IntroductionThe approaching summer promises to be a truly hot time. And if from a meteorological point of view this forecast may not come true due to the action of powerful cyclones, then in the processor market everything has already been determined absolutely precisely. Both leading players, AMD and Intel, have chosen the summer period to update their high-performance platforms. Thus, Intel will launch processors with a fundamentally new Core microarchitecture to the market in mid-summer, and AMD will focus on promoting the Socket AM2 platform, which provides support for DDR2 SDRAM, to the market throughout the summer season.
Although the currently most anticipated processors should be considered the Intel Core 2 Duo family of CPUs, also known by their codename Conroe, AMD, according to tradition that has developed over the past few years, is ahead of its competitor and will begin mass deliveries of its updated processors for the Socket AM2 platform on June 1 . That is why today we will take a closer look at the new products from AMD, postponing the publication of Core 2 Duo reviews for a while, until their official announcement.
Despite the imminent release of very promising Intel processors, the Socket AM2 platform from AMD is attracting a lot of attention. AMD delayed the transition to using DDR2 SDRAM until the last minute, because the K8 processor microarchitecture, which includes an integrated memory controller, benefits primarily not from memory bandwidth, but from its low latency, which the existing DDR2 SDRAM on the market cannot boast of. However, today DDR2 memory speeds have already increased so much that switching Athlon 64 family processors to work with this type of memory can theoretically give tangible dividends in the form of performance gains. Although the first tests of engineering samples of the new platform from AMD did not reveal any particular advantages, now we are talking about serial processors and motherboards. This is the main intrigue of this material. After all, to numerous fans AMD processors I want to believe that Socket AM2 processors will be able to compete on equal terms with Intel Core 2 Duo.
In addition, the updated AMD processors receive a new revision of the core, which, in addition to supporting new types of memory, has some cosmetic changes, which also increase the attractiveness of the Athlon 64 family of processors. Of course, the appearance of Intel processors with the Core microarchitecture will contribute to the outflow of adherents of the current AMD solutions into the “enemy camp”. But it’s too early to draw hasty conclusions, especially since some improvements in K8 processors may be very popular in a number of cases. So, let's take a closer look at AMD processors for Socket AM2 and try to predict how attractive they will be to potential consumers.

Revision F Core: Basics

For use in new processors designed for the Socket AM2 platform, AMD has developed an updated core with the K8 microarchitecture, which received revision number F. Thus, all dual-core and single-core AMD processors with an integrated memory controller that supports DDR2 SDRAM will for now be based exclusively on this core .
The main innovation in the microarchitecture introduced by the core of the new revision was support for DDR2 memory. In the new core, AMD simply replaced the memory controller; fortunately, the Athlon 64 microarchitecture allows such changes to be made without any problems. At the same time, the new memory controller of the Athlon 64 family of processors is not backward compatible with DDR SDRAM. This means that from today DDR memory can be classified as an obsolete solution. Modern platforms from leading processor manufacturers AMD and Intel are now unanimous and require the use of DDR2 SDRAM. Obviously, this should affect the reduction in price of such memory, and in the very near future the cost of DDR2 SDRAM will be set at a lower level than the price of DDR memory modules of the same size.
Returning to the issue of support for DDR2 SDRAM by the memory controller of the revision F kernel, it should be noted that it officially supports memory with a frequency of up to 800 MHz. In other words, AMD managed to implement support for DDR2-800 SDRAM in its platforms before Intel. Naturally, new AMD processors are also compatible with slower DDR2 memory with frequencies of 667 or 533 MHz. But, given the fact that low memory latency is primarily important for the K8 architecture, it is the use of DDR2-800 SDRAM that can give the maximum effect in terms of performance.
It should be noted that traditionally the memory controller of the new core is equipped with a slightly larger number of dividers for the DDR2 operating frequency than is listed in the official specification. Thanks to this, some motherboards will be able to support Athlon 64 family processors for Socket AM2 systems even with DDR2-1067 SDRAM, without overclocking the clock generator. But for now, AMD does not declare work with memory faster than DDR2-800 in its official documents.
In addition to support for DDR2 SDRAM, the revision F kernel boasts some additional innovations. Thus, processors of the Athlon 64 family for the Socket AM2 platform now support virtualization technology, codenamed Pacifica. This is a symmetrical response to Intel technology VT, which appeared in Intel processors with the Presler core.
An equally important circumstance associated with the transition of AMD processors to the revision F core was the reduction in their power consumption. Despite the fact that AMD continues to use the old technological process With 90 nm manufacturing standards (with SOI and DSL technologies), Socket AM2 processors have lower heat dissipation and power consumption than their Socket 939 counterparts. Formally, the transfer of dual-core processors of the Athlon 64 X2 line to a new core made it possible to lower the maximum heat dissipation limit by 19%, from 110 to 89 W, and the maximum heat dissipation of single-core Athlon 64 processors, thanks to the revision F core, was reduced by 30% - from 89 to 62 W.
This increase in efficiency is an equally important improvement of the new core, along with the transition to support for DDR2 memory. Especially in light of the fact that the performance per watt ratio is currently being actively promoted by CPU manufacturers as the main metric for assessing the consumer qualities of their products.
However, the indicated reduction in heat emission of mainstream AMD processors is not all. The fact is that with the release of the Socket AM2 platform and with the manufacturer’s transition to using revision F kernels at the core of its CPUs, it became possible to release additional Energy Efficient processor lines. AMD is going to offer consumers two options for energy-efficient CPUs: with maximum heat dissipation limited to 65 and 35 W. Obviously, processors with a maximum heat dissipation of 65 W will compete with Conroe in terms of thermal and electrical characteristics, and 35-W copies will be intended for use in small, quiet and energy-efficient systems. AMD does not plan to use any special manufacturing technologies to produce energy-efficient processors. Such CPUs will be obtained by simply selecting crystals among all processors of revision F.
The transfer of AMD processors to the Socket AM2 platform will be massive. For the new platform, both dual-core Athlon 64 X2 processors, single-core Athlon 64 and budget Sempron processors will appear simultaneously. Therefore, the kernels of revision F will simultaneously exist in several forms. Possible options and their formal characteristics are given in the table below.


And this is what the core of the Athlon 64 X2 revision F processor looks like.

It should be noted that, despite the appearance of support for DDR2 SDRAM, the revision F kernel does not contain any fundamental improvements in terms of microarchitecture. Since the release of the first processors of the Athlon 64 family, AMD has avoided making any changes directly to the decoders or kernel execution units. That is, roughly speaking, so far we are seeing the development of the K8 architecture only along the extensive path of making small improvements. And this was quite enough for Intel to successfully compete. But now the situation is changing. Intel Core 2 Duo processors coming out this summer have a fundamentally new microarchitecture, distinguished by the ability to execute up to 4 instructions per clock cycle. And it will be quite difficult for AMD processors to compete with them, given that they do not have the same theoretical peak performance. From this point of view, the kernel of revision F, despite all the innovations present in it, is somewhat disappointing. To be honest, we would like more from it, primarily improvements at the microarchitecture level. But AMD engineers cannot offer us anything like that yet.

Socket AM2 platform

Let's take a closer look at what the new Socket AM2 platform offers the user, in addition to support for DDR2 SDRAM.
First of all, it should be noted that formally Socket AM2 is a 940-pin processor socket. At the same time, processors in the Socket AM2 version are neither logically nor electrically compatible with the old Socket 939 and Socket 940 connectors. To protect users from incorrect installation, Socket AM2 processors physically cannot be installed in old motherboards; they are located differently on them legs.

A positive aspect of the transition to Socket AM2 is that from now on AMD will offer a single platform for expensive dual-core and single-core budget processors. The same Socket AM2 motherboards can work with both Athlon 64 X2 and Athlon 64 and Sempron processors.
However, the introduction of a new processor socket does not yet sign the death warrant for the old sockets. AMD promises to continue supporting and supplying Socket 939 products as long as consumers remain interested in this platform.
Socket AM2 also sets new requirements for motherboards in terms of maximum power consumption and heat dissipation of processors. Although we said that the new CPUs with the revision F kernel can boast of reduced power consumption, the platform’s ability to support electrical powerful processors increased. Now the upper limit on current consumption is set to 95 A versus 80 A provided by Socket 939 motherboards. All this can make it possible to use processors consuming up to 125 W, while the maximum power consumption of the Socket 939 CPU was limited to 110 W.
Along with the new, more powerful power supply for Socket AM2 processors, motherboards offer a new cooler mounting mechanism. Now the frame on which the cooler is fixed is screwed to the motherboard not with two, but with four bolts. But at the same time, the fixing “teeth” on the frame remained in the old places.

This means that Socket AM2 motherboards can allow the use of older cooling systems, provided that they were mounted on a standard frame. The same heat dissipation systems that were screwed directly to Socket 939 motherboards cannot be used on new platforms without modification.

Processors for Socket AM2

In the table below we provide full list processors in Socket AM2, which will become available for sale after June 1.

