Is socket fm2 suitable for. Compare AMD FM1 and FM2 sockets

Company AMD introduced the second generation of desktop APUs. chips Trinity are based on the improved Piledriver architecture and also have a powerful integrated video core. Mobile versions of new generation processors from AMD have been offered as part of laptops for almost half a year now. An attractive combination of consumer parameters allowed the company to increase its share in this segment. Let's see if the desktop versions of Trinity, designed for the new platform, will be so successful Socket FM2.

What are the new hybrid processors codenamed Trinity? In the maximum configuration, these chips include a quad-core x86 computing unit with the most advanced AMD architecture at the moment - Piledriver. This is a further development of the Bulldozer architecture, which is used for the fastest AMD FX series chips. In addition, a graphics core is placed on the chip, which the manufacturer refers to the Radeon HD 7000 series.

Trinity, although they are the successors of the Llano processors, however, there is practically nothing left in common between them. Both the computing part and the graphics part in this case are not just improved, they are fundamentally different. Perhaps the only thing that connects the APU of both generations is the 32-nanometer process technology, which is also used for Trinity. Of course, a more advanced technical process would be preferable here, but the production facilities of GlobalFoundries are not yet ready for the mass production of chips using technology thinner than 32 nm.

Trinity's die area is 246 mm² and contains 1.3 billion transistors, while the Llano chip's silicon wafer is 228 mm² and carries 1.18 billion transistors (after a recent update by the manufacturer). The packing density remained about the same, the area increased by about 8%, while the number of semiconductors increased by 10%. Given the timing of the development of the 32-nanometer process technology, we assume that the cost of production of crystals, if increased, then only slightly.

What's new in Trinity? The dual-channel DDR3 memory controller officially supports operation in modes up to DDR3-1866, while it also became possible to use modules with a reduced supply voltage (1.25 V). As you can see, almost half of the crystal is occupied by the graphic part. The integrated GPU has an architecture inherent in chips for discrete adapters of the family Northern Islands. An important innovation is the AMD HD Media Accelerator video encoding/decoding unit. The northbridge functions of the chipset are, of course, now integrated into the processor. In terms of computing power, Trinity has a couple of dual-core x86 modules. Within each of them, the cores are partially dependent, as they share some common resources, in particular, instruction prefetching and real number processing (FP) blocks. Each module has a dedicated 2 MB L2 cache segment. L3 cache memory is not provided here - this is the prerogative of the AMD FX series CPUs. For communication with external devices, the processor has at its disposal 24 PCI Express lanes. Note the support for HDMI, DisplayPort 1.2 and DVI interfaces.

Trinity processors initially operate at fairly high clock speeds. If the Llano chips have just approached the 3 GHz bar, then the older model from the new APU family normally operates at 3.8 GHz, with the ability to accelerate up to 4.2 GHz. Trinity received the latest modification of the dynamic auto-acceleration mechanism AMD Turbo Core 3.0, which, depending on the nature of the load, can automatically increase the frequency of the CPU. Each processor model has its own range: from 200 to 600 MHz.

Integrated graphics

Introducing the term APU(Accelerated Processing Unit), the company initially wanted to emphasize the importance of the integrated graphics unit. Integrated graphics core Trinity, dubbed Devastator, uses architecture VLIW4, which was used for the Radeon HD 6900 of the Northern Islands family. Obviously, the developers have not yet been able to optimize the new GCN (Graphics Core Next) architecture for the needs of the APU, which is used in GPUs for discrete graphics cards of the Radeon HD 7000 series.

Recall that the graphics part of the Llano chips has the VLIW5 architecture. The computing units it includes can theoretically perform more operations in parallel than those with VLIW4. However, in real problems the latter are more efficient. In addition, VLIW4 stream processors, all other things being equal, can operate at a higher clock frequency. It is quite difficult to draw parallels here, but some quantitative indicators are curious. In the full version, the Llano graphics core contains 400 computing units, while the Trinity GPU has 384, but in the latter case, the nominal operating frequency of the graphics unit is 800 MHz, while the predecessor has 600 MHz.

The Devastator core includes 24 texture units and 8 rasterizers. AMD emphasizes that in this case the tessellation processing unit is noticeably accelerated. A hardware block is allocated for working with video data AMD HD Media Accelerator, which includes the most advanced UVD3 video decoding module inherited from the Radeon HD 6000/7000 processor. In addition, the processor contains the AMD Accelerated Video Converter video transcoding unit. Functionally, it is similar to Quick Sync, which Intel uses in its processors.

In general, the Trinity graphics core has excellent functionality. It boasts full support for DirectX 11 with Shader Model 5.0, OpenCL 1.1 and DirectCompute 11. At the same time, the new APUs allow you to connect up to four independent display devices, moreover, support for Eyefinity technology is also announced. Also noteworthy is the support AMD Steady Video 2.0, which allows you to improve the quality of your video by helping to eliminate the effect of shaky images during handheld shooting.

