At the beginning of the year, on January 8, AMD introduced the new AMD Dragon platform, based on the processor of the new AMD Phenom II family. Initially, AMD demonstrated only two processors of this family: AMD Phenom II X4 940 and AMD Phenom II X4 920, which are compatible with the AM2+ socket and support DDR2 memory. Later, processors from the AMD Phenom II family were introduced, compatible with socket AM3 and supporting both DDR2 and DDR3 memory. In this article we will look at the test results of the new AMD Phenom II family of processors.
Lineup of AMD Phenom II family processors
The main difference between the new AMD Phenom II family of processors and the AMD Phenom family of processors is that they are manufactured using a 45-nm process technology using SOI technology, while the AMD Phenom family of processors are manufactured using a 65-nm process technology.
Just like the AMD Phenom family of processors, they are true multi-core processors, that is, all processor cores are implemented on a single chip.
Among the innovations implemented in the new AMD Phenom II processors, one can also note the improved AMD Cool’&’Quiet 3.0 technology. It combines a number of functions that reduce processor power consumption at times when it is underloaded, as well as prevent processor overheating.
When announcing the new processor of the AMD Phenom II X4 family, AMD also pointed out other advantages in comparison with the previous family. In particular, it was noted that new processors execute more instructions per clock (Instruction Per Clock, IPC).
The AMD Phenom II processor family currently includes three series: AMD Phenom II X4 900, AMD Phenom II X4 800 and AMD Phenom II X3 700.
AMD Phenom II X4 900 series processors
Currently, the 900 series processors include two quad-core models: the AMD Phenom II X4 940 and the AMD Phenom II X4 920. Each core of the AMD Phenom II X4 900 series processor has a dedicated 512 KB L2 cache that is shared between all L3 cores. 6 MB cache.
The AMD Phenom II X4 940 processor has a clock speed of 3.0 GHz, and the AMD Phenom II X4 920 processor has a clock speed of 2.8 GHz. These processors are equipped with an integrated dual-channel DDR2 memory controller and support DDR2-667/800/1066 memory.
The AMD Phenom II X4 940 and AMD Phenom II X4 920 processors are compatible with Socket AM2+/AM2 and support HyperTransport 3.0 at speeds up to 3600 MHz (two-way) with bandwidth up to 16 GB/s. Both processors have a TDP of 125 W.
The difference between the AMD Phenom II X4 940 and AMD Phenom II X4 920 processor models lies not only in the clock frequency, but also in the fact that the AMD Phenom II X4 940 processor has an unlocked multiplier, which allows for effective overclocking. In general, if we talk about the overclocking potential of the AMD Phenom II X4 940 processor, then, according to independent sources on the Internet, it is quite large. Thus, there is evidence that the use of liquid nitrogen to cool the processor made it possible to achieve a record clock frequency of 6 GHz, and with conventional air cooling this processor can easily be overclocked to 4 GHz.
We also add that the AMD Phenom II X4 910 processor, which will have a clock frequency of 2.6 GHz, is expected to appear soon.
AMD Phenom II X4 800 series processors
At the moment, the 800 series of processors includes only one model of a quad-core processor - the AMD Phenom II X4 810. However, another model is expected to appear soon - the AMD Phenom II X4 805.
The difference between the 800 series processors and the 900 series processors is the reduced size of the L3 cache and the fact that the 800 series processors implement a memory controller that supports both DDR2 and DDR3 memory. In addition, the 800 series processors are compatible with both Socket AM2+/AM2 and Socket AM3.
Each core of the AMD Phenom II X4 810 processor has a dedicated 512 KB L2 cache and a 4 MB shared L3 cache among all cores. The AMD Phenom II X4 810 processor operates at a clock speed of 2.6 GHz. It is equipped with an integrated dual-channel DDR2 memory controller (DDR2-667/800/1066 memory supported) and a DDR3 memory controller (DDR3-800/1066/1333 memory supported). The TDP of the processor is 95 W.
AMD Phenom II X3 700 series processors
Currently, the 700 series processors include two models: AMD Phenom II X3 720 and AMD Phenom II X3 710. All 700 series processors are tri-core. Each core of the AMD Phenom II X4 720 and AMD Phenom II X3 710 processor has a dedicated 512 KB L2 cache, and a 6 MB L3 cache shared between all cores.
Like the 800 series processors, the 700 series processors have an integrated dual-channel DDR2 memory controller (DDR2-667/800/1066 memory is supported) and a DDR3 memory controller (DDR3-800/1066/1333 memory is supported).
The AMD Phenom II X3 720 processor is clocked at 2.8 GHz, and the AMD Phenom II X3 710 processor is clocked at 2.6 GHz. Another difference between the AMD Phenom II X3 720 and the AMD Phenom II X3 710 is that the AMD Phenom II X3 720 is multiplier unlocked and hence can be easily overclocked.
Testing methodology
The processors were tested in two stages. At the first stage, the performance of processors in various applications was determined, and at the second - in various games.
During testing, each test was run five times, rebooting the computer after each test run and maintaining a two-minute pause after the reboot. Based on the results of five test runs, the arithmetic mean and standard deviation were calculated.
The entire testing process was fully automated, for which a special script was used, which sequentially launched all the necessary tests, rebooted, withstood the necessary pauses, etc. In this test script, the following benchmarks and applications were used to determine performance in various applications:
- DivX Converter 6.6.1;
- DivX Codec 6.8.5;
- DivX Player 6.8.2;
- Windows Media Encoder 9.0;
- MainConcept Reference v.1.1;
- VLC media player 0.8.6;
- Lame 4.0 Beta;
- WinRAR 3.8;
- WinZip 11.2;
- Adobe Photoshop CS4;
- Microsoft Excel 2007.
DivX Converter 6.6.1 with DivX Codec 6.8.5 was used to determine performance when converting a source video file to a DivX video file (Home Theater preset in DivX Converter 6.6.1).
Windows Media Encoder 9.0 (WME 9.0) was used to determine performance when converting a video file recorded in WMV format to a video file with a lower resolution and video bitrate.
The MainConcept Reference v.1.1 application (H.264 codec) was used to determine the performance when converting a source video file recorded in the WMV format into a video file with a different resolution and video bitrate (H.264 HDTV 720p preset).
Lame 4.0 Beta was used to measure performance when converting an audio file from WAV to MP3 format.
DivX Player 6.8.2 was used in conjunction with WME 9.0 to create a multi-tasking test. The point of this test was to start the process of converting the same video file using the WME 9.0 application while playing a video file using the DivX Player 6.8.2 application.
Another multitasking test was to play two video files simultaneously using VLC media player 0.8.6 while simultaneously converting another video file using the WME 9.0 application and converting a WAV audio file to MP3 format using the Lame 4.0 Beta application .
WinRAR 3.8 and WinZip 11.2 were used to determine performance when archiving and unzipping large numbers of digital photographs in TIF format. When compressing data using WinRAR 3.8, the maximum degree of compression and encryption using the AES-128 algorithm were used. When archiving using WinZip 11.2, the maximum degree of compression and encryption using the AES-256 algorithm were used.
We used Adobe Photoshop CS4 to determine system performance when processing digital photographs. Our test with the Adobe Photoshop CS4 application is divided into three subtests. In the first of them, we sequentially applied various resource-intensive filters to the same photo, while simulating the process of artistic processing.
The next subtest, using Adobe Photoshop CS4, simulated batch processing of a large number of photographs. In total, the test involved batch processing of 23 photographs in TIF format.
In the third subtest with the Adobe Photoshop CS4 application, batch processing of RAW photographs was simulated.
Microsoft Excel 2007 was used to determine system performance when performing calculations in Excel spreadsheets. We used two tasks in Excel. The first was to recalculate a spreadsheet, and the second was to simulate a Monte Carlo method for probabilistic economic risk assessment.
Please note that the results of all of these tests depend on the performance of the processor, memory and hard drive. However, they practically do not depend in any way on the performance of the video card.
In all of the tests listed, the result is the time it takes to complete the test task, and the shorter it is, the better.
The following games and benchmarks were used to evaluate processor performance in games:
- Quake 4 (Patch 1.42);
- S.T.A.L.K.E.R.: Shadow of Chernobyl (Patch 1.005);
- S.T.A.L.K.E.R.: Clear Sky (Patch 1.007);
- Half-Life 2: Episode 2;
- Crysis v.1.2.1;
- Left4Dead;
- Call of Juares Demo Benchmark v. 1.1.1.0;
- 3DMark06 v. 1.1.0;
- 3DMark Vantage v. 1.0.1.
In Quake 4, S.T.A.L.K.E.R.: Shadow of Chernobyl, S.T.A.L.K.E.R.: Clear Sky, Half-Life 2: Episode 2, Crysis, Left4Dead and Call of Juares Demo Benchmark tests, the result was the number of frames per second (fps) displayed, and in in the 3DMark06 and 3DMark Vantage benchmarks, the results were presented in dimensionless units (3DMark Score).
During testing, each gaming test (with the exception of 3DMark Vantage v. 1.0.1) was launched at screen resolutions of 1280x800, 1440x900, 1680x1050 and 1920x1200 pixels. At each screen resolution, the game tests were run five times, rebooting the computer after each run and maintaining a two-minute pause after the reboot. Benchmark 3DMark Vantage v. 1.0.1 was run five times in each of four presets (Entry, Performance, High and Extreme).