It should be noted that the correspondence between the frequency, cache size and rating of the CPU for the Socket AM2 platform is the same as for Socket 939 processors. On the one hand, this will allow users to more easily navigate the characteristics of the new processors, but on the other hand, it clearly makes it clear that AMD does not expect a noticeable increase in performance from the transition to a new platform and processor core.
I would like to draw attention to the fact that support for the fastest memory, DDR2-800 SDRAM AMD, is declared only for dual-core processors. Single-core CPUs, according to the official specification, are only capable of working with DDR2-667 memory. This is quite logical, given the increased requirements of dual-core CPUs for memory bandwidth, at least due to the fact that RAM is directly involved in solving core cache coherence issues.
The line of Socket AM2 processors has been significantly expanded thanks to the emergence of energy-efficient processors with two new thermal packages - 65 and 35 W. These processors do not have as high frequencies as their “full-fledged” counterparts and are somewhat more expensive. However, they can be very attractive options in a range of applications, including small, low-noise computers. However, it is unlikely that these processors will favor the preferences of the majority of consumers, including enthusiasts. In other words, we do not yet expect widespread adoption of energy-efficient CPUs.
However, it should be remembered that processors with a reduced thermal package can be easily distinguished by their markings. While the third letter in the marking line of conventional processors is “A”, for a CPU with a thermal package of 65 W it will be changed to “O”, and the most economical processors with a heat dissipation limited to 35 W will be marked with the letter “D” .
Unfortunately, the appearance of processors in the Socket AM2 version will do little to increase the popularity of dual-core CPUs from AMD. The transition to a new platform, although it expands the range of the company’s dual-core offerings, does not entail a reduction in prices for processors with two cores. All Athlon 64 X2 processors will continue to be sold at prices over $300, which is unlikely to have a positive impact on their popularity. Especially considering the fact that Intel company, in light of the imminent appearance of CPUs with a new Core microarchitecture, has thrown a large number of cheap dual-core processors onto the market. For example, the cost of Intel's junior dual-core processor has already dropped significantly below $150. So from these positions, it is Intel that should be considered the main locomotive promoting dual-core CPUs to the market.

Test processors: Athlon 64 FX-62 and Athlon 64 X2 5000+

To test the performance of the new Socket AM2 platform, AMD sent us two processors: Athlon 64 FX-62 and Athlon 64 X2 5000+. The first of them is a dual-core processor aimed at gamers who are ready to do anything (financially) to achieve maximum performance, the second is the senior dual-core processor in the Athlon 64 X2 line.
Athlon 64 FX-62 has the highest frequency among new and old CPUs from AMD at 2.8 GHz. Moreover, it even caught up in frequency with the single-core Athlon 64 FX-57! However, this did not pass without a trace: the maximum heat dissipation of the new product is 125 W, which can be called a kind of record. There are no other equally hot processors among AMD products yet.

The CPU-Z diagnostic utility provides the following information about the Athlon 64 FX-62.

It should be noted that the standard supply voltage of the Athlon 64 FX-62 is 1.35-1.4 V, which is more than that of other dual-core CPUs in the Athlon 64 X2 line.
All this clearly indicates that the frequency potential of 90 nm cores with K8 microarchitecture is coming to an end. However, the results of overclocking the Athlon 64 FX-62 indicate that if you close your eyes to the growing power consumption, you can achieve more.
Thus, our test processor, when its supply voltage was increased to 1.5 V, was able to operate stably at a frequency of 3075 MHz, obtained as 15 x 205 MHz (Athlon 64 FX processors have a variable multiplier).

Heat was removed from the processor using a completely ordinary air cooler from AVC (part number Z7U7414002).

It must be said that overclocking the dual-core Athlon 64 FX-62 processor to frequencies above 3.0 GHz without using special cooling means is quite an impressive fact. Typically, all FX series processors with air cooling allowed increasing their frequency by only about 200 MHz. So, if desired, AMD will be able to increase the standard frequencies of its dual-core processors to 3 GHz. The only thing that can prevent this idea from being implemented is the excessively increasing power consumption and heat dissipation of the CPU. Thus, the power consumption of our test copy of the Athlon 64 FX-62, overclocked to a frequency of 3.075 GHz and operating under full load, according to measurement results, was 192 W (!), which clearly does not fit into the requirements that AMD itself has set for the Socket platform AM2.
The second processor we tested in our laboratory, Athlon 64 X2 5000+, has a standard clock frequency of 2.6 GHz, but is inferior to the FX-62 in terms of second-level cache memory. The cache memory of each of its cores has a volume of 512 KB.

The CPU-Z utility detects this processor as follows.

It is worth noting that all dual-core processors of the Athlon 64 X2 line, including the model with a rating of 5000+, have a supply voltage reduced to the range of 1.3-1.35 V. This, in particular, allows such processors to fit into a thermal package limited by the maximum heat dissipation in 89 W.
A comparison of the electrical characteristics of the new Socket AM2 processors measured in practice allows us to obtain a very interesting picture. As always in our tests, the processor load when measuring the maximum level of power consumption was performed by a specialized S&M utility, which can be downloaded here. As for the measurement technique, it, as usual, consisted of determining the current passing through the processor power circuit. That is, the figures given below do not take into account the efficiency of the CPU power converter installed on the motherboard.

We are already so accustomed to the fact that one of the characteristics of processors with the NetBurst microarchitecture is high heat dissipation. So the numbers shown in the diagram can come as a slight shock. But you can’t argue against the facts. The older AMD processor, Athlon 64 FX-62, today has slightly higher power consumption and heat dissipation than the older dual-core Intel processor, Pentium Extreme Edition 965, which is based on the Presler core revision C1. Approximately the same level of heat dissipation is now demonstrated by older processors in the mass dual-core lines, Athlon 64 X2 5000+ and Pentium D 960. Thus, older AMD processors can no longer be awarded the title of more economical. The latest CPUs from Intel, which are based on the most recent revision of the Presler core, are clearly no worse in this parameter. Thus, the Socket AM2 platform acquired increased tolerances for current and heat dissipation of processors for a reason.
However, let's return to considering the Athlon 64 X2 5000+ processor, namely, let's talk about its overclocking potential. This CPU must be overclocked by increasing the frequency of the clock generator; its multiplier is fixed at the top. However, this does not prevent you from achieving high results. By increasing the supply voltage of our test unit to 1.5 V, we were able to achieve stable operation at a frequency of 2.99 GHz.

The obtained results of overclocking two Socket AM2 processors using a simple air cooler suggest that the frequency potential of the CPU with the revision F core has become somewhat higher than that of previous AMD processors. Thus, the Socket AM2 platform may be quite interesting for overclockers.

Chipsets

Since the connection between logic sets and all processors with the K8 microarchitecture is carried out using the HyperTransport bus, and the memory controller is integrated into the CPU, the transition of the Athlon 64 family to the use of a new socket and DDR2 SDRAM memory does not require the use of any special logic sets. All those chipsets that were used in Socket 939 motherboards can be successfully used in Socket AM2 motherboards.
However, despite this, NVIDIA, which at the moment can be considered the leading supplier of chipsets for AMD processors, marked the release of a new platform from AMD with the announcement of new system logic sets for it. The new chipsets of the NVIDIA nForce family (nForce 590, nForce 570, nForce 550) are positioned by the manufacturer as “specially designed for new AMD processors.” However, there is nothing special in terms of processor support in these chipsets; they are notable only for their advanced capabilities. The simultaneous announcement of new NVIDIA chipsets and the Socket AM2 platform is just a marketing step.
However, switching to a new AMD platform will still require changing the motherboard. In this regard, new chipsets are quite in demand, because most users will probably want to get a new board with more advanced capabilities. It is precisely this category of consumers that the new chipsets from NVIDIA are designed for.
The line of new chipsets from the NVIDIA nForce family includes four products aimed at different target audiences.

All these logic sets are built on the same element base, the basis of which is the nForce 570 chipset. It should be considered the starting point from which the other products - nForce 590 and nForce 550 - stand.
The NVIDIA nForce 570 SLI chipset is a single-chip solution that can be called a further development of nForce 4 SLI.

This chipset supports SLI mode, but only under the PCI Express x8 + PCI Express x8 scheme.
A similar chipset, NVIDIA nForce 570 Ultra, is the same product, but without the ability to activate the SLI mode.

For the most “advanced” part of the gaming community, NVIDIA has also prepared the nForce 590 SLI chipset, which is capable of supporting SLI modes using the PCI Express x16 + PCI Express x16 scheme. In this implementation, to support a second PCI Express x16 graphics slot, the chipset includes an additional chip connected to the processor and MCP via a HyperTransport bus with a width of 16 bits in each direction and a frequency of 1 GHz.

As for the budget NVIDIA nForce 550 chipset, it is the same nForce 570 Ultra, but with somewhat reduced capabilities.

The formal characteristics of the new nForce family chipsets are summarized in the table below:

A study of the characteristics of the new NVIDIA chipsets for the Socket AM2 platform shows that they do not have many differences from the previous generation of chipsets of the nForce4 family. In fact, there are only three main improvements in the new chipsets:

Dual-port gigabit Ethernet controller;
Increasing the number of SATA channels to six;
Long-awaited appearance High Definition Audio.

It must be said that despite such a small list of improvements, NVIDIA presents the new chipsets as a huge step forward, which is facilitated by both the marketing emphasis of some features of the chipsets and the additional features being developed, implemented at the software level.
Without going into details, we note the main technologies present in the chipsets, which are a source of special pride for NVIDIA engineers:

LinkBoost. Automatic overclocking of PCI Express x16 buses to increase bandwidth between GeForce video cards installed in the system;
SLI-Ready Memory. Another name for the previously announced Enhanced Performance Profile technology, which allows the use of memory modules with expanded SPD content, in which, in addition to the main timings, the optimal voltage of the modules and the values ​​of secondary parameters are preserved.
FirstPacket. A technology that allows you to assign high priority to network packets generated by certain applications. NVIDIA uses it to reduce pings in gaming applications.
DualNet. The chipset's dual-port network controller allows you to use both ports separately or together for one connection.
TCP/IP Acceleration. Part of the TCP/IP packet processing procedure traditionally performed by the driver network card, shifted to the hardware capabilities of the logic set.
MediaShield. The chipset's six-port Serial ATA II controller allows the formation of one or more RAID arrays of levels 0, 1, 0+1 and 5.

