by a huge crystal of a processor manufactured according to the 32 nm process technology - it is very easy to provide such a heat sink (sometimes disadvantages can become advantages). Moreover, equipping "box" processors with updated coolers allows you to count on good results even in such a variant, which may also turn out to be cheaper than the "traditional" approach with an OEM processor and some kind of "supercooler". In general, for the limited in means enthusiast the platform is interesting, despite its archaism.

But since testing this platform is still a digression into history, we decided to use a new methodology (including the study of energy consumption issues) to test even more old cpu belonging to the Phenom II X6 family. Until the release of the first FX in 2011 - the top in the company's range. What's more, it's forever. The best decision for old boards with "regular" AM3 and even AM2+. Moreover, as our tests showed, the use of DDR3 is not so necessary for Phenom II family processors, so we won’t be surprised if such systems continue to be used somewhere (after all, Pentium D owners regularly run through the Conference - until now :)). The top-end 1100T would suit us best, but there was none, and the existing 1075T, alas, is not a Black Edition, so it does not turn into an older model in the correct way. However, even with the possibility of overclocking by a multiplier, it is still unknown how correct it is from the point of view of measuring power consumption, and the line itself is so old (2010!) that, as it seems to us, there is no big difference anymore - test 1100T or 1075T . Therefore, there will be a second one - since it exists.

With whom will we compare? It was not for nothing that we mentioned the Core i5-6400 above - the junior quad-core of the modern Intel line directly competes in price with the older AMD models (taking into account, of course, a remark about the video card). According to some readers, last time it was necessary to compare it with solutions for LGA1156, and not having a close price and performance, but still a dual-core Core i3-6320. Therefore, today we will add to the list of subjects and best processor for the mentioned platform, namely the Core i7-880, since the first FX were created, among other things, to compete with them. Unfortunately, however, they came out later than it was necessary to ensure that - already in the days of processors for LGA1155. One of these models (albeit already the third, and not the second Generation Core) has been tested by us at the moment - we will add it to the list of subjects for the sake of completeness. And, at the same time, the fastest Athlon X4 for FM2+ - for the masses. Moreover, for fans of AMD products, these are also direct competitors to some extent: the FX-8370 is certainly “cooler”, but it is also more expensive. Yes, and plus an archaic platform. And among the tested ones, we recall, there is Phenom II X6 1075T, so it will be interesting to see how six, but old cores, compare with modern, but two modules. It's clear that four is more interesting, but the transition from Phenom II (not necessarily six-core) will be simple and inexpensive only if you have a board with AM3+. If there is only AM2 +, then change everything anyway. But if, for example, some Athlon II is installed on such a board, the performance of which is already not enough, the question - to find Phenom II on the secondary market or change the platform, is not at all idle.

As for the other test conditions, all test subjects worked in a system with a discrete graphics card based on the Radeon R9 380 and 16 GB of RAM. The type and frequency of the latter were the maximum supported processors - for all, except for the Phenom II X6 1075T, which we tested with DDR3-1600, which does not cause problems (however, it hardly affects performance either).

Test Methodology

The technique is described in detail in a separate article. Here we briefly recall that it is based on the following four pillars:

• Methodology for measuring power consumption when testing processors
• Methodology for monitoring power, temperature and processor load during testing

And the detailed results of all tests are available in the form of a complete table with the results (in Microsoft Excel 97-2003 format). Directly in the articles, we use already processed data. In particular, this applies to application tests, where everything is 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) and grouped by the areas of application of the computer.

iXBT Application Benchmark 2016

As you can see, if the modular architecture had appeared in 2010, its “life” would have been significantly simplified: a couple of modules are no longer inferior to the Core i5 of that time, and four can convincingly surpass even the quad-core Core i7. But, unfortunately (or fortunately), in 2011, when developing processors for the LGA1155, Intel managed to significantly improve all the characteristics of its products, and so dramatically that since then such "feats" have not been observed for five years. As a result, the older FX had to be positioned not in the segment between i5 and i7, but at the level of the former. So their price is quite consistent with the performance, but nothing more. Moreover, it is clearly seen that the company had no other options - the transfer of Phenom to a thinner production process was unlikely to significantly "spur" them: in order to bypass six old cores, two modules are often enough, and not three or four .

Especially when software it cannot always fully utilize a large number of computation threads, but it is demanding on their quality - including support for modern instruction sets and so on. As a result, even the older FX are now lagging behind the younger Core i5, but it could be worse - as Phenom showed us. Actually, as has been said more than once, intensive architecture improvements usually have their effect not at all in those generations of processors in which they are implemented. But the further - the more important.