Based on the results of five runs, the arithmetic mean and standard deviation were calculated. The entire testing process was fully automated, for which a special script was used, which sequentially launched all the necessary tests, rebooted the computer, withstood the necessary pauses, etc.
The Crysis game was tested with two demo scenes, one of which served to test the graphics processor, and the other to test the central processor in conjunction with the graphics processor, since when played it affects the physical component of the game engine (both demo scenes are included in the game package).
All games were launched in two setting modes: maximum performance and maximum quality. The maximum performance setting mode was achieved by disabling effects such as anisotropic texture filtering and screen anti-aliasing, as well as setting low image detail, etc. That is, this mode was aimed at getting the highest possible result (maximum FPS value). In this setting mode, the result depends to a greater extent on the performance of the processor and to a lesser extent on the performance of the video card.
The maximum quality setting mode was achieved through the use of high detail, various effects, anisotropic texture filtering and screen anti-aliasing. In this setting mode, the result depends to a greater extent on the performance of the video card and to a lesser extent on the performance of the processor.
When testing computers using the method described above, we traditionally use the concept of integral performance assessment and, accordingly, the concept of a reference PC. The fact is that test results alone do not give an idea of PC performance. Indeed, knowing that the conversion time for a video file is 120 seconds, it is still impossible to draw a conclusion about performance, since it is not clear whether this is a lot or a little. That is, the test results are meaningful only if they can be compared with the results of some reference PC. To compare the performance of the tested and reference PCs, the results were normalized, for which the time it took to complete each test task on the reference PC was divided by the time the same task was performed by the processor under test.
To calculate the integral performance assessment for a set of applications, the normalized test results were divided into six groups: video conversion, audio conversion, multitasking tests, working with archivers, working with Photoshop, working with Excel. Next, in each group of tests, the intermediate integral result was calculated as the geometric mean of the normalized results. After this, the geometric mean of the intermediate integral results for all groups of tests was calculated. For convenience of presenting the results, the obtained value was multiplied by 1000. This is an integral assessment of the computer’s performance on a set of applications. For a reference PC, the integral performance result on a set of applications is equal to 1000 points, and for a tested PC it can be either more or less than 1000 points.
In gaming applications, the integral performance result is also calculated, but the approach in this case is somewhat different. Initially, for each game in each setting mode, the formula calculates the weighted average result for all resolutions.
In this formula, results for different resolutions have different weighting coefficients, with the result for a resolution of 1440x900 having the maximum weighting coefficient.
After this, the geometric mean is calculated between the results determined by the formula described above for the maximum quality mode and maximum productivity. The result found in this way is an integral assessment of PC performance in a separate game.
To obtain an integral performance assessment in the 3DMark Vantage test, the geometric mean between the results for all presets is calculated using the formula.
Next, the integral performance estimates in each individual game are normalized to similar results for the reference PC, and the geometric mean is calculated for all normalized integral results. For convenience of presenting the results, the obtained value is multiplied by 1000. This is an integral assessment of the computer’s performance in games. For the reference PC, the overall performance result in games is 1000 points.
As a reference configuration, we used the most productive (and most expensive) computer at the beginning of 2009. The reference PC configuration was as follows:
- processor - Intel Core i7 Extreme 965 (clock frequency 3.2 GHz);
- motherboard - ASUS RAMPAGE II EXTREME;
- motherboard chipset - Intel X58 Express;
- memory - DDR3-1066 (Qimonda IMSH1GU03A1F1C-10F PC3-8500);
- memory capacity - 3 GB (three modules of 1024 MB each);
- Memory operating mode – DDR3-1333, three-channel mode;
- memory timings - 7-7-7-20;
- video card - two GeForce GTX295 video cards in 4-Way SLI mode;
- video driver - ForceWare 181.20;
Let us note once again that our reference PC is very “sophisticated” - it is the most productive and expensive computer at the moment. That is, the integrated performance results of all other computers should be below 1000 points.
Test bench configuration
We tested three processors of the AMD Phenom II family: AMD Phenom II X4 940, AMD Phenom II X4 810 and AMD Phenom II X4 720. In order to ensure the same testing conditions for all three processors and taking into account the fact that the AMD Phenom II X4 810 and AMD processors Phenom II X4 720 supports both DDR2 and DDR3 memory, and the AMD Phenom II X4 940 processor supports only DDR2 memory; the following bench configuration was used to test the processors:
- motherboard - ASUS M3A78-T;
- motherboard chipset - AMD790GX+SB750;
- memory - DDR2-1066 (A-Data);
- memory capacity - 2 GB (two modules of 1024 MB each);
- memory operating mode - DDR2-1066, dual-channel mode;
- memory timings - 5-5-5-15;
- video card - Zotac GeForce GTX295;
- video driver - ForceWare 182.05;
- hard drive - Intel SSD X25-M (Intel SSDSA2MH080G1GN).
Test results
So, after getting acquainted with the testing methodology and the algorithm for calculating integral performance results in applications and games, you can proceed to announcing the test results.
The table shows the execution time of test tasks in seconds for the tested processors and the reference PC, and in Fig. Figure 1 shows the normalized speeds for completing test tasks. In Fig. 2-20 shows the results of testing processors in gaming applications.
Rice. 1. Normalized speeds for completing test tasks
As can be seen from the test results, in non-gaming applications the performance of AMD Phenom II X4 processors is ranked in the following order: Phenom II X4 940, Phenom II X4 810, Phenom II X3 720. Moreover, the performance of the quad-core Phenom II X4 810 processor is approximately 19% higher than the performance of the triple-core processor the Phenom II X3 720 processor, and the Phenom II X4 940 processor is approximately 15% faster than the Phenom II X4 810 processor and 37% faster than the Phenom II X3 720 processor.
|
|
Rice. 2. Test results |
Rice. 3. Test results |
IntroductionThe position of AMD's products in the processor market at present is clearly not enviable: the new K10 microarchitecture, on which AMD fans had high hopes, although it can be considered effective and original, in reality did not allow the company to create processors capable of resisting Intel's. The strengths of the microarchitecture, the main one of which should be the innate quad-core capability, accompanied by a single level 3 cache for all cores, remained in the shadows due to technological problems that prevent AMD from launching the production of processors with frequencies above 2.5 GHz. As a result, the quad-core Phenom X4 processors that AMD can offer today are uncompetitive not only in the face of the new 45 nm Penryn processors, but even in comparison with the older 65 nm Intel products.
Moreover, the performance gap between the Phenom X4 and Core 2 Quad processors is so large that the prospects for establishing at least parity in performance between these products seem very vague. After all, it is obvious that the 65-nm process technology currently used by AMD will not significantly increase the frequencies of the Phenom. As for the transition to a more advanced 45-nm technology, AMD plans it only for the fourth quarter of this year. However, as expected, the 45-nm Deneb processors, which will replace the 65-nm Phenom, will immediately be able to conquer only frequencies not exceeding 3.0-3.2 GHz. And this, apparently, will not be enough to compete with older quad-core Intel processors, so AMD will have to be content with offering only models that attract primarily with a low price for quite a long time.
Realizing this, AMD is trying to promote the platform concept, promoting not processors themselves, but kits that include a CPU, motherboard and video card. With this approach, the insufficient processor performance can be partially compensated by the good capabilities of the GPU, which is what the company's marketing department is pushing for. However, targeting such kits is more interesting for computer assemblers than for end users, who are accustomed to assembling systems from individual components, matching them to each other based on their own preferences. Therefore, it is not at all surprising that neither the AMD Spider platform, which includes ATI Radeon HD-class discrete graphics, nor the Cartwheel with an integrated AMD 780G chipset arouses much enthusiasm among advanced users.
In such conditions, AMD has to look for other ways to the hearts of customers. The main strategy for the company was to set low prices for its products. Simultaneously with the release of Phenom X4 9x50 series processors based on a new kernel revision, free from the "TLB problem", prices for quad-core CPUs have been reduced in proportion to their performance relative to competitor offerings. As a result, AMD today offers the cheapest quad-core processors, which, given this positioning, can find a number of adherents. Similar metamorphoses are occurring with the dual-core Athlon 64 X2 line, which is miserably inferior in performance to modern Core 2 Duo processors. As a result, retail prices for the Athlon 64 X2 have dropped so much that these processors are now perceived as nothing more than budget offerings.
Reducing prices is a good way to maintain sales levels. But at the same time, the advanced part of the computer community is losing interest in AMD products, and the company is no longer perceived as a technology leader. Therefore, AMD was forced to find another original way to stir up interest in its products. This is today's announcement of a unique family of Phenom X3 processors with a tri-core structure. Of course, one of the reasons for the appearance of such CPUs was the direct economic benefit for the manufacturer, who gets the opportunity to “attach” defective crystals of quad-core Phenoms by disabling one of the cores on them. But on the other hand, the release of the Phenom X3 can also be seen as an attempt to contrast at least something with the Intel Core 2 Duo processors, which are superior to the dual-core Athlon 64 X2 from any point of view. Positioned as a mid-range option between the Athlon 64 X2 and Phenom X4, the triple-core Phenom X3 is priced just right against Intel's mid-range dual-core CPUs.
It is on this basis that we will look at the three-core new products proposed by AMD. Modern software is increasingly geared toward multi-threaded environments, so it's possible that the triple-core Phenom X3 could prove an interesting offering as an alternative to dual-core Intel processors. Fortunately, we won't have to remain in the dark about the practical properties of the new Phenom X3. AMD sent us one of the first retail processors of the new series, which we invite you to review in detail.