Like its predecessors, Trinity processors have the ability to operate in Dual Graphics, combining the efforts of an integrated GPU with a discrete graphics card. However, in this case we are still talking about entry-level devices from the Radeon HD 6500/6600 lines.

To help the A10 chips, the manufacturer recommends using the Radeon HD 6670, for the A8 and A6 the Radeon HD 6570 is offered, while for the A4 - HD 6450. In fact, there is the possibility of using Dual Graphics mode, but in the current conditions, such combinations are interesting in cases where a potential owner of a Socket FM2 system already has a video card that can be used as an additional accelerator. The intentional purchase of an adapter of the required class for use in Dual Graphics mode, although it has the right to exist as a deferred upgrade option, however, in general, it loses to the idea of acquiring a faster graphics adapter, which will cost a little more, but in games it will be noticeably more productive than the proposed bundle.

Piledriver architecture

The Piledriver architecture is an upgraded version of the Bulldozer used for the Zambezi (AM3+) chips.

Branch prediction blocks, data prefetching have been improved, the efficiency of working with the second-level cache has been increased, the L1 TLB size has been increased, and the work of the INT and FP module load scheduler has also been improved. In addition, the new F16C instruction sets are now supported, as well as FMA3, which Intel plans to add to its Haswell chips. AVX kits are now available for new APUs, which were not supported by Llano chips. In general, Piledriver does not fundamentally differ from the Bulldozer architecture, it is a modified version with a number of improvements and cosmetic optimizations.

APU Trinity range

At the time of the launch of the new platform, the line of chips Trinity includes six models. Two quad-core processors A10 and A8, as well as one each A6 and A4. As you can see, the number of x86 blocks is not reflected in the name of the APU series. At the same time, there is a dependence of the chip belonging to one or another line, which is determined by the number of computing cores of integrated graphics: A10 - 384, A8 - 256, A6 - 192, A4 - 128. This is another clear example of how the manufacturer wants emphasize the importance of the graphic component.

Line flagship - A10–5800K– operates at 3.8 / 4.2 GHz, its integrated GPU contains 384 computers and operates at 800 MHz. The L2 cache is 4MB and the claimed power consumption is 100W. The second "ten" has the same characteristics, except for the frequency formula. For A10-5700 the base ones are 3.4 GHz, and the dynamic auto-overclocking limit is 4 GHz. This was enough to bring the TDP down to 65W. In A8 models, in addition to a reduced number of video core processing units from 384 to 256, its operating frequency is also reduced to 760 MHz. Formulas for x86 blocks: A8-5600K– 3.6/3.9 GHz, A8-5500– 3.6/3.8 GHz. Single-module A6 and A4 chips, in addition to losing two x86 blocks, have a common L2 cache of only 1 MB. The number of GPUs is reduced to 196 in the case of A6-5400K, and up to 128 - at A4-5300.

As for the cost of the new APUs, the Trinity chips actually play in the same price segment as their predecessors - $50-130. At the same time, the pricing system is interesting. Both A10s are priced at $122. Both the model with an unlocked multiplier and the chip with a lower clock speed and a blocked KU have one recommended cost, which nevertheless has a TDP of 65 W, instead of 100 W for the flagship. The situation is exactly the same with the A8 line of APUs - both models are offered at the same price of $101. For some, higher performance has value, for others, more economical options are preferable. And for those and for others, suitable processors will cost the same price.

As is the case with Llano processors, as well as devices from a competitor, models with the “K” index have an unlocked multiplier. Curiously, now the most affordable model with this feature costs only $67, while the price of the previous generation APU with a free multiplier started at $80. However, the A6-3670K is a quad-core model, while the A6-5400K only has one module with a couple of dependent modules.

For Socket FM2, processors with a disabled graphics core will also be available, which will complete the line of Athlon chips. Given the general concept of APU, it is obvious that separate crystals will not be produced for such models (although, given the area occupied by the GPU, this would make sense), for such processors chips will be used first of all, with certain problems in the graphic part, and if there are fewer of them than the market requires, then full-fledged crystals with a deactivated GPU will also be used.

Socket FM1 and Socket FM2 compatibility

Unfortunately for the owners of systems with hybrid chips of the first wave, the new APUs are neither forward nor backward compatible with the Socket FM1 platform. The processor socket, and, accordingly, the pins on the chip visually have a minimal difference (905 vs. 904), but a different arrangement of the "keys" does not even allow installing Trinity in the old socket.

(left - APU Trinity, right - APU Llano)

AMD for quite a long time gave evasive answers to questions about the compatibility of FM2 and FM1 sockets, so as not to indirectly reduce the demand for processors for the latter. Now this is not necessary. Given that the new APUs are fundamentally different from their predecessors at the level of architectures, it is not surprising that they have their own power subsystem features that were not taken into account in Socket FM1. It was this fact that forced AMD to change the platform.