In addition, along with boards based on the new nForce 590/570/550 chipsets, NVIDIA plans to ship a new nTune 5.0 utility, which has now acquired new capabilities for monitoring and fine tuning systems.
One of the first motherboards based on the NVIDIA nForce 590 SLI chipset was the ASUS M2N32-SLI Deluxe, which we used in our tests.

How we tested

To test the performance of the new Socket AM2 AMD processors, we used the following set of equipment:

Processors:

AMD Athlon 64 FX-62 (Socket AM2, 2.8GHz, 2x1MB L2);
AMD Athlon 64 FX-60 (Socket 939, 2.6GHz, 2x1MB L2);
AMD Athlon 64 X2 5000+ (Socket AM2, 2.6GHz, 2x512KB L2);
AMD Athlon 64 X2 4800+ (Socket 939, 2.4GHz, 2x1MB L2);
Intel Pentium Extreme Edition 965 (LGA775, 3.76GHz, 2x2MB L2).
Intel Pentium D 960 (LGA775, 3.6GHz, 2x2MB L2).

Motherboards:

ASUS P5WD2-E Premium (LGA775, Intel 975X Express);
ASUS M2N32-SLI Deluxe (Socket AM2, NVIDIA nForce 590 SLI);
DFI LANParty UT CFX3200-DR (Socket 939, ATI CrossFire CFX3200).

Memory:

2048MB DDR400 SDRAM (Corsair CMX1024-3500LLPRO, 2 x 1024 MB, 2-3-2-10);
2048MB DDR2-800 SDRAM (Mushkin XP2-6400PRO, 2 x 1024 MB, 4-4-4-12).

Graphics card: PowerColor X1900 XTX 512MB (PCI-E x16).
Disk subsystem: Maxtor MaXLine III 250GB (SATA150).
Operating system: Microsoft Windows XP SP2 with DirectX 9.0c.

Testing was performed with the BIOS Setup settings of the motherboards set to maximum performance.

DDR2 vs DDR: did it make sense?

In anticipation of performance tests of new AMD processors for the Socket AM2 platform, we decided to pay special attention to finding out what can be gained in terms of performance for processors of the Athlon 64 family by switching them to use DDR2 SDRAM. After all, it’s no secret that platforms built on AMD CPUs are very critical of the latency of the memory subsystem. And the transition from DDR to DDR2 SDRAM, although it promises a significant increase in throughput, does not provide a gain in latency.
To get practical data that allows us to draw some conclusions about the benefits that AMD received from using DDR2 SDRAM in its systems, we assembled two similar systems with DDR and DDR2 memory and compared their performance when setting different timings and different memory bus frequencies. As central processing units During testing, we used an Athlon 64 FX-60 for Socket 939 and an Athlon 64 FX-62 slowed down to 2.6 GHz for Socket AM2. Note that for these tests we used 512 MB memory modules, that is, the total amount of memory in the test systems was 1 GB.
First, let's take a look at the results of synthetic benchmarks that measure practical memory bandwidth and latency.

The results obtained in practice confirm the theoretical speculations. DDR2 SDRAM has a higher bandwidth than regular DDR memory, which is higher the higher its frequency. But from the point of view of latency, the picture is completely different. DDR400 SDRAM, operating at minimal 2-2-2 latencies, can only compete with DDR2-800 SDRAM with fairly aggressive (for this frequency) 4-4-4 timings. DDR2-667 SDRAM with the minimum possible timings of 3-3-3 can only achieve approximately the same practical latency as DDR400 with latencies of 2.5-3-3; it cannot compete with fast DDR SDRAM. As for DDR2-533 SDRAM, in terms of latency this memory is guaranteed to be worse than any DDR400 SDRAM.

results SiSoftware Sandra 2007 are quite consistent with the data we obtained using another test, Sciencemark 2.0. In fact, we can already say that only those owners of Socket AM2 platforms who will use either DDR2-800 SDRAM or fast DDR2-667 memory with 3-3-3 latencies can gain performance benefits. The increase in performance in all other cases remains in question and will depend primarily on the nature of the tasks being solved.
From testing the parameters of the memory subsystem, let's move on to considering the speed of operation in complex tests.

The SuperPi test only exacerbates the above statements. Indeed, the Socket AM2 platform demonstrates greater performance than a Socket 939 system with DDR400 memory with 2-2-2 latencies only if it uses DDR2-800 SDRAM.

Some tasks demonstrate a rather weak dependence on the speed of the memory subsystem. However, the low efficiency of DDR2 SDRAM compared to fast DDR400 SDRAM can be seen here too.

Operation speed WinRAR archiver strongly depends on the performance of the memory subsystem. In this case, we see that this task reacts quite sensitively to the growth of throughput. But despite this, only DDR2-800 with 4-4-4 timings manages to show slightly higher results than the Socket 939 platform with memory with 2-2-2 timings demonstrates.

The same can be said when looking at gaming performance. Even the slowest DDR400 memory turns out to be better than some types of DDR2 SDRAM.
So, answering the question posed at the beginning of this section, we can say that there is no direct sense in moving to DDR2 SDRAM in terms of increasing platform performance. Another thing is that the transition to support a newer memory standard may be useful from the point of view of future prospects. The development of DDR SDRAM came to an end, and both manufacturers and JEDEC focused on developing fast memory standards based on DDR2. That is why the choice of AMD should be considered correct. The company waited until DDR2-800 SDRAM, which did not reduce platform performance, became widely available on the market, and switched to new standard memory, looking into perspective. By the way, DDR2 memory is a significant advantage compared to DDR SDRAM in light of the imminent release of a new generation operating system Windows Vista Better availability of large memory modules should also be considered.

Performance

Synthetic tests: PCMark05, 3DMark06 and ScienceMark 2.0

First of all, we decided to test the performance of the processors in question using common synthetic tests.

It should be noted that there is nothing fundamentally new in the results obtained. As shown above, switching AMD processors to use DDR2 SDRAM gives a small performance gain. Therefore, the high level of performance of the new CPU Athlon 64 FX-62 is explained primarily by its high clock frequency of 2.8 GHz. The performance of the Athlon 64 X2 5000+ processor in some cases is inferior to the speed of the Athlon 64 FX-60, since, despite the same clock frequency, this CPU has half the cache memory. However, in those tests for which the size of the cache memory is not important, the Athlon 64 X2 5000+ can outperform any of the Socket 939 CPUs, since in the tested configuration it is equipped with high-speed DDR2-800 memory.

Overall Performance

We assessed overall performance in applications for creating digital content and in office tasks using the SYSMark 2004 SE test, which, moreover, actively uses multithreading.

When working with digital content, AMD processors significantly outperform competing Intel CPUs. As for the new Socket AM2 platform, it does not present us with any surprises in this case.

In office applications, the amount of cache memory is great importance. Therefore, the Athlon 64 X2 4800+ processor for Socket AM2 systems is ahead of the Athlon 64 X2 5000+. I would also like to note the fairly high results shown in this benchmark by the Intel Pentium D 960 processor. As can be seen from the diagram, it is inferior in performance only to AMD FX series processors, which have a much higher price.

Audio and video encoding

When encoding audio and video using DivX, iTunes and Windows Media Encoder we are able to observe a fairly tangible advantage of the new Socket AM2 platform. Streaming video encoding is a task that responds well to increased memory bandwidth. Accordingly, in these tasks the speed of Socket AM2 processors turns out to be approximately 2-4% higher than the speed of Socket 939 processors with similar characteristics.
Apple Quicktime is less enthusiastic about the new platform. When running Socket AM2, the Athlon 64 4800+ processor is even slightly behind its Socket 939 counterpart. However, in any case, we are not talking about fundamental differences in performance even when working with streaming data.

Image and video processing

Until recently, the Intel Pentium Extreme Edition processor remained the unsurpassed leader in Adobe Photoshop and Adobe Premiere. But the release of the high-speed AMD Athlon 64 FX-62 processor changed this state of affairs. Now this processor from AMD receives the title of the fastest product for image processing and non-linear video editing.

Performance in 3ds max 7 and Maya

Unfortunately, the increase in frequency to 2.8 GHz for the Athlon 64 FX-62 processor is not enough to compete with the Pentium Extreme Edition 965 in the final rendering in 3ds max. The thing is that rendering is a highly parallelizable task that can fully load all four virtual cores that a top Intel processor has. However, when rendering in Maya, this picture is not repeated; older dual-core processors from AMD are in the lead in this package.
As for the effect of using AMD DDR2 SDRAM processors, in this case we can talk about its absence or even negativity. Anyway, final rendering- This is not a task for which supporters of AMD processors should switch to a new platform.

3D games

Quite a noticeable performance increase from switching to DDR2 memory can theoretically be obtained in games. The fastest DDR2-800 SDRAM can provide a visible speed increase, reaching 6-7% in some games. However, we are not talking about the qualitative superiority of the new platform yet. At the same time, preliminary test results of the promising Conroe processor show that it will provide a qualitative leap in performance for Intel processors in gaming applications. In other words, although AMD processors continue to maintain a confident lead in games, this balance of power could easily change in the near future. And supporters of the AMD platform need to be mentally prepared for such a turn of events.

Other Applications

Since the performance of the Socket AM2 platform in comparison with the performance of desktop CPUs that support DDR SDRAM seems to be a very interesting question to study, we decided to add several more common programs to the number of test applications.
Using the 7-zip archiver, which very effectively supports multithreading, we measured the speed of data compression and expansion.

We assessed the speed of optical text recognition using the popular ABBYY Finereader 8.0 package.