Simple arithmetic of a tri-core processor
The new family of triple-core AMD Phenom X3 processors (also known under the code name Toliman) hardly needs a detailed introduction, since, if you look at it, there is nothing new in it. These CPUs are based on the same semiconductor crystals that are used in the quad-core Phenom X4. AMD simply blocks one of the cores in them, gaining the opportunity to sell defective chips that could not become the basis of “full-fledged” processors. The idea of disabling part of a semiconductor chip in order to be able to sell scraps from the production of high-end processors is not new, but until now, both AMD and Intel have only used disabling part of the L2 cache memory.As you know, Phenom X4 processors differ from Intel quad-core CPUs primarily in that they have a monolithic structure, and are not assembled from a pair of dual-core semiconductor crystals. Therefore, the probability of a defect appearing in one of the Phenom X4 cores is quite high; it certainly exceeds the probability of defects appearing in the upper, third level cache. That is why, first of all, AMD decided to release three-core processors, and not to offer cheap quad-core processors without a third-level cache. Here, AMD also played into the hands of the block structure of the Phenom X4 - the cores in it are combined only at the L3 cache level, which makes it possible to remove one core from use without making any changes to the microarchitecture and semiconductor crystal.
A direct comparison of the characteristics of the Phenom X4 and Phenom X3 only strengthens the confidence in the close relationship of these processors.
As a result, Phenom X3 processors turn out to be completely similar to their older quad-core brothers in everything except the number of cores.
Today's announcement contains references to three Phenom X3 models, with frequencies of 2.1, 2.3 and 2.4 GHz. All three processors are based on the new B3 stepping, devoid of the notorious “TLB error”. It should be remembered that AMD also produces Phenom X3 models based on the old B2 stepping, but they are not supplied to the retail market.
To avoid confusion in the enormously expanded range of Phenom processors based on the new K10 microarchitecture, we decided to compile a table that shows all the key characteristics of existing modifications.
Three new triple-core processors are highlighted in the table, which will be the first Phenom X3 distributed through the retail network.
Note that all new Phenom X3 have a heat dissipation level of within 95 W, which means they are potentially compatible with a wide range of Socket AM2/Socket AM2+ motherboards, including those in the lower price category. In fact, to achieve compatibility of new tri-core processors with older boards, only a BIOS update is required.
A little more complicated is the issue of Phenom X3 software compatibility. Since this processor is the first CPU with three cores, it may have to face a number of difficulties caused by some applications' inability to detect and correctly use an odd number of cores. However, these particular problems are unlikely to be widespread. For example, during the tests we did not encounter any obstacles, with the exception of the inoperability of older versions of the SiSoft Sandra diagnostic utility.
However, I would like to draw your attention to a fix that appeared a few days ago for the 32-bit operating systems Windows Server 2008 and Windows Vista, designed to solve problems associated with incorrect determination of the number of available cores. Information about this fix can be found on the Microsoft website. This fix fixes potential errors with detecting the number of cores in tri-core processors, but it is not required - even without it, our test Windows Vista Ultimate found all three processor cores perfectly.
Considering that the Phenom X3 is essentially little different from the Phenom X4, the most interesting thing about the new product is the cost. After quite a long period of hesitation, AMD decided to set the following official prices:
AMD Phenom X3 8750 (2.4 GHz) – $195;
AMD Phenom X3 8650 (2.3 GHz) – $165;
AMD Phenom X3 8450 (2.1 GHz) – $145.
Thus, the triple-core Phenom X3 line is positioned by the manufacturer as something between the quad-core Phenom X4 and dual-core Athlon 64 X2. As a result, the new processors logically fit into the existing structure of AMD's offerings and are in a competitive position relative to the dual-core Intel Core 2 Duo processors of the Wolfdale family, the prices of which were reduced last Monday.
But can the three cores of the Phenom X3 processors compete with the two cores of the Wolfdale? This is the question we will try to answer in our testing. Well, first, let’s take a closer look at the sample of the three-core CPU received by our laboratory.
Phenom X3 8750
The triple-core Phenom X3 8750 processor looks exactly the same as its quad-core brothers. Only the marking gives it away - “HD8750WCJ3BGH”.
Just as the first digit “9” in the model number designation indicates that this is a Phenom X4, AMD chose indices starting with the number “8” to designate tri-core processors. The end of the model number with “50”, as in the case of the Phenom X4, indicates that the processor does not have a TLB error, that is, it belongs to the B3 stepping. The second digit depends on the frequency, and for tri-core and quad-core CPUs this correspondence is the same. In other words, the Phenom X3 8750 shown in the photo is designed to operate at a frequency of 2.4 GHz. This is the oldest model in this line to date.
The processor has three (each core has its own) second-level cache with a capacity of 512 KB and a common 2-MB third-level cache. The northbridge built into the processor operates at 1.8 GHz and provides support for dual-channel DDR2 SDRAM, which can operate in both ganged and unganged modes. Accordingly, the CPU uses the HyperTransport 3.0 bus at 1800 MHz, however, it is nevertheless compatible not only with the new Socket AM2+, but also with older Socket AM2 motherboards.
The Phenom X3's standard voltages are set in the range from 1.05 to 1.25 V. Like their older brothers, the processors support Cool"n"Quiet 2.0 energy-saving technology, which, however, is only available on Socket AM2+ motherboards.
How we tested
As already mentioned, the Phenom X3 series of processors falls into the niche between the Phenom X4 and Athlon 64 X2. Therefore, together with the full Phenom X3 line, we tested the senior representative in the dual-core AMD family and the junior model in the Phenom X4 series.The competitors in testing are dual-core processors of similar cost. After the latest price cuts, these are several junior models of the Core 2 Duo line of the Wolfdale family, including a new product, the Core 2 Duo E7200 processor. In addition, older 65-nm representatives of the Core 2 Duo lineup also took part in the tests.
Below is a detailed description of the test systems.
AMD Platform:
Processors:
AMD Phenom X4 9550 (Socket AM2+, 2.2 GHz, 4 x 512 KB L2, 2 MB L3, Agena);
AMD Phenom X3 8750 (Socket AM2+, 2.4 GHz, 3 x 512 KB L2, 2 MB L3, Toliman);
AMD Phenom X3 8650 (Socket AM2+, 2.3 GHz, 3 x 512 KB L2, 2 MB L3, Toliman);
AMD Phenom X3 8450 (Socket AM2+, 2.1 GHz, 3 x 512 KB L2, 2 MB L3, Toliman);
AMD Athlon 64 X2 6400+ (Socket AM2, 3.2 GHz, 2 x 1 MB L2, Windsor).
Motherboard: ASUS M3A32-MVP Deluxe (Socket AM2+, AMD 790FX).
Memory: 2 GB DDR2-1066 with timings 5-5-5-15-2T (Corsair Dominator TWIN2X2048-10000C5DF).
Intel Platform:
Processors:
Intel Core 2 Duo E8400 (LGA775, 3.0 GHz, 1333 MHz FSB, 6 MB L2, Wolfdale);
Intel Core 2 Duo E8200 (LGA775, 2.66 GHz, 1333 MHz FSB, 6 MB L2, Wolfdale);
Intel Core 2 Duo E7200 (LGA775, 2.53 GHz, 1067 MHz FSB, 3 MB L2, Wolfdale);
Intel Core 2 Duo E6750 (LGA775, 2.66 GHz, 1333 MHz FSB, 4 MB L2, Conroe);
Intel Core 2 Duo E6550 (LGA775, 2.33 GHz, 1333 MHz FSB, 4 MB L2, Conroe).
Motherboard: ASUS P5K3 (LGA775, Intel P35, DDR3 SDRAM).
Memory: 2 GB DDR3-1333 SDRAM with timings 6-6-6-18 (Cell Shock DDR3-1800).
Graphics card: OCZ GeForce 8800GTX (PCI-E x16).
Disk subsystem: Western Digital WD1500AHFD (SATA150).
Operating system: Microsoft Windows Vista x86.
Performance
Overall performance
SYSmark 2007, which we rely on as a test reflecting the integrated performance of processors, demonstrates a rather interesting result. As expected, overall the Phenom X3 is slower than the youngest quad-core AMD processor. However, their performance is not at all higher than the speed of the Athlon 64 X2 6400+, which shows approximately the same result as the Phenom X4 9550. Thus, it turns out that if we draw conclusions based only on the given diagrams, we can say that the existence of a market niche for the Phenom X3 is far-fetched. And these processors may be of interest only in a small number of applications that can load all three cores with work “to the fullest.”
In light of the above, it is not at all surprising that the Phenom X3 is inferior in speed to Core 2 Duo processors, even to the cheapest models E7200 and E6550. It turns out that in a wide range of tasks, with normal, not narrowly targeted use, even three cores with the K10 microarchitecture cannot withstand two cores with the Core microarchitecture. And the main problem of Phenom processors is, obviously, insufficiently high clock speeds.
However, let’s not rush to final conclusions, but let’s see how the new Phenom X3 performs in various types of applications.
3D games
In anticipation of the final graphs, let us remind you that to study processors in games, we specifically use a low resolution of 1024x768. This allows us to focus specifically on the “game” speed of the CPU and abstract from the influence of the GPU on performance - if high resolutions were used, the GPU would become the limiting factor.