Chipsets

Despite the fact that Socket FM1 and Socket FM2 are incompatible with each other, the chipsets used on the previous generation platforms are quite suitable for the new one. chips AMD A55, as well as AMD A75 we will see in the composition of motherboards for Socket FM2. In general, there is nothing to be surprised here. Considering that the core functions of chipsets are taken over by central processors, their role in modern platforms is largely reduced to servicing peripherals. And here innovation does not happen so often. If there are already certain complaints about the functionality of the AMD A55 (lack of SATA 6 Gb / s), then the AMD A75 cannot be called obsolete. The latter became the first chipset in the industry with an integrated native USB 3.0 controller. And the rest of the "kit" is quite on the level.

To make the announcement of Socket FM2 even brighter, AMD also introduced a new chipset that will be used for this platform - AMD A85X. One of its key differences from the A75 is the ability to split the PCI-E x16 bus into two devices (x8 + x8), and, as a result, the ability to create CrossFire configurations with a pair of discrete video cards. In addition, the A85X now supports 8 instead of 6 SATA 6 Gb/s ports and allows you to create RAID 5 disk arrays. It also provides FIS-Based Switching channelization capabilities. In terms of support and configuration of the USB bus, no changes: 4 USB 3.0 ports, up to 10 USB 2.0 ports and up to two USB 1.1.

The Socket FM1 platform did not provide the ability to use two graphics adapters in the system. Such configurations are the lot of rather enthusiastic gamers or experienced crunchers. Obviously, in the case of Socket FM2, AMD wants to make the most versatile platform that could interest users with different needs in terms of performance and functionality.

Upgrade prospects

Given the experience with the release of the first generation APU platform, AMD hastened to assure potential buyers of new solutions that Socket FM2- this is serious and for a long time. At least one more generation of hybrid chips will use this connector, and accordingly, they will be able to be installed on motherboards that will now go on sale.

The lack of upgradeability, and the very short life of Socket FM1 are important reasons for the generally restrained enthusiasm for the previous generation platform. Yes, we can agree that this is not the segment in which the issue of modernization is paramount. However, for users who pay money for a new solution, the prospect of an upgrade is often important even if in reality the need for it does not arise before complete obsolescence. With Socket FM2, everything should be in order in this respect. It will remain relevant for at least 2-3 years.

All motherboard manufacturers have already presented their solutions with Socket FM2 connectors. It is curious that vendors have focused on models with different chipsets. Someone presented a whole set of devices based on the most affordable AMD A55 and several boards based on the top AMD A85X, not at all attracting the A75, while someone, on the contrary, relied on the latest chipset, diversifying their offers based on it as much as possible. All this means that the range of devices for Socket FM2 will be very wide, so it will be easier for users to choose a device in accordance with their requirements. As for prices, in our opinion, the range here will be only slightly wider than in the case of motherboards for Socket FM1 - $50-120.

Processor AMD A10-5800K

The top model of the new APU Trinity line came to us for testing - AMD A10-5800K.

Motherboard Gigabyte GA-F2A85X-UP4

To study the Socket FM2 platform, we used the older model in the current line of motherboards from Gigabyte - GA-F2A85X-UP4 based on the new AMD A85X chipset.

The board complies with the latest specification Ultra Durable 5 which involves the use of high-quality energy-efficient components. Power stabilizer eight-phase (6+2). The power circuit uses powerful IR3550 assemblies, as well as chokes with ferrite cores. A digital controller is used to control the VRM parameters.

The layout of slots for expansion cards is optimal. Three PCI-E x16, the same number of PCI-E x1 and one PCI. The latter does not require an additional controller, since support for this bus is still implemented in AMD chipsets. Considering the number of PCI Express lanes, the nuances of using slots cannot be avoided. The first slot is set to full speed by default. When using two video cards, the first and second slots are switched to x8+x8 mode. The third full-format PCI-E x16 has a bandwidth of x4, while if the nearest PCI-E x1 is used, then the lower PCI-E x16 will also provide data transfer rates at the x1 level. Gigabyte GA-F2A85X-UP4 allows you to fully realize the advantages of the A85X chipset - the model allows you to create a configuration with two video cards based on AMD chips that will work in CrossFireX mode.

On board Gigabyte GA-F2A85X-UP4 there is a gentleman's set of overclockers - Power, Reset, Clear CMOS buttons, as well as a status LED. The board is expectedly equipped with two BIOS chips, and the graphic version of the 3D BIOS is used as a UEFI shell, which is conceptually already well known to us from the manufacturer's previous boards.