But here - nothing matters: there would be one fast stream. In such conditions (which is not a secret), AMD processors have a hard time, but it's easy to see that they had a chance to be the fastest on the market in 2010.

But in this case - and hypothetical was not. However, judging by the small difference between FX and Phenom (and not even the older one), it is clear that no one was involved in optimizing such work scenarios at all: anyway, the performance for those times was not bad.

As we have already written more than once, relatively old integer code is the best that can be encountered in the life of AMD modular processors. And it is clearly seen that, in general, they were developed for such applications: after all, six-core Phenom II in 2010 could no longer compete with quad-core Core i7 in such tasks, but for quad-module FX it was a feasible task. Unfortunately, at the end of 2011 (when the first processors of this family finally appeared physically) it became much more complicated.

Actually, an aria from the same opera - as we have already noted, data packaging is similar to text recognition in terms of the logic of work. And the results too.

The obvious outsider here is the Core i7-880, but simply because LGA1156 only supported SATA300. As we have already noted, in order for the difference to become generally noticeable, it is necessary to use fast SSD with which there were difficulties in those years. Now it's gone, so it's not much, but it does. But AMD endowed its chipsets with support for the new interface even then, so in this case there were no rough edges at all.

As we have mentioned more than once, various SMT technologies are “alien” to the program, but the number of “hardware” cores and their quality are relevant, which, for example, results in the fact that the modern younger Core i5 is faster than the old Core i7. And even not so fundamentally old - not only 880, but also 3770 was left behind. The first one also lagged behind the FX-8370, which is a common thing. And here are six very old architectural cores in Phenom II... They can overtake two modules of modern AMD processors, but with great difficulty - they can't cope with three.

What do we have in general? The FX-8370 is about 1.5 times faster than the Athlon X4 880K - a normal gain by doubling the cores and adding L3 cache. But, unfortunately, this is not enough to compete with modern processors Intel that equal prices do not fully compensate. If only because the buyer of the Core i5-6400 can do without a discrete graphics card, but the one who chooses FX cannot. But if he still plans to buy it, it turns out something close to parity - until now. True, prices are not its cause, but rather a consequence - it is not for nothing that they have been declining all the years.

Why the situation turned out to be exactly like this - in principle, the results can also be assumed. We do not know exactly what years the main part of the development of the modular architecture fell on, but we can assume that it was earlier than 2011 - after all, it was then (and after several delays) that the first processors for AM3 + already began to be sold. Had this happened a year earlier, when quad-core processors such as the Core i7-870/880 cost around three to five hundred dollars, the effect would have been noticeable - comparable to the release of the first Athlons. At the same time, dual-module processors (including models with an integrated GPU) would be suitable for replacing a quad-core Phenom or Core 2 Quad, while three-module ones would look fine against the background of Phenom II X6 (or instead of them) and Core i5. But in the end, the processors had to compete not with models for LGA1366 or LGA1156, but with the brand new (at that time) LGA1155, which is still not bad against the background of newer Intel platforms. Which, however, have become even better, and the old FX have been living on the market without major changes since 2012. Which has to be compensated for by prices that were at first between the Core i5 and i7, then at the level of the older i5, then the middle ones, and now the younger ones. Since the consumer characteristics of processors approximately correspond to such prices. Only here Core i5 are very cheap processors for production, and FX are expensive. So it's time to break this vicious circle - the further, the more difficult it is. Let's hope everything works out this year.

Energy consumption and energy efficiency

However, with regard to energy consumption, even in those years it was not all smooth sailing, but from the point of view of modernity, 200 W is very frightening. It is clear that this includes what "passed" through the board to power the video card - but it is the same for everyone. But the “gluttony” of the three-chip platform is its pure feature and “greetings from the 2000s”: modern ones are much more economical. However, if you pay attention to the actual needs of the processor, then it also reached 140 W, i.e. for AMD, exceeding the TDP level is just a common thing (although some people still try to scold Intel for this in the old fashioned way). But the Phenom II X6 looks better at first glance. But do not forget that this is not the oldest model of the line, firstly, and that power consumption makes sense only in conjunction with performance, secondly.