The performance situation of the Phenom X3 may differ in different games, but nevertheless, two characteristic types of behavior of these CPUs can be distinguished. In those games where performance does not scale well with more than two processor cores (in other words, those that do not fully support quad-core processors), the Phenom X3 results are unsatisfactory. So, in Quake3, Half-Life 2 Episode Two and, oddly enough, Crysis, the new triple-core processors are inferior to the Athlon 64 X2 6400+, not to mention Intel products.
However, there is another group of gaming applications, including Unreal Tournament 3, World in Conflict and Lost Planet: Extreme Condition. Performance in these games is highly dependent on the number of available processing cores, so the new Phenom X3 doesn't look so bad here. At least they are not inferior to the older Athlon 64 X2, and sometimes they are even able to compete with Core 2 Duo processors. Moreover, not only the previous generation, but also with the new Core 2 Duo E7200.
Media Content Encoding
The state of affairs when encoding media content is entirely determined by the quality of optimization of codecs for multi-core architectures. Apple iTunes, which is well optimized for dual-core processors only, runs significantly faster on Athlon 64 X2 and Core 2 Duo based systems. When using the DivX video codec, which has average optimization for multi-threaded environments, the Phenom X3 processors lag behind the dual-core Athlon 64 X2 6400+, which has a one and a half times higher frequency, only slightly. However, they still seriously fall short of the speed of dual-core Intel processors. But the popular H.264 video codec x264, which brilliantly loads processors with a large number of cores, allows you to fully unleash the potential inherent in the Phenom X3. When testing CPU speed in this codec, the three-core new products not only outperform the Athlon 64 X2, but also demonstrate performance at the level of junior Wolfdale.
Final rendering
The final rendering is just an excellent example of tasks with a well-parallelized load. Therefore, it is not at all surprising that in these tests the Phenom X3 family performs exactly as AMD wanted. The performance of the new tri-core processors clearly falls into the gap between the speed of the younger Phenom X4 and the older Athlon 64 X2. At the same time, the triple-core Phenom X3 quite successfully competes with dual-core Core 2 Duo processors, including their 45-nanometer models. The only pity is that this state of affairs is rather an exception to the general rule.
Other applications
Dual-core processors handle Adobe Photoshop better than the Phenom X3. Although many of the filters in this program can parallelize the workload, the results suggest that AMD's triple-core processors are primarily lacking in clock speed.
Rendering video in Adobe Premiere is similar to 3D rendering. Here the Phenom X3 performs quite well.
Archiving to WinRAR is also faster on the Phenom X3 than on the older Athlon 64 X2. But the Wolfdale Core 2 Duo E8000 series processors, which have a more capacious L2 cache, demonstrate much better results.
The popular computer algebra package works much more efficiently on dual-core processors with Core microarchitecture, although it uses multi-cores very well, as can be seen from the superiority of triple-core AMD processors over the dual-core Athlon 64 X2 6400+.
The results of testing processors in the popular chess program are another consolation for AMD fans. Yes, there are applications in which Phenom X3 processors can perform no worse than the younger Core 2 Duo, and, if desired, a significant number of such programs can be found.
Overclocking
Although the triple-core Phenom X3 processors are based on the same B3 stepping as AMD's quad-core CPUs, their overclocking capabilities should be examined separately. After all, reducing the number of cores working simultaneously entails a reduction in heat generation, which in theory can open up space for obtaining better overclocking results.It should be noted that the Phenom X3 8750 processor we have, like other CPUs in this line, has a fixed multiplier. Therefore, its overclocking should be done by increasing the frequency of the clock generator. This process is not as simple as we would like. The point is that, as explained in article specifically dedicated to this issue, this frequency is associated not only with the resulting clock speed of the processor, but also with the frequencies of the north bridge, memory, and HyperTransport 3.0 bus built into the processor. Therefore, when increasing the frequency of the clock generator, we should not forget about the need to reduce the corresponding coefficients and dividers involved in generating the frequencies of the north bridge, HyperTransport bus and DDR2 SDRAM.
For example, by increasing the processor supply voltage to 1.45 V, we were able to increase the clock generator frequency from the standard 200 to 260 MHz while maintaining processor stability. However, at the same time, the multipliers for the frequencies of the north bridge and the HyperTransport bus had to be reduced from the nominal value of 9x to 7x, which made it possible to keep the corresponding frequencies within limits close to the standard ones.
In this state, overclocked to 3.1 GHz, our Phenom X3 8750 processor demonstrated completely stable operation, which was tested using an hour-long run of the Prime 25.5 utility. To remove heat from the overclocked processor, we used a Scythe Mugen (Infinity) air cooler.
It should be noted that the achieved frequency of 3.1 GHz is the best result of overclocking a processor with K10 microarchitecture, obtained in our laboratory. Thus, we can hope that the Phenom X3 processors are more overclocking friendly than their quad-core counterparts. However, final conclusions can be drawn after obtaining more extensive statistics based on testing more than one CPU instance.
Energy consumption measurement
To complete the picture, we measured the power consumption of systems (without a monitor) built on the processors participating in testing, operating in nominal mode. System configurations were kept the same as in the performance tests. Energy-saving technologies Enhanced Intel SpeedStep and Cool'n'Quiet 2.0 have been activated. The load on the processors was created by the Prime95 25.5 program.
As would be expected, triple-core processors turned out to be more economical than their quad-core relatives due to the smaller number of cores. At the same time, due to the low clock frequency, their power consumption is inferior to that of the dual-core Athlon 64 X2 6400+. However, the Phenom X3 family is completely unable to compete in terms of efficiency with dual-core Intel processors.
conclusions
AMD Phenom X3 is without a doubt a very interesting processor. If only because this is the first CPU in the industry to have a tri-core design and a monolithic design. And, despite the fact that this was the first time we encountered such a non-standard CPU, its use in a familiar hardware and software environment did not create any serious problems. This processor turned out to be fully compatible with the existing infrastructure, which indicates that AMD has chosen the right strategy for eliminating defects in the production of quad-core Phenom X4.As for the consumer qualities and market prospects of the new product, everything is far from so clear. All the main problems of processors with the K10 microarchitecture could not but affect its tri-core carriers - first of all, the Phenom X3 processors, like the Phenom X4, are sorely lacking in clock speed. However, they are still in a slightly more advantageous situation compared to quad-core CPUs, since AMD positions them as competitors to the dual-core Intel Core 2 Duo.
However, a worthy confrontation between Core 2 Duo and Phenom X3 is not always achieved - but only in those applications in which the performance scales well across more than two cores. Unfortunately, there are very few such applications, so in most cases the Phenom X3 loses to Intel processors of similar cost. However, they exist, and these include, in particular, final rendering, individual video processing and encoding tasks, and some others.
Accordingly, we have to admit that another AMD initiative does not have much chance of success. Phenom X3 may be a good niche product, but it will not be widely popular. Low-end Intel processors belonging to the Wolfdale family, having a similar price, offer higher average performance, lower heat dissipation and power consumption, and significantly better overclocking potential. AMD is unlikely to decide to significantly reduce prices for the Phenom X3, since they are based on a monolithic quad-core semiconductor crystal, the production cost of which is relatively high. To be fair, it should be added that if AMD decides to further reduce the cost of the Phenom X3 series, then these CPUs may well become a worthy alternative to the Core 2 Duo E4000 and Pentium Dual Core processors.
To what has been said, it remains to be added that the Phenom X3 cannot always be recommended for upgrading the existing fleet of Socket AM2 systems. The fact is that older dual-core Athlon 64 X2 processors in a number of cases are capable of providing better performance, although at higher heat dissipation.
With the release of the Phenom II family of processors, AMD was able to regain the attention of users, strengthening its significantly weakened position in the processor market. AMD recently switched its CPUs to support DDR3 memory, thereby releasing models with a new design - Socket AM3, which complemented the solutions on the market with socket AM2 and AM2+ that support DDR2. A special feature of the new processors is full compatibility with motherboards equipped with the AM2+ socket, which made it possible for many users to upgrade at minimal financial cost without replacing their motherboard.
The main advantage of boards for Socket AM3 lies in the support for faster DDR3 memory, which in itself makes these solutions more relevant and modern. On the other hand, it is known that due to higher latency, the advantages of low-frequency DDR3 memory modules over regular DDR2 tend to zero. At the moment, in terms of price, approximately parity has been established between memory of different standards, with the exception of high-frequency “overclocking” DDR3 kits, the cost of which is by no means affordable. A pair of sticks designed for frequencies of 1600 MHz and higher are still more expensive than an equivalent set of older DDR2 operating at 1066 MHz. And the cost of motherboards with a progressive Socket AM3 connector is higher than their analogues for AM2+ processors.
Despite the price factor, users are still looking at the new type of memory, and it becomes interesting to look at the dependence of the performance of AMD processors at different memory frequencies and timings. To do this, we will compare the three-core and quad-core Phenom II processors at operating memory frequencies from 800 MHz (DDR2) to 1600 MHz (DDR3), which will make it possible to identify not only differences in performance between the AM2+ and AM3 platforms, but also to track the dynamics of the dependence of the results on RAM bandwidth.
Our testing used the Phenom II X3 720 BE and Phenom II X4 955 BE processors, operating at nominal 2.8 and 3.2 GHz, respectively. We specifically selected two processors with different processing power and number of cores in order to identify the relevance of high-frequency memory modules with higher bandwidth for both older representatives of the Phenom II family and mid-class models.