Of the interesting features of the model, we note the technology Dual Clock Gen. The board has a microcircuit with an additional clock generator (the main one is in the chipset). According to the manufacturer, it allows you to achieve stable operation at higher bus clock speeds (~135-150 MHz), which may be of interest to owners of APUs with locked multipliers who want to boost their processor. Although, of course, given AMD's pricing policy for Trinity chips, it is better for enthusiasts to initially look towards models with the "K" index.

The board has a full set of video outputs: DVI, HDMI, DisplayPort and D-Sub. In this case, you can simultaneously connect up to three display devices with any combination of interfaces. Note that the DVI port operates in Dual-Link mode, allowing the use of monitors with resolutions up to 2560×1600.

The disk subsystem will allow you to connect 8 drives via SATA 6 Gb / s: seven internal and one using eSATA. As for the peripherals, the user has six USB 3.0 ports. Four of them are implemented using the chipset, two more use an additional Etron EJ168 controller.

In general, the board leaves a rather pleasant impression. A decent set of features for an older solution, nothing superfluous and at the same time a good start for the future.

Performance

To evaluate the possibilities AMD A10-5800K, we have selected worthy opponents for him. First of all, it's a processor AMD A8-3850. This chip differs from the older model of the previous generation APU line (A8-3870K) only by a 100 MHz lower clock speed and a blocked processor multiplier, while the integrated graphics part is used by the most productive Radeon HD 6550D. From the main competitor, a model of the same price category is presented - a dual-core processor Intel Core i3-3220 from the new line of 22nm Ivy Bridge chips. First of all, let's check how the CPU block works.

Trinity's compute performance is on average slightly better than Llano's (+5-10%), although given the notable architectural differences, the difference may vary depending on the applications used. In some cases, first-generation APUs with four full cores can even be faster than a pair of dual-core modules running at a much higher frequency. In applied tasks, Trinity is not lost against the background of dual-core Intel Core i3, offering quite decent performance for its price. In single-threaded tasks, the processor from Intel will definitely have an advantage, the phenomenal efficiency of the Intel Core architecture makes itself felt. But in multi-threaded tasks, the number of computing units decides a lot, and here AMD's quad-core CPUs have an advantage. Of course, Intel processors with the same number of cores are even more productive, but they are significantly more expensive.

During the test of the new APU, we also decided to evaluate the effectiveness of the bundle CPU+GPU in applied tasks, using for these purposes the Musemage graphic editor, which uses the resources of the graphic core to perform various operations. The list of stages included the SVPMark benchmark, which also knows how to connect graphics for video processing.

The range of programs interspersed with heterogeneous computing is gradually expanding. Moreover, this is not only synthetic software for tests, but also applied applications. The pace, of course, leaves much to be desired, but there is hope that such developer initiatives will be strongly encouraged by hardware manufacturers. This is the rare case when the interests of both competitors coincide. Intel is also focusing more and more on the performance and capabilities of its integrated video with each successive architectural iteration. The Ivy Bridge chips here have noticeably succeeded in comparison with their predecessors, and in the expected Haswell, the graphics core should receive an even more significant increase in performance. In the meantime, AMD has a much stronger position here.

In 3D synthetics, Trinity has a very solid performance increase of 40-45%. Of course, in the overall standings, the increased performance of the x86 block is also taken into account, but this is not bad after all. 6000 points in 3DMark Vantage is almost the level of the Radeon HD 6570, that is, a discrete graphics card, which is now offered for $50-60. The indicators of Intel HD Graphics 2500 look noticeably more modest against the background of "embeddings" from AMD.

Intel offers separate processor modifications equipped with Intel HD Graphics 4000 graphics. In the case of dual-core models of the Ivy Bridge line, this is the Core i3-3225. It also has a working clock speed, like the Core i3-3220 - 3.3 GHz, but is equipped with a full-fledged graphics module with 16 computing units (the HD Graphics 2500 has only six), although it costs $ 20-25 more. At the time of writing, we did not have such a model, however, in order to include in the review not only the results of the Intel HD Graphics 2500, but also the results of Intel's most powerful integrated graphics solution at the moment, we used the Core i7-3770K. It appears only in gaming tests with embedded video. This will allow a more balanced assessment of the current position and potential capabilities of the integrated GPUs of both companies.

In real games, the A10-5800K is again very confidently ahead of the A8-3850. The advantage is no longer as great as in the case of tests from Futuremark, however, an increase of 25-35% can also be considered an excellent result. In addition, the average 30 fps at 1920×1080 resolution already allows not only to view pictures in not the most simple games.

Solutions from Intel are expectedly less hasty, especially in the case of lightweight GPUs. It would seem that Intel HD Graphics 4000 has just managed to get close to Llano's performance, as Trinity chips again make this mission impossible. We hope that with the release of Haswell there will be intrigue again.