And from this point of view, the modular architecture was a clear step forward. We also note that FX behave better than Athlon - if only because the shared L3 cache (which is not available in processors for FM2/FM2+) has a positive effect on performance, but is not too gluttonous. The truth takes up a lot of space, which is why its implementation in processors with integrated GPUs turned out to be impossible. But in general, it becomes clear why the company did not make the FX shrink for the 28 nm process technology: in the APU it allowed to increase the graphics power, but the processor cores would not give anything or almost nothing. And the wake-up call “rang the alarm” five years ago: Intel managed to achieve the performance level of 45-nanometer processors, but at the cost of excessive power consumption (who said "NetBurst"?). And then the situation only worsened.

iXBT Game Benchmark 2016

And can these processors work well in a gaming computer? Generally speaking, yes - after all, the main load falls on the video card. But how many possibilities of the latter will be "disappeared" because of the processor? This question is especially not idle, by the way, for users of boards with AM2 + or "regular" AM3, where Phenom II X4 / X6 is the best available without changing the platform, and the once popular Athlon II with so. modern times, they don’t “pull” anything at all.

The case when "single-threaded performance" is critical, which puts everything AMD processors into an uncomfortable position. The performance of even the (already) inexpensive R9 380 is held back by all the test subjects. But you can play with comfort on all the same.

And here everyone copes close to the maximum possible. And by the way, pay attention - the old Phenom II is noticeably better than the new Athlon.

It's worse here, however, again, Phenom II is no worse than any Core 2 Quad or Core i5/i7. And FX are already able to "wrestle" with newer i5 / i7.

But in the newer game of the series, Phenom II is kept on an equal footing (already on an equal footing) only with Athlon. Which, however, is quite enough for practical use - but it could be better. At least at the level of FX, which in FHD already allows the selected video card to "give it its all" to the fullest.

And here everyone is about the same - there are differences only in the mode with reduced resolution. And, funny, they are more in favor of AM3 + than vice versa.

When everything is determined by the video card, processors from five or six years ago are also good. The most powerful of them, of course. But they cost a little later began to be very cheap.

FX is behaving well, Phenom II time has expired, alas. On the other hand, if such a processor already exists, then it is not at all necessary to change it in a gaming computer - there will be no noticeable effect. It's better to put a more powerful video card.

Here Thief clearly "votes" for powerful platforms - and considers only the modern range of Intel as such. From one side. On the other hand, it cannot be said that something does not work at all. There are about 40 frames - if you want to save money on changing the platform, this can be considered sufficient.

Here in this pair, the dependence of the frame rate on the performance of processors is already there. But, in fact, so what? The absolute results of all the test subjects are more than sufficient for a comfortable game. So in the end we come to the conclusion that for an inexpensive gaming computer"The old oak will still make noise." Naturally, if it is already there (or can be purchased very cheaply). And, of course, given the fact that even for budget modern video cards, such a processor can be a "limiting factor." Not in the sense that it will not be possible to play, but in the fact that the performance, nevertheless, will be lower than potentially possible. But even this will not always happen.

Total

In principle, we didn’t get anything unusual in the end - the platform is formally “live” and up-to-date, but in fact it has not been updated for a long time. Whether updates are needed or not is a matter of debate. Some, for example, do not like that Intel is constantly upgrading something, almost without changing the performance of processors. On the other hand, for the same money, performance is constantly (albeit slowly) growing, and the need to change platforms is primarily due to their functionality. As a result, some top motherboard from five years ago, for example, looks dull and pale against the background of even the most budgetary modern proposals, the price of which is five times lower. If nothing is touched, then the performance will not grow, and otherwise the characteristics of the computer will remain typical of five to seven years ago. Another question is that in many cases this is quite enough, and in the case of a reasonable pricing policy, "historical" platforms are quite suitable for practical application, until they physically disappear from service, which will obviously happen even later than the end of sales.

After the breakthrough of the beginning of the "zero" AMD safely returned to its usual state of always catching up and, despite the rather interesting and, no doubt, advanced technical solutions, does not even try to compete with Intel in terms of sales.

As of mid-2009, the company accounts for about 14.5% of the microprocessor market.
At the same time, once branded "chips" of AMD chips - for example, 64-bit instruction extensions or a RAM controller built into the processor - have long been used in the chips of the main competitor.

AMD products today occupy two very narrow niches: ultra-budget processors for building economy-class computers and high-performance models offered three to five times cheaper than comparable Intel chips.

This explains the fact that you can find AMD processors of various families and generations on store shelves - from prehistoric Sempron and Athlon based on the well-deserved K8 architecture for Socket 939 to cutting-edge six-core Phenom II X6.