Processor Specifications
Basic data on processors is included in the following table:
| AMD Phenom II X4 955 BE | AMD Phenom II X3 720 BE | |
| Core | Deneb | Heka |
| Technical process, nm | 45 SOI | 45 SOI |
| Connector | AM3 | AM3 |
| Frequency, MHz | 3200 | 2800 |
| Factor | 16 | 14 |
| Clock generator | 200 | 200 |
| L1 cache, KB | 128 x 4 | 128 x 3 |
| L2 cache, KB | 512 x 4 | 512 x 3 |
| L3 cache, KB | 6144 | 6144 |
| Supply voltage, V | 0,875-1,5 | 0,850-1,425 |
| TDP, W | 125 | 95 |
We also provide a couple of screenshots of the CPU-Z utility with data on the processors in question:
Test configuration
The Socket AM2+ platform was tested on the following configuration:
- Processors AMD Phenom II X3 720 BE, Phenom II X4 955 BE;
- Cooler: Thermalright Ultra-120 eXtreme;
- Motherboard: MSI 790XT-G45;
- Video card: Point of View GF9800GTX 512MB GDDR3 EXO (@818/1944/2420 MHz);
- Memory: OCZ OCZ2FXE12004GK (2x2GB DDR2-1200);
- Sound card: Creative Audigy 4 (SB0610);
- Hard drive: WD3200AAKS (320 GB, SATA II);
- Power supply: FSP FX700-GLN (700 W);
- Operating system: Windows Vista Ultimate SP1 x64;
- Video card driver: ForceWare 190.62.
- Motherboard: MSI 790FX-GD70;
- Memory: Kingston KHX1600C9D3K2/4G (2x2GB DDR3-1600).
Since the Ungaged mode is enabled by default, it was used as the main one. In Gunged mode, additional tests were carried out only at the maximum frequency of DDR2 and DDR3 memory, since it would be logical to assume that it is with higher memory bandwidth that the operating features of the memory controller will be more noticeable.
We also conducted a number of additional tests at an increased frequency of the NB north bridge built into the processor, at the frequency of which the memory controller and third-level cache operate. Theoretically, by increasing the NB frequency we should get quite a noticeable performance increase. Again, to identify the dependence of performance on this factor, we performed the test only at the maximum memory frequency. Unfortunately, due to lack of time, we had to limit ourselves to tests only on Socket AM3 in combination with DDR3.
For both processors in each testing mode, the same timings were set, the Drive Strength parameters were left in Auto mode.
Test Modes
Memory modules with this frequency are the most common and affordable. Latencies 5-5-5-18 are standard for this memory (with the exception of overclockers with low timings). However, recently many modules designed for CL6 have appeared on the market, but they usually work without problems at lower latencies.
For the Phenom II X3 720 BE and Phenom II X4 955 BE at a given DDR2 memory frequency, all timings were fixed at the following values:
The maximum operating mode for AMD processors is DDR2 memory.
In the first case, we used fairly high timings, which were set to the following values:
More relevant mode with CAS Latency 5.
Memory latencies were set for processors to the following values:
The memory settings are identical to the previous configuration, but the controller operates in Ganged mode.
Officially, Phenom II processors only support DDR3-800/1066/1333 memory, but top-end motherboards allow a nominal frequency of 1600 MHz. The values of 800 MHz and 1066 MHz are of little interest, since even the cheapest DDR3 memory kits currently available on the market are rated at 1333 MHz. That is why DDR3-1333 and DDR3-1600 modes were used for our testing.
For the first mode, delays were set, which in general do not differ much from the standard timings of cheap DDR3-1333 modules.
With memory modules designed for a frequency of 1600 MHz, not everything is so clear in terms of timings. Some of the kits operate at such frequencies at CL9, but most modern overclocking kits are initially designed for 8-8-8 (or even 7-7-7) level timings, so this is the configuration that was used for our tests.
It’s just that in this “high-speed” mode, the Phenom II X3 720 BE completely refused to function normally and no manipulations helped achieve stability at such timings. Only with delays of 9-10-10-24 did the system work without failures. So, with a memory frequency of 1600 MHz, we had to limit ourselves to testing only one Phenom II X4 955 BE. We also note that such “incompatibility” was an isolated case for us, and the Phenom II X2, and even the Athlon II X2 (which will appear in our next articles) worked with DDR3-1600 memory without any problems.
Since the Phenom II X3 720 BE worked only with DDR3-1333 MHz, it was at this memory frequency that we tested both processors in Ganged controller mode.
Tests with an increased frequency of the built-in north bridge in the processor (NB) were carried out at different memory frequencies, respectively, for the younger model at DDR3-1333, for the older one at a memory frequency of 1600 MHz.
All timings are identical to the DDR3-1333 7-7-7-20 mode.
All timings are identical to the DDR3-1600 8-8-8-24 mode.
Test results
Lavalys Everest Memory Benchmark
Below is the data from the memory subsystem performance test built into the Lavalys Everest program. To reduce the error, this benchmark was run five times for each mode. The letter U in the diagrams denotes the Unganged mode, and G, respectively, Ganged.
Very noticeable growth with increasing memory bandwidth. With DDR2 in Ganged mode, we get an even more than 8% increase, but even when using DDR3 in this mode, the gain in reading speed is negligible.
Here, memory timings and its frequency have almost no effect on the result, but there is a slight drop when working in Ganged mode. But the increase from increasing the frequency of the built-in north bridge is very high.
The huge difference in the Ganged controller mode on the AM2+ and AM3 platforms immediately catches your eye. If on the first one the activation of this mode leads to only a slight drop in results, then on AM3 the difference reaches 20%. There is also a very noticeable lag when using DDR2-800 memory, but between DDR2-800 and DDR3-1333 (or even DDR3-1600) the difference is much smaller.
Overall, memory latency is still slightly reduced when Ganged is activated. The difference between DDR2-1066 and DDR3-1333 turns out to be less than between DDR2-800 and DDR2-1066, and the lag in the configuration with DDR2-800 is most noticeable on the older processor.
PCMark Vantage
The latest version of the PCMark application shows inconsistent results. Initially, it was planned to compare our processors in the PCMark Suite, Memory Suite and Productivity Suite test sets, but the spread of results in the first and last was quite large and the final data was completely inadequate. Only the indicators in the Memory Suite were enviably stable, and that is what we present here.
But this test is practically indifferent to memory frequency and other settings, but there is still a slight drop in results when the Ganged mode is activated. Overclocking NB traditionally brings some gains.
WinRar 3.90 b1
The built-in performance test was run seven times.
This application turns out to be quite sensitive to changes in memory frequency; the performance increase from NB is also noticeable, although it is quite small. But the Ganged mode again negatively affects the final result.
7-Zip 4.65
The built-in performance test was run five times.
This archiver no longer reacts to changes in memory bandwidth. If on the older quad-core processor there is still at least some trace of positive dynamics in the growth of results with increasing RAM frequency (in Ganged there is again a slight decrease in the final score), then already on the Phenom II X3 the difference between all modes is calculated in hundredths of a percent, all differences are determined by the error measurements, which is why it is no longer possible to trace any dependence using these data.
Paint.Net 3.36
For tests, a special benchmark version 3.20 was used. To increase the accuracy of the results obtained, the test was run seven times. Note that the spread of results after each test run on the older processor was smaller than on the younger one, and, most likely, the results of the Phenom II X3 again should not be considered very accurate due to the influence of a larger error.
Performance varies slightly between modes. It is noticeable that in Ganged mode the test execution time is slightly faster. Phenom II X3 in combination with DDR3-1333 turns out to be slower for some reason than in combination with DDR2-1066, while Phenom II X4 with DDR3 shows better results than with DDR2. However, let's not forget about the greater impact of the error on the Phenom II X3. This factor may have also caused a certain drop in performance with increasing NB frequency, while on the Phenom II X4 we again see a completely expected increase in results in this mode.
CineBench 10
In this application, the test was repeated three times for each mode.
And again, the difference in the results is so insignificant that it can be attributed to an error, but some patterns in the results are visible. The performance increase with increasing memory frequency, although miniscule, is present. The Ganged mode in the multiprocessor test leads to a slight decrease in the final score.
A surprise awaits us when we look at the results of this test. For unknown reasons, on the Socket AM2+ motherboard they are higher than on the Socket AM3.
But according to the processor test data, everything looks quite adequate and with DDR3 memory processors show better results. On the Phenom II X4, only DDR3-1600 outperforms DDR2-1066 (5-5-5-18); on the Phenom II X3, even with DDR3-1333, the result is not inferior to DDR2-1066.
The Last Remnant
A special gaming benchmark was used, which was run three times.
This game responds quite well to changes in RAM bandwidth. The difference between the slowest DDR2 configuration and the fastest DDR3 configuration reaches 8%. The influence of the Ganged mode manifests itself in different ways: on the AM2+ platform with DDR2 memory we see an increase in results, but on the AM3 platform there is already a drop in performance. Increasing the frequency of the NB block has a very positive effect on performance, and the older processor benefits from this more than the younger one.
Far Cry 2
Game version 1.03. All settings are set to Medium, including the Performance section values (physics, fire, trees). The test included two 7-run cycles of the Ranch Small demo.
In the game Far Cry 2 we again see a good dependence on the memory subsystem. So, without any overclocking of the processor itself, just raising the frequency of the NB block and using fast DDR3-1600, we achieve a gain of 13% (on the Phenom II X4) over the “slowest” mode with DDR2-800. And in general, as can be seen from the results, DDR2-800 slightly “limits” the potential of both processors. As for the Ganged mode, performance is reduced in it.