Integrated video capabilities are highly dependent on the performance of the memory subsystem. Let's see how in the case of the A10-5800K RAM bandwidth affects gaming performance.

If we compare AMD processors under such conditions, then, as we see, in most cases, the A10-5800K has a slight advantage (2-5%). Mafia II, in which the system with the new APU received a 10% increase, can rather be considered an exception. In addition, reverse situations are also possible, as evidenced by the results in Lost Planet 2, where the A8-3850 outperformed the newcomer by almost 5%. However, in any case, the rivalry here takes place only between AMD chips. The results that a PC with a dual-core Core i3-3220 demonstrates are beyond their reach. The gap from the pursuers is 7-18%. Even despite the smaller number of compute units, the dual-core Ivy Bridge chip is extremely effective in games, and here even twice the number of compute units cannot help AMD processors. On the other hand, the difference does not look depressing and the main weather here is made by a discrete graphics card.

In general, the increase in computing performance of Trinity is relatively small and is on average at the level of 5-15%. Despite the fact that full-fledged Llano computing cores still turn out to be preferable to dual modules in some cases, due to internal architecture improvements, as well as higher clock frequencies, Piledriver-based chips manage to outperform their predecessors. The possibilities of integrated graphics pleased more. A 30% advantage over its predecessor, which before the advent of Trinity was a kind of benchmark in terms of the capabilities of the built-in GPU, inspires optimism.

Energy consumption

Having gained a general idea of the performance of the Trinity APUs, we were also interested in evaluating the power consumption of the new AMD processors. The declared TDP parameter for the A10-5800K is 100 W, let's look at the real performance in typical tasks.

During the load on the computing units (rendering in Cinebench), the consumption of Llano and Trinity is approximately at the same level. But the increase in the power of the graphics core did not go unnoticed. In games where it is the GPU that is more loaded, the power consumption of the A10-5800K is 18 W higher than that of its predecessor. The manufacturing process has remained the same, but higher clock frequencies make themselves felt. At the same time, it is worth noting that in rest mode, in which the processor is often found most of the time, the energy efficiency of the new APUs is higher. However, here it is worth making an allowance for the fact that different motherboards are used for both processors, which can affect the absolute figures.

Dual-core Intel Core i3 as a whole demonstrate exemplary efficiency. The CPU consumes a minimum of energy for computing tasks, but when evaluating performance in games, it is worth considering a significant difference in the speed of solutions.

Results

Platform Socket FM2 and processors Trinity are quite an interesting option for building sufficiently powerful multimedia PCs. Compared to its predecessors, the performance of Piledriver computing units has not increased as much, while the integrated graphics capabilities have been improved by a third, reaching the performance of entry-level discrete graphics cards. At the moment, this is a serious advantage of AMD solutions. At the same time, the range of Trinity chips is exactly the same as that of Llano. Considering the balanced price, they will look very organically as part of inexpensive universal solutions “for everything”. And although recently mobile systems have been increasingly purchased for such tasks, new APUs in the desktop version will also find their customers.

All motherboards with FM2+ and FM2 connectors are fully compatible with Socket FM2 processors(AMD Trinity and Richland, Ax-5000. Ax-6000 and AMD Athlon X4 7x0 / X2 3x0 series APUs) and CPU with Socket FM2+. But processors with Socket FM2+ (AMD Kaveri Ax-7000 and Athlon X4 8x0) can only be installed on motherboards with the same socket, but they cannot be run on FM2 boards.

Specially Socket FM2+ boards were developed for AMD Kaveri solutions on AMD A58, A68, A78 and A88 chipsets. They remain compatible with Socket FM2 processors, so we recommend buying them for a possible further upgrade (if you decide to go this way).

Is it worth building a computer on Socket FM2+?

Yes, is it even worth it to do such things? It's May 2016, when Intel released high-performance Skylake, and AMD is preparing to introduce the all-new AMD Zen and APU Excavator for Socket AM4 in the fall. The new platform should be some kind of revolutionary step, as marking time has led AMD to surrender its position in the market for both processors and video cards. Therefore, we do not recommend buying those processors that are sold now, since the competition towards the end of the year will be able to dot the "i" and lower the price tags to the proper level. If you really want to update your desktop in the near future, and replacing the processor with a more powerful one is not in your plans in the future, then you can take a closer look at Socket FM2+. But we are waiting for AMD Zen...

Historical study of the company's first integrated platform

As experience shows, articles devoted to testing "old" (by the standards of the computer market) systems are usually no less popular than reviews of "hot" new products. And no wonder: even when their owners are no longer satisfied with the current level of performance, it is still interesting to compare it with the demonstrated new computers - at least in order to understand what is worth switching to (and whether it is worth it). Of course, it is impossible to test absolutely everything released by manufacturers at least in the last five years, but some iconic processors are quite. Especially when they are interesting in themselves as stages in the development of the industry or allow us to draw conclusions about some other products. In particular, that is why we decided (since the opportunity arose) to repeat one test of the year before last, but using modern software. Yes, yes, we will again talk about the AMD FM1 platform.