In addition, we also tested the speed of test systems in the popular computer algebra package Mathematica, a new version which became capable of taking advantage of multi-core CPUs.

conclusions

Summing up everything that has been said about the new platform from AMD, we can only admit that the support for DDR2 SDRAM introduced in it is a small evolutionary step forward. Tests show that you should not expect any performance boost from simply changing DDR SDRAM to DDR2 SDRAM. Moreover, in order to see at least some effect from replacing the memory, in tests it is necessary to use the fastest DDR2 SDRAM with a frequency of 800 MHz and minimal timings. The currently widespread DDR2-667 SDRAM may not at all allow for a performance increase compared to Socket 939 platforms equipped with DDR400 SDRAM with low latency.
In conclusion, I would like to add that the emergence of the Socket AM2 platform working with DDR2 SDRAM should not be assessed as an ordinary event. Despite the fact that at the moment Socket AM2 systems do not have obvious and indisputable advantages over the Socket 939 platform, in the future the effect of this transition will become more than clear. Undoubtedly, DDR2 memory is much more promising today. It more dynamically increases its frequency and bandwidth, becomes cheaper faster and, in addition, allows you to create DIMM modules of larger capacity. As a result, AMD will undoubtedly benefit from the fact that it relied on DDR2. Moreover, at a very opportune moment: now no one will scold the manufacturer for such a step, either from the standpoint of performance or from the point of view of the price aspect.
However, at the moment AMD is not experiencing real pressure from Intel. Processors from this manufacturer continue to be leaders in almost any application. This is also facilitated by an increase in the frequency of older models of dual-core processors Athlon 64 X2 to 2.6 GHz, and Athlon 64 FX-62 to 2.8 GHz. Of course, there is a danger that the current state of affairs will be reversed with the advent of new Intel processors with Core microarchitecture. However, it is too early to talk about this yet.
I must say that after getting acquainted with AMD processors with the revision F core, some disappointment remains in my soul. The fact is that the company's engineers once again got away with cosmetic alterations and abandoned deep microarchitectural improvements. It is precisely this attitude of AMD towards improving its own processors that will sooner or later lead to the fact that the Athlon 64 family will lose the “arms race” to competing processors. Unfortunately, at the moment there is no information about the planned significant changes in the K8 microarchitecture.

Then for the initial assessment you can use the table:

	Motherboard AM2	Motherboard AM2+	Motherboard AM3	Motherboard AM3+	Motherboard AM4	Motherboard FM1	Motherboard FM2	Motherboard FM2+
Processor AM2
Processor AM2+
Processor AM3
Processor AM3+
Processor AM4
Processor FM1
Processor FM2
Processor FM2+

After the initial comparison, you must definitely check the availability of a specific model in the compatibility lists of the motherboard manufacturer.

What is SocketAM4 and what is it compatible with?
SocketAM4 is an AMD processor socket for high-performance processors with Zen microarchitecture (Ryzen brand) and subsequent ones. Processors with this socket have 1331 pins, support DDR4 memory and contain up to 24 PCI-E 3.0 lanes. Processors with Socket AM3+/FM2+ are not physically compatible with AM4 motherboards; in addition, the mounting of the processor cooling system has changed, and a new cooler will be required for the new socket. Both high-performance processors without built-in video and APUs with built-in graphics are available for Socket AM4.

At what frequencies does RAM run with AMD Ryzen?
As you know, AMD Ryzen processors work with DDR4 memory and have a built-in dual-channel memory controller. Depending on the number of modules per channel and memory rank, the memory operating frequency differs. This situation is not new - in server systems this has generally become a problem, so they came up with ones that, all other things being equal, work faster than “regular” RDIMM memory modules.
In any case, specifically AMD Ryzen processors work with RAM as follows:

Memory type	Number of modules per processor	Memory rank	Maximum memory speed
DDR4	2	Single Rank	2667 MHz
		Dual Rank	2400 MHz
	4	Single Rank	2133 MHz
		Dual Rank	1866 MHz

Will SocketAM4 processors be compatible with SocketAM3+ boards?
Will not. AM4 processors are both physically and electrically incompatible with legacy sockets.

What is SocketAM3+ and what is it compatible with?
, mechanically and electrically compatible with SocketAM3 (despite the slightly larger number of pins - 942, may also be called SocketAM3b in some sources), but designed to support new AMD Zambezi-core processors based on the Bulldozer architecture like the AMD FX 8150. All old ones are also supported by them, and, naturally, such boards only work with and are compatible with the previous ones.

Will SocketAM3+ processors be compatible with
Judging by all the signs, they won’t.(For example, due to the larger diameter of the processor legs.) A board on an older chipset that will be able to support SocketAM3+ processors after a BIOS update can be distinguished by the characteristic black color of the socket, but such boards may lose some of the functionality related to energy saving and monitoring. This information may be updated in the future.

What is SocketAM3 and what is it compatible with?
SocketAM3 is a further development of SocketAM2+, its main difference is the support of boards and processors with this type of DDR-III memory connector.
have a memory controller that supports both DDR-II and DDR-III, so they can work in SocketAM2+ boards (the performance of a specific processor in a specific board must be checked using the CPU Support List on the motherboard manufacturer’s website), but the reverse situation is not possible, SocketAM2 and SocketAM2+ processors do not work.

What types of memory do boards with SocketAM3 support?
- Only DDR-III with frequencies from 800 to 1333 MHz, both unbuffered (“regular”) and with ECC, i.e. absolutely the same memory that is used by motherboards with LGA1155, LGA1156 and LGA1366 connectors for .
With currently produced SocketAM3 processors, memory of the PC10600 type can operate at the nameplate frequency of 1333 MHz only if one module is installed per channel, and when two modules are installed on each channel of the memory controller (when a total of three or four memory modules are installed), their frequency is forcibly reduced up to 1066 MHz.
Registered memory is not supported; memory with ECC (without Registered!) is supported only by Phenom II processors for this socket.
The memory organization is the same as in Socket939/940/AM2/1156, i.e. dual-channel, and to achieve optimal performance it is necessary to install two or four (preferably identical in pairs) memory modules in accordance with the instructions for the motherboard.

What is SocketAM2+, and how is it different from just AM2?
SocketAM2+ is an upgraded version of SocketAM2, featuring support for HyperTransport version 3.0 with a frequency of up to 2.6 GHz, as well as improved power circuits.
As a rule (exceptions are extremely rare and are associated with the individual characteristics of specific motherboards), absolutely all SocketAM2 processors work perfectly in all SocketAM2+ boards. The situation with backward compatibility is worse; not all SocketAM2 boards support SocketAM2+ processors (compatibility in each specific case must be checked on the motherboard manufacturer’s website), secondly, reducing the HyperTransport frequency leads to a noticeable drop in performance compared to the “native” SocketAM2+ board.
Also, when using Phenom SocketAM2+ processors, the boards allow you to use DDR-II memory of the PC-8500 type at the rated frequency without overclocking (when installing one module per channel).

What is Socket AM2?
– a new socket for AMD desktop processors working with dual-channel DDR-II memory, replacing Socket939.

How many legs does he have?
– 940, but it is in no way compatible with Socket940 itself (the legs are located differently), which is why it was called Socket AM2. (Its “descendants” SocketAM2+ and SocketAM3 also have 940 contacts)

Which ones are and will be produced for the new connector?
– Athlon64 (single-core, production will be discontinued in 2007), Athlon64 X2, Athlon64 FX (actually older versions of Athlon64 X2), Sempron (Athlon64 with a reduced second-level cache), corresponding Opterons will appear soon (actually Athlon64 X2 with ECC support ( not Registered!) memory)

What types of memory do boards with SocketAM2 support?
- Only DDR-II with frequencies from 400 to 800 MHz, specifically - PC4200 (533 MHz), PC5300 (667 MHz), PC6400 (800 MHz), i.e. absolutely the same memory that is used by motherboards with an LGA775 connector on Intel chipsets 945/955/965. Registered memory is not supported; memory with ECC (without Registered!) is supported only by Opteron processors for this socket.
The memory organization is the same as in Socket939/940, i.e. dual-channel, and to achieve optimal performance it is necessary to install two or four (preferably identical in pairs) memory modules in accordance with the instructions for the motherboard.
Installing high-speed memory modules such as PC6400, or modules with reduced timings, is justified only in the case of older models of dual-core processors - with single-core Athlon64 and Sempron, installing faster memory does not affect the overall performance of the system.

Do the Socket AM2 versions of processors differ from their Socket939 counterparts in anything other than the type of memory they support?
- No, no fundamental differences were found for users; moreover, the integral performance of systems with equal-rated and equal-frequency processors, but working with DDR-II and DDR memory, respectively, is generally approximately the same. But for Socket AM2, processors are and will be released that are basically absent in the Socket939 version, for example, Athlon64 FX62, Athlon64 X2 5200+, etc. SocketAM2 processors also support AMD Virtualization (“Pacifica”) virtualization technology.

Will new processor models for Socket939 be released?
- No, moreover, the production of both motherboards and processors for this socket has already ceased.

What chipsets are used in Socket AM2 boards?
- The same as in Socket754/Socket939, there is no fundamental difference between the sockets from the point of view of the chipset. But on the new generation of chipsets for AMD processors, boards with the old connectors will no longer be produced.

What coolers can be used with SocketAM2 processors?
- Coolers designed for Socket754/Socket939/Socket940 are suitable if they are attached to the plastic teeth of the fasteners installed on the motherboard, but previously released coolers that have their own fasteners to the motherboard cannot be attached to the AM2 socket due to changes in the number and location of fasteners holes. To use such coolers, you need to purchase their upgraded version or (possibly!) a separate mounting kit.
The power connector for the processor cooler on Socket AM2 motherboards is completely similar to the PWM 4-pin used in LGA775 boards and is compatible with older 3-pin connectors.