Game version 1.2. The tests were carried out in the Crysis Benchmark Tool, the standard CPU-benchmark was run (the bat file for running it is located in the bin 64 folder). This demo includes a scene in which the hero destroys several houses with a grenade launcher, and it puts the maximum possible load on the central processor due to the abundance of fragments and other active objects. The test included five cycles of 4 runs of the test “demo” each.
And this game shows quite a good dependence on the memory subsystem. Once again, the older processor benefits more from higher memory frequencies than the younger one. For the first, the difference between DDR2-800 and DDR3-1600 is 10%, for the second, the difference between DDR2-800 and DDR3-1333 is just over 4%. DDR2-1066 with delays of 5-5-5-18 loses even to DDR3-1333 (7-7-7-20). When the memory controller operates in Ganged mode, the results are slightly reduced, but increasing the NB frequency, as usual, improves performance.
We also note that in this test on an older processor there is practically no difference between DDR3-1333 and DDR3-1600, which indicates that even at a frequency of 1333 MHz (and latencies of 7-7-7-20) the memory practically does not limit the potential Phenom II X4 955 BE in this application.
conclusions
It's time to sum up the results of our testing. In general, it can be noted that the difference between the new AM3 platform and the older AM2+ is not very significant. In some tests, these differences tend to zero, but in some applications (especially games and archivers) there is a significant advantage of Phenom II processors in conjunction with DDR3 memory.
Also, these differences are largely due to the power of the processor itself, as we saw from the examples of the Phenom II X3 720 and Phenom II X4 955, because in percentage terms the greater increase from the use of faster memory modules was observed in the second processor. So for the younger dual- and triple-core Phenom II and Athlon II models, the problem of choosing memory is less relevant, since this will have a minor impact on the final performance. However, we would still recommend using a minimum of DDR2-1066 even at normal timings, since in some applications the slow DDR2-800 slightly “limites” the potential of even mid-class processors.
In some applications, DDR2-1066 (5-5-5-18) turns out to be faster than DDR3-1333 (7-7-7-20), but more often they are either on par or DDR3 still has the advantage. Moreover, this pattern manifests itself on all processors; it will simply be more pronounced on more powerful ones. So for older CPUs, it is, of course, more advisable to use the Socket AM3 platform in combination with high-speed DDR3 memory modules.
Regarding the Ganged operating mode, we can say that in most tests it leads to a drop in performance, and where its activation has a positive effect, the gain from this is small. Therefore, it is no coincidence that by default the boards operate in the more efficient Unganged mode. Another interesting thing is that on different platforms, activating this mode has different effects on the final performance. In particular, in the game The Last Remnant in Ganged mode with DDR2 we see an increase in the result, but with DDR3 there is already a drop. This, however, once again confirms that for a modern multi-core system based on Socket AM3 this mode is contraindicated, and for Socket AM2+ this parameter is less important. By the way, in Ganged mode the stability of the memory subsystem also decreases—during testing we had to slightly increase the voltage on the NB and RAM.
It is also worth noting the benefits of increasing the frequency of the north bridge built into the processor, along with which we are also increasing the frequency of the L3 cache. Even in the nominal operating modes of the considered processors, this has the most positive effect. The boost from overclocking the NB to 400 MHz is sometimes no less effective than moving from DDR2 to DDR3. In percentage terms, this increase in performance was greater on the older processor, and it is logical to assume that as the CPU frequency increases, the gain from overclocking the NB will be even more relevant. So when overclocking the Phenom II, this parameter will play an important role, and in order to fully unleash the potential of AMD processors when increasing their frequency, it is necessary to at the same time increase the NB frequency. But this also requires an increase in the corresponding voltage, which entails an increase in the overall temperature of the processor, and when overclocking a processor it is not always possible to achieve the same high NB frequencies as during its nominal operation. However, we will look at how this affects processor overclocking in practice in one of the following materials...
We thank the following companies for providing test equipment:
- AMD for the Phenom II X4 955 BE processor;
- MSI for 790XT-G45, 790FX-GD70 boards and Phenom II X3 720 BE processor;
- Special educational equipment for memory Kingston KHX1600C9D3K2/4G;
- for the WD3200AAKS hard drive.
AMD's current policy in terms of processor production is extremely clear. All efforts are aimed at creating Deneb chips for the Phenom II X4 9*0 line of processors. However, the production of such high-tech crystals is far from simple, even by modern standards. The percentage of defects is so high that recycling it irrevocably would lead to a significant increase in the cost of full-fledged working chips. That is why, having successfully systematized the rejected crystals, AMD provided naturally discounted models, united in the Phenom II X4 8 * 0 line (Deneb core); Phenom II X3 7*0 (Heka core) and even Phenom II X2 5*0 (Callisto core). You can get acquainted with the characteristics of some representatives of all lines, the Phenom II family, by looking at the table below.
|
Name |
Phenom II X4 945 |
Phenom II X4 910 |
Phenom II X4 810 |
Phenom II X4 805 |
Phenom II X3 720 |
Phenom II X3 710 |
Phenom II X2 550 |
|
Technical process, nm |
|||||||
|
Core |
|||||||
|
Connector |
|||||||
|
Frequency, MHz |
|||||||
|
Factor |
|||||||
|
HTT/Bclk |
|||||||
|
L1 cache, KB |
|||||||
|
L2 cache, KB |
|||||||
|
L3 cache, KB |
|||||||
|
Supply voltage, V |
|||||||
|
TDP, W |
|||||||
|
Limit temperature, °C |
|||||||
|
Set of instructions |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
RISC, IA32, x86-64, NXbit, MMX, 3DNow!, SSE, SSE2, SSE3, SSE4a |
|
|
Approximate price as of 07/14/09, $ |
The Phenom II X2 550 Black Edition processor tested today is the fastest in its line, however, it is this model range that has undergone the most significant “amputations” compared to all Phenom II representatives. The series lost two whole cores, with the same amount of third-level cache memory. But, first things first - first, about its packaging.
Appearance of packaging
Let us remind you that the Phenom II X2 550 is “crowned with the title” Black Edition. Accordingly, the packaging, according to AMD tradition, is exclusively black without any “flashy” logos.
The blue square on the front of the package displays the main advantages of the model. This is a fairly high clock speed of 3.1 GHz, a total cache memory of 7.0 MB, and is also designed for installation in the Socket AM3 processor socket.
Equipment
The packaging of the boxed model of the Phenom II X2 550 BE did not bring any surprises, but it did not cause disappointment either.
The delivery includes:
- Processor Phenom II X2 550 Black Edition;
- CPU cooler FOXCONN(N)1A018E000;
- Installation instructions and three-year warranty;
- Sticker on the system unit.
The “lightweight” model of the FOXCONN(N)1A018E000 cooler is already familiar to us firsthand. This model is included with all “stripped down” Phenom II models. However, its effectiveness in cooling the Phenom II X2 550 Black Edition processor being reviewed today will be tested in practice and described below.
There is a sticker included. Let us remind you that it was absent in the first tested models of the Phenom II family. Based on letters from readers, we received information that all models of the Phenom II family of new batches are equipped with a sticker.
AMD processor Phenom II X2 550 Black Edition
After examining the heat distribution cover of the Phenom II X2 550 Black Edition processor, it became known that its place of production is Malaysia. The marking is represented by the alphanumeric combination HDZ550WFK2DGI, which can be deciphered as follows:
- HD – AMD K10.5 architecture processor for workstations;
- Z – processor with a free multiplier;
- 550 – model number indicating the family (first digit) and the position of the model within the family (the remaining digits - the higher the number, the higher the operating clock frequency);
- WF – processor thermal package up to 80 W at a supply voltage in the range of 0.875 – 1.425 V;
- K – the processor is packaged in a 938 pin OµPGA (Socket AM3) case;
- 2 – the total number of active cores and, accordingly, the amount of L2 cache memory 2x512 KB;
- DGI - Callisto core (45 nm) stepping C2.
It should be noted that there is some discrepancy with the labeling. The letter combination DGI was used to mark the previously reviewed processors Phenom II X3 710 and Phenom II X3 720 Black Edition. which have a Heka core, which assumes the presence of three active computing cores. But the Phenom II X4 810 processor, also reviewed earlier, is labeled as FGI, and has four active computing cores, but a “cut down” third-level cache. Well, the most surprising thing is that the full-fledged processors Phenom II X4 920 and Phenom II X4 940 are also labeled DGI, although they did not fall under the “scalpel”. However, the Phenom II X2 550 Black Edition processor we are considering today is dual-core.
The back of the processor exposes the 938-pin package. This is Socket AM3. Let us remember that it is backward compatible with the AM2+ connector, and the memory controller built into the processor can work with DDR2 and DDR3 memory.
Specification:
|
AMD Phenom II X2 550 BE |
|
|
Marking |
|
|
CPU socket |
|
|
Clock frequency, MHz |
|
|
Factor |
15.5 (starting) |
|
HT bus frequency, MHz |
|
|
L1 cache size, KB |
|
|
L2 cache size, KB |
|
|
L3 cache size, KB |
|
|
Number of Cores |
|
|
Instructions support |
MMX, 3DNow!, SSE, SSE2, SSE3, SSE4A, x86-64 |
|
Supply voltage, V |
|
|
Thermal package, W |
|
|
Critical temperature, °C |
|
|
Technical process, nm |
|
|
Technology support |
Cool'n'Quiet 3.0 |
Proprietary technologies:
- AMD CoolCore- hardware technology allows you to disable currently unused processor units to reduce power consumption and heat dissipation; no driver or activation in the BIOS is required.