Why go back to her? First, despite its short life, it was, one might say, a turning point in the development of the market: this is the first platform whose integrated graphics were not implemented on the principle of "what was", but really suitable for (albeit limited) gaming or "non-graphic computing". In 2011, this was fresh and relevant - we recall that the then Intel proposals supported technologies already existing in discrete GPUs only to an extremely limited extent. AMD, on the other hand, has implemented full functionality and performance at the level of junior discrete graphics cards of the same year, and not some distant past. Actually, even later, performance competition continued to remain only intra-company - especially if we consider the budget segment, in which FM1 could fully replace only FM2, and later FM2 +, but not the updated LGA1155 or LGA1150. Last year, however, processors with a more powerful GPU were released for the latter than in any AMD APUs, but they are significantly more expensive. And what about the budget processors for the latest LGA1151? Something is possible, but for this it is desirable to compare the solutions of both companies directly and on an equal footing.

The processor component of the first AMD APUs is also interesting in its own way, albeit archaic: it goes back to the 2009 Athlon II. Despite their venerable age, such processors are still used by many, so they are also worth testing. But it's not really necessary to do so. As earlier tests showed, the performance of the A4-3400 roughly corresponds to the younger Athlon II X2 215/220, while the analogue of the A8-3870K is the older processors on the same chip, already sold under the Phenom II X4 840/850 brand. Moreover, the correspondence in this case is almost complete: the same number of cores similar in microarchitecture (and, accordingly, in supported technologies) allows us to expect that even when the software changes, the processors will still behave in a similar way. So, having tested the two mentioned processors for FM1, we will get an estimate of the performance range of budget processors for AM3. And accurate enough. And Intel processors for the LGA775 platform also fall into the same range - somewhere from Pentium E5x00 to Core 2 Quad Q9500. Here the comparison, of course, is already more rough, but also worthy of attention.

In general, no matter how you look at it, it's worth spending some time on the very first generation of AMD APUs. Today we will deal with this.

Test stand configuration

CPU	AMD A4-3400	AMD A6-3500	AMD A8-3870K	AMD A8-7650K
Kernel name	Llano	Llano	Llano	Kaveri
Production technology	32 nm	32 nm	32 nm	28 nm
Core frequency std/max, GHz	2,7	2,1/2,4	3,0	3,3/3,8
Number of cores (modules) / computation threads	2/2	3/3	4/4	2/4
L1 cache (total), I/D, KB	128/128	192/192	256/256	192/64
L2 cache, KB	2×512	3×1024	4×1024	2×2048
L3 cache, MiB	-	-	-	-
RAM	2×DDR3-1600	2×DDR3-1866	2×DDR3-1866	2×DDR3-2133
TDP, W	65	65	100	95
Graphics	Radeon HD 6410D	Radeon HD 6530D	Radeon HD 6550D	Radeon R7
Number of GPUs	160	320	400	384
Frequency std/max, MHz	600	433	600	720
Price	-	-	-	T-12650703

For the reasons mentioned above, we are most interested in two processors, but we will test three (since they already exist), adding the A6-3500 to the list of test subjects. Also interesting in its own way, since it occupied a special position in the model line: a three-core (the only one of all) with a good (if not the best) GPU, TDP 65 W and massively available (unlike exotic quad-cores for this platform with such a heat pack). And again, in terms of gaming performance, at least some We need A6, but there are no others.

We will compare this trio first of all with the A8-7650K: this is a much more modern and serious solution from the company, but the slowest of the newer generations of processors we tested. Over time, we plan, if possible, to test cheaper offers for FM2 + (fortunately, this platform still retains good positions in this segment), but so far there are none - we will limit ourselves to an assessment from above: the old A8 versus the new one.

CPU	Intel Celeron G3900	Intel Pentium G3260	Intel Pentium G4500T
Kernel name	skylake	Haswell	skylake
Production technology	14 nm	22 nm	14 nm
Core frequency std/max, GHz	2,8	3,3	3,0
Number of cores/threads	2/2	2/2	2/2
L1 cache (total), I/D, KB	64/64	64/64	64/64
L2 cache, KB	2×256	2×256	2×256
L3 cache, MiB	2	3	3
RAM	2×DDR3-1600 / 2×DDR4-2133	2×DDR3-1333	2×DDR3-1600 / 2×DDR4-2133
TDP, W	51	53	35
Graphics	HDG510	HDG	HDG530
EU quantity	12	10	23
Frequency std/max, MHz	350/950	350/1100	350/950
Price	T-13475848	T-12649809	T-12874617

Plus three Intel processors: a modern Celeron and two Pentiums - one is just as modern, and the second is a little outdated, but processors for the LGA1150 platform are still popular. Why choose the Pentium G4500T? We need some Intel processor with a GT2 video core (which is now also available in the Pentium), but the older G4520 is a clear overkill, because in terms of processor performance it very often overtakes even modern A10s. So we decided to take a slower model, even if it was energy efficient - AMD and Intel's proposals have already diverged so much in this parameter that it still makes no sense to compare them directly.