What power supplies can be used with Socket AM2 boards?
- The same as with Socket939/PCI Express boards, i.e. ATX 24+4, and in most cases - 20+4 if there is sufficient power reserve in the +12V circuit.

Many people, when assembling a PC, or when purchasing a ready-made solution based on a particular processor, are faced with the concept of “socket”. Let's guess: half have no idea what it is or what it is intended for. In this article we will look at what this term represents, as well as the main sockets of AMD processors.

The Reds have always been distinguished by a loyal policy regarding the replacement of processor sockets: maximum preservation of compatibility with obsolete chips, a single fastener for cooling systems (AM2-AM3+ generations), easy BIOS flashing and more. But how the company’s technologies developed is the topic of this article.

To put it very briefly, a socket is a special connector on the motherboard into which the CPU is inserted. This design is created as an alternative to soldering, which greatly simplifies chip replacement and system upgrades as a whole. The second advantage is the reduction in cost of MP production.

And now about the pulp. The socket “accepts” only a certain type of processor. In other words, the contact pad of various connectors is significantly different from each other. Moreover, the type of mounts for cooling systems also often differs, which makes almost all sockets incompatible with each other.

AMD processor sockets

We would like to present you with a list of the most current AMD processor sockets at the moment, as well as describe the supported technologies for each. The list will consist of the following candidates:

1. Socket AM4+;
2. Socket TR4;
3. Socket AM4;
4. Socket AM3+;
5. Socket AM3;
6. Socket AM2+;
7. Socket AM2.

Let's get to the educational program, gentlemen.

1.Socket AM4+

The AM4+ processor socket should theoretically debut in April 2018 to support 12nm Zen+ processors (but this is not certain). It is known that motherboards with this socket will support the new X470 chipsets, which indicates higher CPU overclocking to frequencies previously unattainable with the X370.

Additionally, there is support for XFR 2 and Precision Boost 2 technologies. A nice feature of the new product is full compatibility with all existing representatives of the Ryzen 1000 series. It will be enough just to update the UEFI-BIOS firmware.

There is no information about AMD processors on this socket yet.

2. Socket TR4

A completely new socket developed by AMD engineers in 2016 for Threadripper family processors and visually similar to SP3, but not compatible with Epyc models. The first LGA connector of its kind in “red” design for consumer systems (previously only PGA versions with “legs” were used).

Supports processors with 8-16 physical cores, 4-channel DDR4 memory and 64 PCI-E 3.0 lanes (4 of which are on the X399 chipset).

Processors running on this socket:

• Ryzen Threadripper 1950X (14 nm);
• Ryzen Threadripper 1920X (14 nm);
• Ryzen Threadripper 1900X (14 nm).

3. Socket AM4

A socket introduced by AMD in 2016 for microprocessors based on the Zen architecture (14 nm). It has 1331 pins for connecting the CPU and is the first connector from the company that supports DDR4 RAM. The manufacturer claims that this platform is unified for both high-performance systems without an integrated graphics core and future APUs. The socket is supported by the following motherboards: A320, B350, X370.

Among the main advantages, it is worth noting support for up to 24 PCI-E 3.0 lanes, up to 4 DDR4 3200 MHz modules in 2-channel mode, USB 3.0/3.1 (natively, not using third-party controllers), NVMe and SATA Express.

Processors running on this socket:

Summit Ridge (14 nm):

• Ryzen 7: 1800X, 1700X, 1700;
• Ryzen 5: 1600X, 1600, 1500X, 1400;
• Ryzen 3: 1300X, 1200.

Raven Ridge (14 nm):

• Ryzen 5: 2400G, 2200G.

Bristol Ridge (14 nm):

• A-12: 9800;
• A-10: 9700;
• A-8: 9600;
• A-6: 9500, 9500E;
• Athlon: X4 950.

4. Socket AM3+

This socket is also called AMD Socket 942. It is essentially a modified AM3, developed exclusively for processors of the Zambezi family (i.e., the familiar FX-xxxx) in 2011. Backwards compatible with the previous generation of chips by flashing and updating the BIOS (not supported on all MP models).

Visually different from its predecessor in the black color of the socket. Among the features worth noting is the memory management unit, support for up to 14 USB 2.0 and 6 SATA 3.0 ports. In parallel with the socket, 3 new chipsets were presented: 970, 990X and 990FX. Also available are the 760G, 770 and RX881.

Processors running on this socket:

Vishera (32 nm):

• FX-9xxx: 9590, 9370;
• FX-8xxx: 8370, 8370E, 8350, 8320, 8320E, 8310, 8300;
• FX-6xxx: 6350, 6300;
• FX-4xxx: 4350, 4330, 4320, 4300;

Bulldozer (32 nm):

• Opteron: 3280, 3260, 3250;
• FX-8xxx: 8150, 8140, 8100;
• FX-6xxx: 6200, 6120, 6100;
• FX-4xxx: 4200, 4170, 4130, 4100.

5. Socket AM3

A processor socket that first appeared on the market in 2008. Designed with low-cost to high-performance systems in mind. It is a further development of the AMD AM2 socket and differs from its predecessor, first of all, in support of DDR3 memory modules, as well as higher bandwidth of the HT (HyperTransport) bus. The socket is supported by the following motherboards: 890GX, 890FX, 880G, 870.

All processors released for socket AM3 are fully compatible with socket AM3+, when the latter only supports mechanical interaction (identical arrangement of PGA pins). To work on newer boards you will have to reflash the BIOS.

You can also install AM2/AM2+ family chips in the socket.

Processors running on this socket:

Thuban (45 nm):

• Phenom II X6: 1100T, 1090T,1065T, 1055T, 1045T, 1035T.

Deneb (45 nm):

• Phenom II X4: 980, 975, 970, 965, 960, 955, 945, 925,910, 900e, 850, 840, 820, 805.

Zosma (45 nm):

• Phenom II X4: 960T.

Heka (45 nm):

• Phenom II X3: 740, 720, 710, 705e, 700e.

Callisto (45 nm):

• Phenom II X2: 570, 565, 560, 550, 545.

Propus (45 nm):

• Athlon II X4: 655, 650, 645, 640, 630, 620, 620e, 610e, 600e.

Rena (45 nm):

• Athlon II X3: 460, 450, 445, 435, 425, 420e, 400e.

Regor (45 nm):

• Athlon II X2: 280, 270, 265, 260, 255, 250, 245, 240, 240e, 225, 215.

Sargas (45 nm):

• Athlon II: 170u, 160u;
• Sempron: 190, 180, 145, 140.

6. Socket AM2+

The AMD socket appeared in 2007. It is similar to its predecessor down to the smallest detail. Developed for processors built on Kuma, Agena and Toliman cores. All processors belonging to the K10 generation work perfectly on systems with an AM2 socket, but you will have to put up with a “cut” in the HT bus frequency to version 2.0, or even 1.0.

The socket is supported by the following motherboards: 790GX, 790FX, 790X, 770,760G.

Processors running on this socket:

Deneb (45 nm):

• Phenom II X4: 940, 920.

Agena (65 nm):

• Phenom X4: 9950, 9850, 9750, 9650, 9600, 9550, 9450e, 9350e, 9150e.

Toliman (65 nm):

• Phenom X3: 8850, 8750, 8650, 8600, 8450, 8400, 8250e.

Kuma (65 nm):

• Athlon X2: 7850, 7750, 7550, 7450, 6500.

Brisbane (45 nm):

• Athlon X2: 5000.

7. Socket AM2

It first debuted under the name M2 in 2006, but was hastily renamed to avoid confusion with Cyrix MII processors. Served as a planned replacement for amd 939 and 754 sockets. The socket is supported by the following motherboards: 740G, 690G, 690V.

As an innovation, it is worth noting support for DDR2 RAM. The first processors on this socket were single-core Orleans and Manila and dual-core Windsor and Brisbane.

Processors running on this socket:

Windsor (90 nm):

• Athlon 64: FX 62;
• Athlon 64 X2: 6400+, 6000+, 5600+, 5400+, 5000+, 4800+, 4600+, 4200+, 4000+, 3800+, 3600+.

Santa Ana (90 nm):

• Opteron: 1210.

Brisbane (65 nm):

• Athlon X2: 5050e, 4850e, 4450e, 4050e, BE-2400, BE-2350, BE-2300, 6000, 5800, 5600;
• Sempron X2: 2300, 2200, 2100.

Orleans (90 nm):

• Athlon LE: 1660, 1640, 1620, 1600;
• Athlon 64: 4000+, 3800+, 3500+, 3000+.

Sparta (65 nm):

• Sempron LE: 1300. 1250, 1200, 1150, 1100.

Manila (90 nm):

• Sempron: 3800+, 3600+, 3400+, 3200+, 3000+, 2800+.

Results

AMD are such entertainers. Perhaps they themselves are surprised at the number of processor architectures that they have developed over their long history. It is noteworthy that the vast majority of older processors still work and pair perfectly with newer motherboards (if we are talking about the gap between sockets AM2 and AM3).

The most progressive connector at the moment, AM4, and its successor, AM4+, should receive support at least until 2020, which indicates potential backward compatibility of platforms with some minor limitations in functionality.

The relatively long lifespan and good stability of the “5.0 method” led to the fact that we tested all current families of processors with its help (and in some cases, more than one or two representatives of each), and there was still time left to work on excursions into history :) In general, from a practical point of view, they are no less important than tests of new products - many old platforms still have and work, so the question of “how many grams” can be won with an upgrade does not apply to idle people. And to answer this accurately, you need to know both the performance of new processors and the level of outdated ones. You can, of course, use the results of tests conducted a long time ago, but they all relate to software versions that have been popular for a long time, and it tends to change. Therefore, new tests are needed. It is quite difficult to carry out - and the processors themselves still need to be found, and other environments must be prepared to meet the requirements of the methodology. Therefore, for example, within the framework of the main version of the testing methodology, we basically cannot touch on Socket 754, since it is impossible to find 8 GB DDR SDRAM and a board on which all this will work. There is a similar problem with Socket 939, but it’s possible to cope with the newer (but, in principle, equivalent to the previous one in terms of performance) AM2 platform. What we will actually do today, fortunately, we managed to find as many as five suitable processors. More precisely, seven, but two stood out too much from the general range in terms of performance, which is why they were considered last time. And today is the era of late AM2 and even AM2+.