Enhanced virus protection technology (NX bit / Enhanced Virus Protection). Supported by operating systems starting from Windows XP SP2, it is designed to prevent the spread of some viruses that use a buffer overflow error (for example MSBlaster and Slammer), i.e. allows you to prohibit the execution of program code located in memory areas intended for data.
128-bit SSE block and SSE4a instruction set. Includes 6 new instructions to fully and efficiently support the respective applications.
AMD Virtualization (AMD-V)- improved technology that allows you to simultaneously run two independent Operating Systems on one PC.
AMD Cool 'n' Quiet 3.0 Technology provides an effective reduction in power consumption, thereby enabling the creation of quieter computing systems. For the technology to function, support/activation in the BIOS and a software driver is required.
Dual Dynamic Power Management Technology- Provides independent power to all processor cores and memory controller for optimal performance and power consumption.
By tradition, confirmation of the characteristics is a screenshot of the CPU-Z program.
However, even the latest version of the CPU-Z program brought a surprise. Pay attention to the Code Name cell. Codename for the core is Deneb, when the real name should be Callisto. Most likely, this embarrassment is due to the fact that the Phenom II X2 processor line is quite “fresh” and the authors of the CPU-Z program at the time of creating version 1.51 simply did not know that this processor model would exist.
The Cahce section of CPU-Z showed the cache distribution. 128 KB L1 cache per core. 512 KB of L2 cache also per core and a total of 6 MB of L3 cache.
DDR3 memory worked at the “native” frequency of the controller built into the processor, 1333 MHz, with a corresponding set of timings.
During testing we used Processor Test Stand No. 1
| Motherboards (AMD) | ASUS M3A32-MVP DELUXE (AMD 790FX, sAM2+, DDR2, ATX)GIGABYTE GA-MA790XT-UD4P (AMD 790X, sAM3, DDR3, ATX) |
| Motherboards (AMD) | ASUS F1A75-V PRO (AMD A75, sFM1, DDR3, ATX)ASUS SABERTOOTH 990FX (AMD 990FX, sAM3+, DDR3, ATX) |
| Motherboards (Intel) | GIGABYTE GA-EP45-UD3P (Intel P45, LGA 775, DDR2, ATX)GIGABYTE GA-EX58-DS4 (Intel X58, LGA 1366, DDR3, ATX) |
| Motherboards (Intel) | ASUS Maximus III Formula (Intel P55, LGA 1156, DDR3, ATX)MSI H57M-ED65 (Intel H57, LGA 1156, DDR3, mATX) |
| Motherboards (Intel) | ASUS P8Z68-V PRO (Intel Z68, sLGA1155, DDR3, ATX)ASUS P9X79 PRO (Intel X79, sLGA2011, DDR3, ATX) |
| Coolers | Noctua NH-U12P + LGA1366 KitScythe Kama Angle rev.B (LGA 1156/1366)ZALMAN CNPS12X (LGA 2011) |
| RAM | 2x DDR2-1200 1024 MB Kingston HyperX KHX9600D2K2/2G2/3x DDR3-2000 1024 MB Kingston HyperX KHX16000D3T1K3/3GX |
| Video cards | EVGA e-GeForce 8600 GTS 256 MB GDDR3 PCI-EASUS EN9800GX2/G/2DI/1G GeForce 9800 GX2 1GB GDDR3 PCI-E 2.0 |
| HDD | Seagate Barracuda 7200.12 ST3500418AS, 500 GB, SATA-300, NCQ |
| power unit | Seasonic SS-650JT, 650 W, Active PFC, 80 PLUS, 120 mm fan |
Select what you want to compare the AMD Phenom II X2 550 with
Knowing the approximate drop in performance when testing triple-core models compared to quad-core models of the same Phenom II family, it was not difficult to guess the performance of dual-core models of the same family. The clock frequency of the Phenom II X2 550 Black Edition increased by 100 MHz compared to the Athlon II X2 250 and the presence of 6 MB of third-level cache gave a slight increase in performance. Otherwise, the standard dependence of the number of computing cores on performance, adjusted for clock frequency. But this slight increase in performance allows you to try to compete with equal-frequency dual-core Intel processors, especially when taking into account the cost of these processors.
Efficiency of a boxed cooler
The FOXCONN(N)1A018E000 cooling system, which comes with all processor models of the Phenom II X4 8** and Phenom II X3 7** lines, did not demonstrate much efficiency. This was especially evident when testing the Phenom II X4 810 processor, although when the processor was running at “standard” voltage and frequencies, it coped with its responsibilities.
Let us remind you that this cooler consists of a solid aluminum radiator, the dimensions of which are 30x68x77 (HxWxD) mm. The central thermal column is square-shaped in section; heat-dissipating ribs extend diagonally from it, four of which are thickened, because They also serve as a fan mount.
The radiator is fastened with a “traditional” clip, which fits into the corresponding “grooves” in the radiator.
The fan is marked as FOXCONN PV701512F2BF 1G. Its standard size is 70 mm, and its height is only 15 mm, which means that it is low-profile. The fan drive is equipped with a PWM (PWM) converter, which makes it possible, when connected to the appropriate 4-pin connector, to automatically adjust the speed of rotation of the impeller. The maximum rotation speed of the blades during testing reached ~3000 rpm, while the noise level can be described as moderate and does not stand out from the background of other fans in the system. For a more realistic idea of the effectiveness of a “boxed” cooler in cooling the dual-core processor Phenom II X2 550 Black Edition, it was given the most serious opponent Scythe Kama Angle . Moreover, the rotation speed of the blades of the latter was maximum, i.e. 1200 rpm In parallel with monitoring the processor temperature, the power consumption of the system as a whole was measured to assess the energy efficiency of the Phenom II X2 550 BE processor. Energy-saving technologies C1E and Cool`n`Quiet have been disabled due to possible distortion of results.
First, measurements were made at “standard” frequencies and voltages. The clock frequency is 3100 MHz, and the processor supply voltage is 1.34 V, i.e. the one installed by the GIGABYTE GA-MA790XT-UD4P motherboard in AUTO mode.
As you can see, the “boxed” cooler was able to “maintain” the temperature under load at 58°C, which is 8°C less than that of the Phenom II X4 810 and as much as 18°C more than the performance cooler Scythe Kama Angle. The energy efficiency of the Phenom II X2 550 Black Edition processor in idle mode is almost the same as that of the dual-core Athlon II X2 250, which AMD positions as more economical. But under load, processor consumption diverges significantly. This is due to the presence of a large third-level cache memory in the Phenom II X2 550 Black Edition.
Efficiency of using DDR3
The Phenom II X2 550 Black Edition processor is capable of working with both DDR2 and DDR3 memory. Despite the fact that at the moment DDR3 memory is almost equal in price to DDR2 memory, new AM3 motherboards will be able to use it. That is why we present to your attention comparative tests of the Phenom II X2 550 Black Edition processor using DDR3-1333 and DDR2-800 memory.
|
Test package |
Result |
Decrease in productivity, % |
||
|
We use DDR3 |
We use DDR2 |
|||
|
Rendering |
||||
|
Shading, |
||||
|
Tom Clancy's H.A.W.X. Demo, |
||||
|
DirectX 9 |
||||
|
DirectX 10 |
||||
The overall average performance drop was only 2,68% . Only the buyer himself can decide whether this is a lot or a little, focusing on the difference in price. In any case, if you have a motherboard with a Socket AM3 connector and DDR3 memory at an affordable price, you shouldn’t give up the extra three to five frames per second.
Overclocking
Since the Phenom II X2 550 processor model we are considering today is the Black Edition, which implies a free, not blocked, multiplier, we decided to try to overclock it without increasing voltage, because not all motherboards have the ability to change the processor supply voltage in a sufficient range.
Stable operation was achieved at a processor clock frequency of 3817 MHz. What's on 23% higher than the nominal clock frequency. It is worth noting that the multiplier changed from x15.5 to x19.0, while the reference bus frequency remained unchanged.
The processor temperature when overclocked without raising the voltage using a “boxed” cooler increased by only 2°C at idle, and by 3°C under load, but still remains acceptable. But power consumption increased by 12 watts and amounted to 237 watts, which, even without the results of overclocking with increased voltage, makes you think not only about a productive cooler like the Scythe Kama Angle, but also about a powerful power supply, as well as a good motherboard that will be capable of " feed" the processor during overclocking.
By raising the voltage to 1.44 V, it was possible to achieve stable system operation at a processor clock frequency of 3939 MHz. In this case, the multiplier value was x19.5. Compared to the “standard” clock frequency, the increase was 27%. In fact, this is a very solid overclocking, since not a single “brother” model of the Phenom II family was able to achieve stable operation at this clock frequency. For example, the Phenom II X3 720 Black Edition model was able to overclock to only 3608 MHz at a rather dangerous voltage of 1.536 V. The Phenom II X4 810 model, which is not a representative of the elite Black Edition division, was overclocked in the classical way, i.e. raising the reference frequency, and reached a clock frequency of only 3445 MHz at 1.44 V. The only exception can be considered the Phenom II X4 940 Black Edition processor, which reached a clock frequency of 3811 MHz at a voltage of 1.44 V. However, do not forget that this is a representative of a full-fledged line that can only work with DDR2 memory, which naturally affected the results of its overclocking.