Be that as it may, AMD is now betting on the K10 architecture, so we will talk about processors based on it.
These include the Phenom and Phenom II, as well as their budget variant, the self-consciously named Athlon II.

Historically, the first K10-based chips were the quad-core Phenom X4 (codenamed Agena), released in November 2007.
A little later, in April 2008, the tri-core Phenom X3 appeared - the world's first central processors for desktop computers, in which three nuclei are located on one crystal.

In December 2008, with the transition to a 45-nanometer process technology, the updated Phenom II family was introduced, and in February the chips received a new Socket AM3 connector.
The serial production of the quad-core Phenom II X4 began in January 2009, the tri-core Phenom II X3 - in February 2009, the dual-core Phenom II X2 - in June 2009, and the six-core Phenom II X2 - literally just now, in April 2010.

The Athlon II, a modern replacement for the Sempron, is a Phenom II that lacks one of its most important virtues - a large L3 cache shared by all cores.
Available in dual, triple and quad versions.
The Athlon II X2 has been in production since June 2009, the X4 since September 2009, and the X3 since November 2009.

AMD K10 architecture

What are the fundamental differences between the K10 and K8 architectures?
First of all, in the K10 processors, all cores are made on the same chip and are equipped with a dedicated L2 cache.
Phenom/Phenom 2 chips and server Opterons also have L3 cache memory shared by all cores, the volume of which is from 2 to 6 MB.

The second major benefit of the K10 is the new HyperTransport 3.0 system bus with a peak bandwidth of up to 41.6 GB/s in both directions in 32-bit mode or up to 10.4 GB/s in one direction in 16-bit mode and up to 2, 6 GHz.
Recall that the maximum operating frequency previous version HyperTransport 2.0 is 1.4 GHz and peak throughput- up to 22.4 or 5.6 GB / s.

A wide tire is especially important for multi-core processors, while HyperTransport 3.0 provides the ability to configure the channel, which allows each core to provide its own independent line.
In addition, the K10 processor is able to dynamically change the bus width and operating frequency in proportion to the natural frequency.

At the same time, it should be noted that at present, the HyperTransport 3.0 bus in AMD chips operates at a much lower speed than the maximum allowed.
Three modes apply depending on the model: 1.6 GHz and 6.4 GB/s, 1.8 GHz and 7.2 GB/s, and 2 GHz and 8.0 GB/s.
The manufactured chips do not yet use two more modes laid down in the standard - 2.4 GHz and 9.6 GB / s and 2.6 GHz and 10.4 GB / s.

K10 processors integrate two independent RAM controllers, which speeds up access to modules in real-life conditions.
The controllers are capable of working with DDR2-1066 memory (models for AM2+ and AM3 socket) or DDR3 (chips for AM3 socket).

Since the controller integrated into the Phenom II and Athlon II for Socket AM3 supports both types of RAM, and the AM3 socket is backwards compatible with AM2+, the new CPUs can be installed on older AM2+ boards and work with DDR2 memory.

This means that when buying a Phenom II for an upgrade, you do not have to immediately change the motherboard, and also purchase a different type of RAM - as, for example, is the case with Intel i3/i5/i7 chips.

K10 architecture microprocessors feature a range of upgraded power-saving technologies - AMD Cool'n'Quiet, CoolCore, Independent Dynamic Core, and Dual Dynamic Power Management.

This sophisticated system automatically reduces the power consumption of the entire chip in idle mode, provides independent power management for the memory controller and cores, and is able to turn off unused processor elements.

Finally, the cores themselves have also been significantly improved.
The design of fetch blocks, branch and branch prediction, and scheduling was redesigned, which made it possible to optimize the kernel load and, ultimately, increase performance.

The bit width of SSE blocks was increased from 64 to 128 bits, it became possible to execute 64-bit instructions as one, support for two additional SSE4a instructions was added (not to be confused with the SSE4.1 and 4.2 instruction sets in Intel Core processors).

Here it is necessary to mention a design defect found in server Opterons (codenamed Barcelona) and in Phenom X4 and X3 of the first releases - the so-called "TLB error", which at one time led to a complete cessation of supply of all Opterons of revision B2.
In very rare cases, under high load, a design flaw in the L3 Cache TLD block could cause the system to become unstable and unpredictable.

The defect was considered critical for server systems, which is why the shipment of all released Opterons was suspended.
For desktop Phenom, a special patch was released that disables the defective unit using BIOS tools, but at the same time, processor performance dropped noticeably.
With the transition to revision B3, the problem was completely eliminated, and such chips have not been found on sale for a long time.