Test Methodology

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

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

And the detailed results of all tests are available in the form of a complete table with the results (in Microsoft Excel 97-2003 format). Directly in the articles, we use already processed data. In particular, this applies to application tests, where everything is normalized relative to the reference system (as last year, a laptop based on Core i5-3317U with 4 GB of memory and a 128 GB SSD) and grouped by the areas of application of the computer.

iXBT Application Benchmark 2016

The A8-3870K's four "full-weight" cores still allow it to compete with entry-level dual-core Intel processors in these programs, but they are already slower than a couple of dual-threaded modules of modern FM2 + solutions. The successes of the rest of the subjects, of course, are much more modest. And the most noteworthy is that the A4-3400 is already about twice as slow as the Celeron G3900. What's wrong here? Both processors are commonplace dual-core models without any SMT technologies and operating at almost the same frequency, but they differ by half. So just counting the cores doesn't say anything about performance even in a multi-threaded environment: the level of old dual cores (recall that A4-3400 is also comparable to Athlon II X2 or Celeron/Pentium for LGA775) is about half that of modern ones. But we have not yet taken the oldest model - the first representatives of this class (such as Athlon 64 X2 or Pentium D) are even slower. And the first quad-core processors only roughly correspond to modern dual-core ones, which also gives food for thought.

Especially in those conditions when they fail to "turn around in full force" - as in Photoshop, for example. Note that in this group of applications, generally speaking, modern Celeron and Pentium do not shine for many reasons. But they “do not shine” against the background of peers, and not at all representatives of outdated architectures.

A single-threaded (mostly) application, where the new AMD microarchitectures do not look the best. The old ones are even more convincing to some extent - 3870K almost caught up with 7650K, despite the significantly lower clock frequency. But this has long been a struggle in the "basement", so you can not pay much attention to it: it works - and okay.

Audition is a little more loyal to multi-core processors, although in principle this does not change anything - only the A4-3400 looks even worse than in the previous case.

But in a simple multi-threaded integer, the old A6 and A8 are still quite good - despite their very advanced age, they can somehow compete with budget processors. But if there are only two cores (as in all A4s) or three low-frequency ones (a feature of the A6-3500), nothing good comes of it. As expected.

Due to the lack of a common cache memory, "athlon-shaped" and "during life" did not shine in such tasks, but, nevertheless, older models, as we see, can now compete at least with Celeron. The younger ones (which do not have a head start in terms of the number of cores, which affects the packing time) behave worse, but it cannot be said that it is completely terrible.

Already within the framework of AM3, the company provided its chipsets with support for the SATA interface, which was also preserved in the FM1 disk controllers, so, in principle, processors for the latest platform can normally “load” a fast solid-state drive with work, almost as good as modern devices. In more complex scenarios, nuances are possible, but from the point of view of normal domestic use, there are no problems.

As we have already noted, this program is not too good for "virtual multithreading" technologies, which played a bad joke on the new AMD A8: it turned out to be almost indistinguishable from the old one. However, the computing capabilities of either one or the other, and even more so of junior processors for FM1, are generally low from the point of view of today, so “serious work” is not their forte. But they get the job done. Slowly but surely.

So, what do we have in the bottom line? Even the A8-3870K is generally comparable only to modern Celerons. Of course, there are cases when it looks more or less good against the background of the latter, thanks to the presence of four cores, but it also happens that quantity can not be used, but with the quality, everything is clear. The funnier thing here, however, is not this, but the fact that AMD's progress in improving integrated platforms in general turned out to be almost worse than that of Intel, although the latter company is most often criticized. A8-7650K, of course, is not the fastest processor in the family, but even from Athlon X4 880K with a discrete video card and 16 GB of memory we got only 129.5 integral points - the A8-3870K already gave out only 20% less. Moreover, this is not a top segment at all - even initially, the processors were positioned approximately as competitors to the Core i3. The latter, we recall, grew up one and a half times, so they left to fight on other fronts. Mostly with themselves or with Intel processors of a higher class, but of earlier release years. But the "APU" remained almost at the same level in terms of processor performance, despite the change in architecture and other improvements. But perhaps progress has been more noticeable in other areas?