Test bench configuration

CPU	Athlon 64 X2 3800+	Athlon 64 X2 5200+	Athlon 64 FX-62	Athlon 64 X2 6000+
Kernel name	Windsor	Windsor	Windsor	Windsor
Production technology	90 nm	90 nm	90 nm	90 nm
Core frequency, GHz	2,0	2,6	2,8	3,0
	2/2	2/2	2/2	2/2
L1 cache (total), I/D, KB	128/128	128/128	128/128	128/128
L2 cache, KB	2×512	2×1024	2×1024	2×1024
RAM	2×DDR2-800	2×DDR2-800	2×DDR2-800	2×DDR2-800
Socket	AM2	AM2	AM2	AM2
TDP	65 W	89 W	125 W	125 W

Unfortunately, we didn’t get our hands on a single single-core Athlon 64. More precisely, one was found in storage, but studying it showed that it was a model for Socket 939. Which is a pity, since at first only such models made it into the mass segment - on At the time of the announcement of the platform, the company estimated the minimum dual-core processor (which was 3800+) at as much as $303 (the reason is clear - there were still several months left before the release of the Core 2 Duo, and the Pentium D had lower performance than the Athlon 64 X2). But we found the legendary 3800+, and not even the ADA3800, but the ADO3800 - it cost $20 more, but had a TDP of only 65 W, which for that time was quite “cool” for a dual-core model.

Unfortunately, we could not find any other junior “classic” 90 nm dual-core processors or any representatives of the 65 nm process technology. So conclusions about the dual-core family will have to be drawn on the basis of the mentioned “initial” 3800+ and three models (since two of them appeared after this family lost the status of devices with maximum performance) of a high level: 5200+, 6000+ and FX- 62. Strictly speaking, we could do without the latter, since testing it will not bring us any exclusive information - the clock frequency is exactly in the middle between the two other participants. But we couldn’t pass by the processor, which at the time of the announcement was sold at a price of around 1250 (!) dollars, having the opportunity not to pass it. A legend after all. Although it has been greatly devalued over the past years, the processor once rightfully occupied its price level, being the most productive x86 solution on the market.

CPU	Phenom X4 9500	Phenom II X4 940
Kernel name	Agena	Deneb
Production technology	65 nm	45 nm
Core frequency, GHz	2,2	3,0
Number of cores/threads	4/4	4/4
L1 cache (total), I/D, KB	256/256	256/256
L2 cache, KB	4×512	4×512
L3 cache, MiB	2	6
UnCore frequency, GHz	1,8	1,8
RAM	2×DDR2-1066	2×DDR2-1066
Socket	AM2+	AM2+
TDP	95 W	125 W

And for comparison, two models of subsequent generations are already Phenom. The first damn thing is lumpy in the form of the Phenom X4 9500 and the breakthrough Phenom II X4 940. Again, the latter is not so interesting, since we tested the Phenom II line under AM3, and they differ only in the supported memory, but formally the 940 is the best that was made under AM2+. In practice, many boards with this socket can use more productive solutions, thanks to the backward compatibility of the two platforms, but the formal status is also a reason to get acquainted :)

As for the first Phenoms, we have a representative of the very first generation - with the so-called “TLB bug”. Its discovery forced the company to switch to the corrected B3 stepping (such models are easily distinguished by the fact that their number ends in “50”), and BIOS patches appeared to ensure stable operation of already sold processors. At one time, we tested one of the engineering samples of Phenom with the TLB patch enabled and disabled and came to the conclusion that its use reduces performance by an average of 21% (in some programs - several times). Well, since this error did not always spoil the user’s life with system instability, many, naturally, preferred to disable this fix if possible at their own peril and risk.

Unfortunately, using modern software, this is already very difficult to do, unlike the days of Windows XP - Microsoft built the bug fix directly into its OS. This started with SP1 for Windows Vista and, naturally, migrated to Windows 7. In principle, there are ways to disable this “parking brake”, but we did not do this, since most users do not do this. And from the point of view of testing processors in modern software Tweaks like these are not correct. But it’s worth remembering about their capabilities, if someone still has to use a computer based on the first generation Phenom (and, according to reviews, performance increases on models with the correct stepping). As well as the fact that simply disabling the TLB-patch in Setup when working under modern Windows operating systems no longer affects anything (we did a quick check of this to make sure). Or, by the way, this situation can be considered as another reason not to rush to install a new OS on old computer, and without that it’s not too fast so that there is a desire to work with the “latest” versions of application software on it - it’s better either “the old fashioned way”, or, after all, start an upgrade.

In general, this is the set of subjects. Strongly skewed in favor of the fastest models and generally not covering many of the once popular branches of the Athlon family tree, however, we will test what we managed to scrape together.

CPU	Celeron G530T	Celeron G550	Pentium G860	Core i3-2120T
Kernel name	Sandy Bridge DC	Sandy Bridge DC	Sandy Bridge DC	Sandy Bridge DC
Production technology	32 nm	32 nm	32 nm	32 nm
Core frequency GHz	2,0	2,6	3,0	2,6
Number of cores/threads	2/2	2/2	2/2	2/4
L1 cache (total), I/D, KB	64/64	64/64	64/64	64/64
L2 cache, KB	2×256	2×256	2×256	2×256
L3 cache, MiB	2	2	3	3
UnCore frequency, GHz	2,0	2,6	3,0	2,6
RAM	2×DDR3-1066	2×DDR3-1066	2×DDR3-1333	2×DDR3-1333
Video core	HDG	HDG	HDG	HDG 2000
Socket	LGA1155	LGA1155	LGA1155	LGA1155
TDP	35 W	65 W	65 W	35 W
Price	N/A(0)	N/A(0)	N/A()	N/A()

With whom to compare? We decided to take four processors from modern Intel products. Celeron G530T and G550 have the same clock speed as the Athlon 64 X2 3800+ and 5200+, respectively (the second pair also has the same “lower” level cache capacity; however, Celeron has a common L3, while Athlon has a separate one L2, but the number is the same). The Pentium G860 is no longer the fastest of Intel processors priced under $100, after the appearance of the G870, but it has exactly 3 GHz frequencies, like the 6000+. Well, to complete the picture, there is another energy-efficient processor, namely the Core i3-2120T, operating at a frequency of 2.6 GHz, fortunately, we recently compared it with the Core 2 Duo of the same era as the older Athlon 64 X2, and indeed A direct comparison of the equal-frequency G550, 2120T and 5200+ is extremely interesting and revealing. It is clear that all these models are a priori somewhat lower than the Phenom II X4, but we have already analyzed this family (albeit in a different design) in detail, and have also compared it with modern (and not so modern) Intel processors more than once.

CPU	A4-3400	A6-3670K	Phenom II X2 545	Phenom II X3 740
Kernel name	Llano	Llano	Callisto	Heka
Production technology	32 nm	32 nm	45 nm	45 nm
Core frequency, GHz	2,7	2,7	3,0	3,0
Number of cores/threads	2/2	4/4	2/2	3/3
L1 cache (total), I/D, KB	128/128	256/256	128/128	192/192
L2 cache, KB	2×512	4×1024	2×512	3×512
L3 cache, MiB	—	—	6	6
UnCore frequency, GHz	—	—	2,0	2,0
RAM	2×DDR3-1600	2×DDR3-1866	2×DDR3-1333	2×DDR3-1333
Video core	Radeon HD 6410D	Radeon HD 6530D	—	—
Socket	FM1	FM1	AM3	AM3
TDP	65 W	100 W	85 W	95 W
Price	N/A()	N/A(0)	N/A()	N/A(0)

And four more models from the AMD range. Firstly, A4-3400 and A6-3670K. The second, after a recent price reduction, “lives” at the level of older Pentiums, and the first is comparable to Celeron. In addition, the FM1 platform is interesting to us because it offers the buyer a good level of integrated graphics - higher than the discrete graphics from the AM2 heyday. Accordingly, if someone has not yet raised their hand, throw it away system unit five years ago, the cheaper FM1 can stimulate this process. An additional convenience is that both processors operate at a clock frequency of 2.7 GHz, i.e. exactly between 5200+ and FX-62. And two old Phenom IIs, operating at a clock frequency of 3 GHz, are also asking to be included in the list of test subjects: X2 545 and X3 740. From a practical point of view, of course, it’s too late to remember them, but from a theoretical point of view, they’ll do.