It’s not for nothing that the “boxed” cooler is missing from the table above. Its effectiveness turned out to be extremely insufficient– the system “freezes” under load. But Scythe Kama Angle once again demonstrated his “icy heart”. Difference between the highest recorded temperature in nominal mode and during acceleration with increasing voltage was only 6°C and as much as 32°C lower than the critical temperature declared by AMD. Power consumption during overclocking with increased voltage increased by another 23 watts. The assumptions were confirmed; to operate the Phenom II X2 550 Black Edition processor in an overclocked state with increased voltage, you will need a high-performance cooler, a good power supply and a motherboard with a high-quality processor power supply. We suggest evaluating the increase in performance of the overclocked Phenom II X2 550 Black Edition in the following table.
|
Test package |
Result |
|||
|
Rated frequency |
Overclocked processor |
|||
|
Rendering |
||||
|
Shading, |
||||
|
Fritz Chess Benchmark v.4.2, knodes/s |
||||
|
Tom Clancy's H.A.W.X. Demo, |
||||
|
DirectX 9 |
||||
|
DirectX 10 |
||||
When overclocking the processor to 27% the average increase in productivity was 16.4%. This nonlinearity is due to the fact that the “standard” processor frequency is 3100 MHz, which is quite high even by today’s standards, and not all tasks depend only on the core clock frequency.
Activating blocked kernels
At the moment, it’s no secret that for all modifications of the “cut down” processors of the AMD Phenom II family, you can try to unlock and restore previously disabled blocks. Naturally, to claim that all models are capable of “unlocking” is an absolute fallacy. However, the culprit of today's review made us sweat... The approach applied to the Phenom II X3 720 Black Edition processor was unsuccessful, i.e. by setting the option Advanced Clock Calibration (ACC) in meaning AUTO no changes were noticed. Using the method of “scientific poking” and studying the information posted on the Internet, the following values were set for the BIOS items in the Advenced Clock Calibration section.
- EC Firmware Selection
- Advanced Clock Calibration
- Value (all cores) [-2%]
Fingers crossed, the system was launched, and a few minutes later a wonderful screenshot of the task manager window and the CPU-Z program was taken.
The dual-core Phenom II X2 550 Black Edition processor has turned into the non-existent quad-core Phenom II X4 B50 BE! Now we have in our hands a full-fledged Deneb processor with a “starting” clock frequency of 3100 MHz. Let us recall that the Phenom II X3 720 Black Edition processor, with the Advanced Clock Calibration option set to AUTO, just like today’s Phenom II X2 550 BE, became quad-core and received the non-existent “official name” Phenom II X4 20. A system with an already quad-core Phenom II X2 The 550 Black Edition was surprisingly absolutely stable. No nuances in operation were noticed during testing.
|
Test package |
Result |
Productivity gain, % |
||
|
Phenom II X2 550 |
Phenom II X2 550 |
|||
|
Rendering, CB-CPU |
||||
|
Shading, |
||||
|
Fritz Chess Benchmark v.4.2, knodes/s |
||||
|
Tom Clancy's H.A.W.X. Demo, |
||||
|
DirectX 9 |
||||
|
DirectX 10 |
||||
This is where productivity has improved significantly! This suggests that increasing the clock frequency above ~3.0 MHz does not provide as much performance gain as increasing the number of active cores. Thus, activating two previously blocked cores at a constant clock frequency gave an average increase in performance 46% . At times almost linear with the number of cores. That is why there were special hopes for successful overclocking of the newly-made quad-core processor.
The fact of stable operation at a clock frequency of 3838 MHz of the Phenom II X2 550 Black Edition processor with two unlocked cores speaks for itself. This is the highest result that was achieved when overclocking any processor from the Phenom II family in our test laboratory. In this case, the voltage supplied to the processor was 1.4 V. Which does not even go beyond the limits set by the manufacturer. Fabulous! By paying ~$110 you can get a processor whose characteristics lie between the most expensive and advanced models of the Phenom II X4 955 Black Edition and Phenom II X4 945 family today.
Naturally, one fact of fantastic characteristics is not enough. This is why the decision was made to pit the unlocked and overclocked Phenom II X2 550 Black Edition against some of the most powerful processors from competing Intel. So, the competitors will be Core 2 Quad 9550 and Core i7 940 , previously reviewed on our website, as well as the “brotherly” Phenom II X4 940. To achieve more adequate results, only processor tests will be compared.
Futeremark PCMark`05 showed an absolutely linear dependence of performance not so much on the number of computing cores but on the processor clock frequency.
CrystalMark has already shown more realistic performance results for multi-core processors. Although the threefold difference between the Phenom II X2 550 Black Edition, operating at standard frequencies with two computing cores, and the Phenom II X2 550 Black Edition, overclocked to 3838 MHz with two cores unlocked, seems unrealistic. Nevertheless, the quad-core Phenom II X2 550 Black Edition processor operating at a clock frequency of 3838 MHz confidently holds superiority over its far from weak competitors, which are two to three times more expensive.
Futeremark PCMark`06, in turn, gave the most interesting results, showing that in general, gaming performance primarily depends on the video subsystem, and only then does processor performance become important.
Well, coming to the conclusions of this article, I would like to show the results of “dancing with a tambourine” over the extraordinary Phenom II X2 550 Black Edition processor in all its glory.
|
Test package |
Result |
Productivity gain, % |
||
|
Phenom II X2 550 |
Phenom II X2 550 |
|||
|
Renderin, |
||||
|
Shading, |
||||
|
Fritz Chess Benchmark v.4.2, knodes/s |
||||
|
Tom Clancy's H.A.W.X. Demo, |
||||
|
DirectX 9 |
||||
|
DirectX 10 |
||||
The overall average performance increase when activating two cores and overclocking to 3838 MHz was 67.45% (!). In the editor's memory this is first case such an increase in performance directly, and not in the percentage of characteristics. Moreover, the acceleration noticeable to the “naked eye” is observed not only in specific synthetic tests, but in real applications and games. And the most amazing thing is that such an increase in productivity was obtained without the use of “sophisticated” technologies, such as cooling with liquid nitrogen, multi-level freon units, or even the ubiquitous water cooling system. Paradox or incredible luck? Most likely, the second, since it was not possible to find more suitable logical reasons.
Conclusion
As for using the AMD Phenom II X2 550 Black Edition processor in nominal mode, this prospect is a little doubtful. The fact is that the Phenom II X2 550 Black Edition is generally inferior to the more technologically advanced Athlon II X2 250 processor, both in terms of power consumption, heating, and even cost. After all, although in terms of performance at nominal frequencies there is a slight superiority of the Phenom II X2 550 Black Edition, But the same energy consumption, heating and price completely neutralize it. Here I would like to advise you to either save a little and get almost the same performance at lower costs, or, conversely, add a little and buy at least a three-core processor. The only thing that can improve the situation a little is if you look at this processor as a Black Edition model with mandatory subsequent overclocking. But if you choose the AMD Phenom II X2 550 Black Edition as an object of real overclocking with an attempt to unlock two more cores, then its purchase will be more than justified!
As already stated, 67% "free" The increase in absolute productivity does not require any explanation. There is simply no processor that is more profitable for an overclocker today. However, please note that:
have “good” luck, since world practice shows that not every processor of the Phenom II family “cut down in cores” can be unlocked. However, if there is an opportunity to “choose”, not using it would be not only stupidity, but real laziness!
To use the Phenom II X2 550 Black Edition processor overclocked to 3838 MHz with two cores activated, you should, at a minimum, acquire a good power supply, an appropriate motherboard with a high-quality processor power supply and an 8-pin power connection, as well as an alternative high-performance cooler.
In this article we will talk about choosing the optimal video card for AMD AM3 and FM1 processors:
- Phenom X6 1035T, 1045T, 1055T, 1065T, 1075T, 1090T, 1100T
- Phenom X4 910, 920, 925, 940, 945, 955, 960T, 965, 970, 975, 980
- Athlon II X4 620, 630, 635, 640, 645, 655
- Athlon II X4 631, 641, 638, 651, 651K
Due to the unstable economic situation, many PC users do not want or do not have the opportunity to change the platform, “sitting” on the old one for as long as possible. Therefore, many people face the question of choosing the optimal combination of an old multi-core CPU and a more or less modern video card. We will try to select the most comparable solutions from those available on the market.
Video card for AMD Phenom X6 1035T, 1045T, 1055T, 1065T, 1075T, 1090T, 1100T and AMD Phenom X4 910, 920, 925, 940, 945, 955, 960T, 965, 970, 975, 980
These processors are close in performance to solutions from the AMD FX-4000 and FX-6000 lines. Consequently, older four- and six-core models when overclocked will be able to work in tandem with video cards of the levelAMD Radeon R7 370/RX 460 And NVIDIA GeForce GTX 750 Ti. We recommend using the younger ones together with solutions of the levelAMD Radeon R7 360 And NVIDIA GeForce GTX 750.
Video card for AMD Athlon II X4 620, 630, 635, 640, 645, 655, 631, 641, 638, 651, 651K
We recommend using the more productive solutions listed above in conjunction with video adapters of the level AMD Radeon R7 360 And NVIDIA GeForce GTX 750. As for models with low frequencies, somewhat outdated ones are best suited to them. AMD Radeon R7 250/R7 250X And NVIDIA GeForce GTX 650 / GT 740.