	Motherboard	RAM
AM2	ASUS M3A78-T (790GX)	8 GB DDR2 (2x800; 5-5-5-18; Unganged)
AM3	ASUS M4A78T-E (790GX)	Corsair Vengeance CMZ8GX3M2A1600C9B (2×1333; 9-9-9-24; Unganged)
FM1	Gigabyte A75M-UD2H (A75)	G.Skill F3-14900CL9D-8GBXL (2×1866/1600; 9-10-9-28)
LGA1155	Biostar TH67XE (H67)	Corsair Vengeance CMZ8GX3M2A1600C9B (2×1333/1066; 9-9-9-24 / 8-8-8-20)

A small note about the RAM frequency - although officially all dual-core AM2 processors support DDR2-800, for 5200+ and 6000+ the real memory frequencies are somewhat different from the theoretical ones: 746 and 752 MHz, respectively, which is due to a limited set of dividers (which we are talking about already mentioned last time). The difference from the standard mode, however, is small, but it may have an effect somewhere in comparison with the FX-62, which operates in a “canonically correct way,” since its frequency is completely divided by 400 (the 3800+ too, but, naturally, these “monsters” » a priori not competitors). And all Phenoms (both the first and second generations) support DDR2-1066, but only in the “one module per channel” configuration, which for obvious reasons does not suit us: the volume required “according to the standard” for the technique is 8 GB with two modules We were unable to provide it. In general, these are also small things, but we focus on them to reduce the number of subsequent questions :)

Testing

Traditionally, we divide all tests into a number of groups, and show the average result for a group of tests/applications in diagrams (you can find out more about the testing methodology in a separate article). The results in the diagrams are given in points; the performance of the reference test system from the 2011 sample site is taken as 100 points. 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 within the framework 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 presented both converted into points and in “natural” form.

Interactive work in 3D packages

The almost identical results of the three Phenom IIs once again show that these tests are unable to utilize more than two computation threads. It would seem that the ideal situation is for the older Athlon 64 X2 - high-frequency dual-core processors with a relatively large and fast L2. But... even the 6000+ lags behind not only the A4-3400 with a frequency of 2.7 GHz, but also the two-GHz (!) Celeron G530T, and the results of the others in this situation need not even be mentioned. In general, over the past years, processor architectures have stepped far forward (not overnight, but the overall progress is good), which cannot be ignored. There were, of course, extremely unsuccessful steps along this path, such as the first Phenom. The lion's share of responsibility for the failure of the 9500 lies with the TLB "patch", but even without this one cannot count on high results from the first K10 - low-frequency models with a small (by modern standards) cache memory capacity, and even slow. And the cores here, we repeat, are useless.

Final rendering of 3D scenes

They are useful in these subtests, but the Phenom X4 9500 still managed to overtake only a portion of dual-core processors, and even then not the fastest. The reason is simple - low frequency. And cache memory is important for these tasks. Although it is clear that be it a carcass or a stuffed animal These processors had to be released (at least for such loads), since the Athlon 64 X2 was even slower, and AMD did not have any other processors at that time. Later, Phenom II X4 turned out to be an excellent work on fixing the mistakes, so they are still relevant in the quad-core modification. By the way, the fastest processors for FM1 (Athlon II X4 651 and A8-3870K) in this group show a result of 124 points, i.e. almost the same as what became available to AM2+ “holders” almost four years ago. Not so bad, in general :) Well, unless, of course, you focus too much on the fact that the Core i7-920, which appeared at the same time at a fairly close price, is capable of 182 points.

Packing and Unpacking

A very indicative group of tests. Firstly, the terrible results of the Phenom X4 9500 were predetermined in advance: at one time, including a “patch” for the TLB slowed down the engineering sample three times. However, even without it, the Phenom at 2.6 GHz (and not 2.2 as here) was only slightly ahead of the Athlon 64 X2 6000+, so we can even say that its performance has improved slightly over the past years, the reason for which is the support for multi-threading new versions of 7-Zip. But it also did not allow (this is the second observation) the Phenom II X4 940 to overtake at least the three-core Phenom II X3 740, which has a higher cache memory frequency and works with faster DDR3 RAM. The third curious point is that the Athlon 64 X2 6000+ scores exactly 100 points: the same as the reference Athlon II X4 620, which operates at a lower frequency. But it cannot reach the Celeron and others like them at the same frequency. And the A4-3400 (2.7 GHz, 2x512 KB L2) is faster than the Athlon 64 X2 5200+ (2.6 GHz, 2x1024 KB L2).

Well, one more interesting result (albeit from a slightly different story): the Core i3-2120T is approximately equal to the Phenom II X3 740. Although the second has twice the L3 capacity, almost 15% higher frequency, and there are three cores, which, all other things being equal , still better than two cores with Hyper-Threading support.

Audio encoding

The cache is unimportant - pure mathematics, so the Phenom X4 9500 was able to demonstrate relatively good (within the scope of this article, of course) results: it outperformed all the processors we took for comparison that support a smaller number of computation threads, and also operate at a higher frequency Core i3-2120T not radically faster. However, the dual-core Pentium G860 is not at all much slower, and it also managed to overtake the equal-frequency triple-core Phenom II X3 740. Apparently, it is for this reason that the “classic” three-core processors have died for a long time (three-module FX are a slightly different story). And the Athlon 64 X2 6000+ managed to outperform the Celeron G530T and A4-3400: new instruction sets and other improvements in modern architectures are not used in these subtests, so the high frequency saved the day. Although, of course, if we remember that it is one and a half times higher than that of the 530T... But let’s not talk about sad things - there is already more than enough of it. In particular, all other Athlon 64s, including the once legendary FX-62, are, for obvious reasons, even slower. And the 3800+ is only slightly faster than modern single-core models (such as the Celeron G460/G465 equipped with HT support), despite the fact that there is no alternative to multi-core for this group of tests.

Compilation

For once, the FX-62 managed to beat both the Celeron G530T and the A4-3400 - a pyrrhic victory, but a victory. At least, compared to other groups of tests. Another thing worth paying attention to is that the FX-62’s results are closer to 6000+ than 5200+, although in terms of core frequency it is exactly in the middle between them - the features of the K8 line memory controller are of considerable importance under such a load. Accordingly, the defeat of the Phenom X4 9500 was predetermined - the TLB-patch “kills” L3 performance so much that only the presence of four cores allowed this processor to overtake the Athlon 64 X2 6000+ and even almost catch up with the Celeron G550. Well, we also had no doubt that the Phenom II X4 940 would be the best of all test participants - the frequency is high (the rest are either the same or slower), four full-fledged cores and 6 MiB of L3 speak for themselves.

Mathematical and engineering calculations

But here the benefit from multithreading is small, so the 940 only slightly outperformed the 545, but lagged behind the 740. However, this is also a good result, even if it is only suitable for intra-company competition - professional packages have a certain “pro-Intel” essence, and this is of no help no escape. But AMD clearly did not stand still - even though the A4-3400 loses to the Celeron, its “specific” (per unit clock frequency) advantage over the Athlon 64 X2 is about 20%.

Raster graphics

Some of the tests are multi-threaded, some are not, so among AMD products the Phenom II X3 looks quite sufficient to solve such problems: the 940 turned out to be only slightly faster than the 740 due to slow memory and lower cache frequencies, and the A6-3670K “hangs out” on that the same level due to the complete absence of the latter and a lower clock frequency. But, generally speaking, high-frequency Celeron and Pentium look best here, and low-frequency ones are also not bad. “Old” AMD processors cannot be saved by either the frequency or the number of cores - the Athlon 64 X2 6000+, which has become common, lags behind the A4-3400.

Vector graphics

As we have already established, these programs are undemanding in terms of the number of computation threads, but their performance depends on the cache memory, so it is not surprising that three equal-frequency Phenom IIs showed similar results with a slight loss of 940 - there the L3 frequency is 200 MHz lower . But this is just the Sandy Bridge level with a frequency of 2.6 GHz (i3 is slightly faster than Celeron due to the “extra” megabyte of cache memory), and one of the best Athlon 64 X2 managed to overtake only the A4-3400 and the two-GHz Celeron. The other representatives of the line are even slower, and for the Phenom X4 9500 such a load promises an inglorious defeat - the core frequency is low, and this is not the first time that the TLB patch has had a disgusting effect on cache memory performance. However, it is obvious that even without it we would have gotten results only slightly higher than those of the Athlon 64 X2 3800+, which is clearly not enough to compete with modern processors.

Video encoding

The Phenom X4 9500 once again managed to outperform some relatively modern dual-core processors: the cache doesn’t interfere much with it, and there are still four cores. But slow. Athlon 64 X2 cannot suffer from the “TLB bug” for obvious reasons, so this bug will also be fixed, but their cores are just as slow architecturally, and there are only two of them. And even the frequency doesn't help much. The results of the Athlon 64 X2 3800+ and 6000+ are especially indicative - they are almost twice as inferior to the equal-frequency Celeron G530T and Pentium G860. And the 5200+ is a third slower than the A4-3400 with a comparable clock speed. In general, the big thing can be seen from a distance - just a little over six years ago there was simply no line on the market better than the Athlon 64 X2, and now it is simply unable to compete even with budget models from both AMD itself and Intel. The Phenom II X4 940 is capable of this with ease, but it is a significantly newer processor, and its brothers now live in the budget sector. The Phenom II X4 955, for example, the company has been shipping in bulk since September for $81, but what distinguishes it from the 940? Only support for DDR3 memory and +200 MHz to cores and L3. By the way, we remember that at the time of the announcement the recommended price of 940 was neither more nor less, and 275 full dollars - quickly in modern world processors are being devalued :)

Office software

The vast majority of tests in this group are single-threaded, and do not use intensive improvements in modern architectures, so for such applications the Athlon 64 X2 is quite sufficient. Unless, of course, energy costs are a concern - the 6000+ traditionally lags behind both the G530T and the A4-3400, but these processors do not require hundreds of watts at all. It is clear that the “old people” are also not fully loaded with such work, so they will get by with a few dozen, but “a few” is more in their case. And you will also need some kind of additional video. But in general, it’s enough for work. Which is quite consistent with the fact that many people in offices still use a variety of Celeron or Sempron devices, even slower ones than we recently tested. Accordingly, the Athlon 64 X2 3800+ will be at least no worse, and if you use some kind of voracious antivirus, it will be much better :)

Java

The Phenom X4 9500 once again had a blast, since there are still four cores, and the cache memory and its performance are not particularly important here, but in its case, “to the fullest” only means a result equal to the Celeron G550. However, taking into account the fact that above, as a rule, everything was much worse, and such a victory over oneself (and over the patches) inspires respect. What about the other participants? As usual: the Athlon 64 X2 is unsuccessfully trying to catch up with at least some modern budget processor, and the Phenom II X4 demonstrates that it can very well be considered one :)

Games

There was a time when the Athlon 64 (not even the X2) were the best gaming processors. Now, let’s face it, even the Phenom II X4 and younger Core i3 can only apply for this position “through pull,” not to mention dual-core models. Modern dual-core models. And not the ancient ones, to which even laptop processors can be considered competitors only in the terminology of Russian tenders :) Regarding the Phenom X4 9500, we’d better refrain - just as in a hanged house it’s not customary to talk about rope, so in comments to the results of one of the most “cash-loving” groups should not remember the “TLB martyrs”.

Multitasking environment

By the way, even here this founder of multi-core AMD processors failed to overtake earlier dual-core models from the same manufacturer - the last Chinese warning to those who like to buy “cores for the sake of prospects” without regard to what kind of cores they are. Otherwise, everything is the same as usual - Athlon 64 X2 are unable to cope with at least a two-gigahertz Celeron or dual-core Llano (by the way, the younger Athlon II X2 has the same performance as the A4), and the Phenom II X4 940 is just a Phenom II X4 . Not a bad processor for about a hundred dollars, even if it was worth almost three hundred at one time - devaluation, sir.

Total

In the end, we have what was expected - a mishmash of one-, two-, and multi-threaded tests (which is, in fact, an exact projection of modern software; including that which is difficult to benchmark, and, therefore, test methods are equally doesn't fit well) did best processor for Socket AM2+ approximately equal to the equal-frequency Pentium. Two conclusions follow from this - good and bad. The first is due to the fact that the compatibility of this platform with AM3 is almost complete - unlike owners of LGA775 systems, owners of a good motherboard with AM2+ and a sufficient amount of DDR2 memory can upgrade their computer to a very good level. Not top-end, of course, but the Phenom II X6 1100T has a “weighted average” performance of 159 points, and the Phenom II X4 980 has 143 points. Minus the inevitable 5% (or so) for slower memory - we get somewhere between 150 and 135 points. And the maximum for LGA775 is 132 points. And even then, only if you’re lucky enough to find a Core 2 Quad Q9650 somewhere on the secondary market for a reasonable price, since “during its lifetime” it never dropped below $316 wholesale, and if it also works on an existing board: despite the name the same socket, LGA775 are four limited compatible platforms (however, problems are also possible with the oldest AM2 boards). AMD, on the contrary, continues to sell both the 980 and 1100T for $163 and $198, respectively. To a certain extent, it is a little expensive, but if you really want to “boost” the system by replacing only the processor, such costs may well turn out to be optimal (in any case, a new set of Core i5, a board with LGA1155 and memory will cost much more).

And now the bad news, which directly follows from the good news - using a board with AM2+ together with a processor for AM2 or AM2+ does not make any sense. And it’s not even necessary to take a closer look at the top-end models for AM3 mentioned above - besides them, AMD has a lot more in its assortment. And not only among new processors, but also among inventory in retail stores or on the secondary market. Where can you buy some kind of Athlon II X3 or even X4 very cheaply - since now the manufacturer values ​​the younger Phenom II X4 at only 80-90 dollars. Is there any reason? Yes, I have. After all, even the best Athlon 64 X2, as we saw today, is inferior to the A4-3400, and this processor is approximately equal to the Athlon II X2 215. Note that the X2 is also the best. Well, replacing, for example, an Athlon 64 X2 3800+ with a long-discontinued Athlon II X4 630 will simply double the average performance.

It is clear that all these arguments are justified only if the existing board supports AM3 processors: otherwise it is easier to change the platform (to LGA1155, FM1 or FM2 - without much difference). And it’s even more clear that it makes sense to bother with them only when the performance of the existing computer is no longer enough. In the end, many people still somehow use Pentium 4, Athlon XP, or Celeron and Sempron (and even slower ones than we recently tested). Accordingly, the Athlon 64 X2 3800+ will already seem to them something no less reactive than the famous Pink Panther (after all, even within the framework of AM2 it is 53 points versus 30 for the Sempron 3000+), and the owner of it will seem like a person taken to heaven in flesh, like one of the biblical prophets :) But that’s all.

Despite the fact that in the summer of 2006 the Athlon 64 X2 3800+ was a dream (and the Athlon 64 FX-62 was a pipe dream) of many users, today one can only look at their results with a grin or nostalgic sadness. Moreover, the process of devaluation began back in 2006 - the FX-62 was the “king of the hill” for only one quarter, after which it was inferior not even to the top-end, but only to the close Core 2 Duo (over the past years, the ratio, by the way, has not actually changed: according to the latest method, the FX-62 scored 73 points, and the E6600, above which there were also the E6700 and X6800, scored all 77). Well, later both companies went far ahead. Let's emphasize - both.

Of course, Intel's success looks more clear: the Celeron G530T has a frequency of only 2 GHz and a TDP of 35 W (including the graphics core). But the A4-3400 outperforms the same old guys to a similar extent. Yes, of course, it requires 2.7 GHz for this (that is, the specific performance is about a third lower than that of the “bridges”), and the thermal package is already 65 W, but the A4 has a rich internal graphics world that is more powerful. Moreover, both of these processors are not new products: they were announced last year and are already giving way to faster “successors” on the shelves, and AMD has launched a new architecture. It caused a lot of criticism at the start, but at least it all happened without the kind of scandal that accompanied the release of the first Phenoms. Moreover, it is worth noting that even if there were no notorious “TLB bug” and the need to fix it, Phenom X4 still could not count on good results. Simply because even the best model in the line with the index 9950 (which the company did not get right away) operated only at a frequency of 2.6 GHz. The closest analogue from the modern line is A6-3650 with the same frequency. And, by the way, the cache memory capacity is the same, despite the L3 of the first Phenoms - a total of 4 MiB in both. Even if the A6 had separate, but full-speed, the Phenom had only L2 as such.

Well, how the performance of the “old” and “new” AMD cores compares was clearly shown by today’s testing - the “extra” 100 MHz and the increased cache still did not prevent the FX-62 from falling behind the A4-3400 by almost 10%. Accordingly, a similar picture would be seen when comparing the Phenom X4 9950 with the A6-3650. The latter has a result of 110 points, i.e. the best that 9950 could hope for - 100 points. Reference. Which are typical for the Athlon II X4 620 (by the way, with the same frequency of 2.6 GHz; we have already seen something similar) or... Celeron G550/G555 :) What can we say in this case about the younger representatives of the line, where the frequencies are also low ? Let's say that without problems with TLB, the 9500 would have caught up with the FX-62 (at one time, our testing showed that the patch reduces overall performance by about 21%) - what would it change? Nothing!

In general, the best that can be said about the Agena chip processors are the debug versions of the Stars family, through work on which (and improvement of the technical process, of course) we managed to move on to the truly successful Deneb, which still remains relevant. No other advantages were found in them. Unlike FX, where it immediately became possible to evaluate not only the minuses, but also the pluses. And how AMD knows how to work on errors is very clearly seen in the example of the first and second generation Phenom. Well, there’s only a little time left before the release of Piledriver, so let’s cross our fingers and expect similar results :)

We thank the companies, "" and « »
for help in setting up test benches

Compatibility of Socket AM2, AM2+, AM3 and AM3+ processor sockets

Socket AM3+
Socket AM3+ is a continuation of Socket AM3, mechanically and electrically compatible with Socket AM3 (despite the slightly larger number of contacts - 942, it may also be called SocketAM3b in some sources). Designed to support new AMD processors based on the Zambezi core with the Bulldozer architecture (for example, AMD FX 8150). Socket AM3+ is compatible with Socket AM3 processors and coolers for Socket AM2/AM3.

Socket AM3
Socket AM3 is a further development of Socket AM2+, its main difference is the support of boards and processors with this type of DDR3 memory connector. Socket AM3 processors have a memory controller that supports both DDR2 and DDR3, so they can work in Socket AM2+ motherboards (processor compatibility must be checked on the CPU Support List on the motherboard manufacturer’s website), but the reverse situation is not possible, Socket AM2 and Socket AM2+ processors do not work in Socket AM3 boards.

Socket AM3 motherboards support RAM DDR3 with a frequency from 800 to 1333 MHz (including with ECC). With currently produced Socket AM3 processors, memory type PC10600 will operate at a nominal frequency of 1333 MHz only if one module is installed per channel, and when two modules are installed on each channel of the memory controller (when a total of three or four memory modules are installed), their frequency is forced down to 1066 MHz. Registered memory is not supported; ECC (non-Registered) memory is only supported by Phenom II processors for this socket. The memory architecture is dual-channel, so to achieve optimal performance it is necessary to install two or four (preferably identical in pairs) memory modules in accordance with the instructions for the motherboard.

Socket AM2+
Socket AM2+ is an upgraded version of Socket AM2. The differences are support for HyperTransport 3.0 technology with frequencies up to 2.6 GHz and improved power circuits.
Basically, all Socket AM2 processors work perfectly in all Socket AM2+ boards (there are exceptions related to individual technical features some motherboards). Not all Socket AM2 motherboards support Socket AM2+ processors (compatibility in each specific case must be checked on the motherboard manufacturer's website). Secondly, reducing the HyperTransport frequency leads to a noticeable drop in processor performance compared to Socket AM2+ motherboards. Also, when using Phenom Socket AM2+ processors, the boards allow you to use DDR2 RAM (for example, PC-8500) at the rated frequency without overclocking (when installing one module per channel).