Energy consumption and energy efficiency

Actually, it is clearly visible - for the sake of which everything was started: the faster A8-7650K consumes energy much more carefully than the A8-3870K. Moreover, we note that the technical processes are in principle comparable: the processors for FM1 were the first to use the 32 nm process technology, and they managed to improve it only by one step. Moreover, it is small: Intel immediately switched from 32 to 22, and now to 14 nm, and AMD has mastered only the transition from 32 to 28 nm. Therefore, now there is no direct competition between companies. But do not forget that AMD also managed to limit the needs of their devices somewhat - it used to be even worse.

True, of course, against the background of what Intel has achieved, all successes are too lost. But they did something - so, well done. The first "APUs" were not only slow, but also very inefficient. For comparison - Core i3-2120 even in a system with a discrete graphics card (which, as we know, only spoils the results) had an "energy efficiency" score of 2.15 points, i.e. more than one and a half times higher than that of "peers » family A8. But so far, we have practically not touched on graphics, which was very weak in the early Intel processors, and AMD integrated platforms were purchased mainly for the sake of it. Let's see what's good for now.

iXBT Game Benchmark 2016

Usually we present in the articles the results of only those games that at least one of the participants copes with at least one resolution. In this case, we decided to move away from this practice, since we initially have a notorious favorite in the form of the A8-7650K, to which all the rest are not competitors. Therefore, we will consider in detail only those games that the A8-3870K can somehow cope with - there are not so few of them.

For example, the “tanks” that not the newest Intel processors can cope with in the minimum settings mode. When using the same video cards, they also turn out to be winners - because of the high "single-threaded" performance. But the power of integrated graphics is still different, which leaves its mark. In particular, in FHD mode, even the old A8-3870K easily beats all Intel processors with GT1 GPUs. Moreover, even the low-frequency A6-3500 under the same conditions outperforms the most modern Celeron, and, moreover, the Pentium for LGA1150. A4-3400 cannot accomplish such "feats", but you can play on it. And even try to do it in the "full" resolution mode - peers from Intel were not capable of this.

Things are much worse with "ships", however, in general, older models for FM1 cope with them better than modern Celerons, not to mention the "previous" Pentiums. The latter are completely inferior to the younger A6. Pentium G45x0 is faster, of course, and how many years newer. In general, only the A4-3400 unambiguously lost its position, but no one doubted this - even “during its lifetime” it belonged to the budget segment itself.

Both new Celerons and slightly older Pentiums in this, to put it mildly, not new game, if they can compete with anyone, only with the A4-3400. And in order to somehow catch up with the A8-3870K, representatives of the G45x0 family are already needed. So far this is how it is. What fades a little except perhaps against the background of the indicators of the new A8, but new ones - after all, we are studying five-year-old processors (if anyone has forgotten).

The A8-3870K nominally handled the game in HD - the Pentium G4500T did the same. It's clear that it doesn't matter not enough, but more - for processors for FM2 +, for example. And the Pentium G3260, announced at the beginning of 2015, looks very funny, but has not managed to catch up with the youngest A6 of 2011 in any way :)

In this case, things look a little better for Intel, but only if you do not remember the difference of several years. AMD also did not stand still, after all, so the new A8 went far ahead. Intel processors too - but in comparison with their own predecessors in the main.

The picture is already familiar: the Celeron G39x0 lags behind even the younger old A6, the Pentium G32x0 still loses to the no less ancient A4, the G4500T somehow struggles with the A8-3870K, and the A8-7650K rises menacingly above all this :)

Is it generally possible to consider FM1 a gaming platform as of today? No, of course not. Actually, even FM2+ is suitable for this role only conditionally - we have always adhered and continue to adhere to the opinion that if gaming is one of the intended purposes when purchasing a computer, a discrete video card has no alternative. But you can play some games (if restraint) play on the IGP. From the point of view of today's article, the most important thing is that to this day this five-year-old platform, in general, is not inferior to modern low-end Intel solutions. More precisely, Pentium and Core i3 with HDG 530 GPU are no worse than the older A8 for FM1, but all models up to HDG 510 inclusive (and the old “numberless ones”) at best reach the level of the younger A6. And then A4. That is, the backlog at one time was very good, which is not surprising - after all, even the A4-3400 has a complete analogue of the Radeon HD 6450, which is still de facto sold under the name Radeon R5 230. The built-in Radeon 6550D in the older A8 is closer to video cards of a slightly different level - about the Radeon HD 5570. In general, in those years such discrete video cards were in demand, but here an integrated solution. Which looks pale against the background of the new proposals of AMD itself, but after all, how many years have passed. And Intel processors are reaching this level only now, i.e., almost five years after the appearance of the FM1 platform, or about six years - if you count from the company's first GPUs integrated "under the cover" of the processor (albeit on a separate chip).

Total

The first thing to be sure to note in the conclusions is that we did not encounter any problems during testing, despite using the latest version of Windows and a modern set of programs. Yes, of course, video drivers for old "APUs" are already available only through Windows Update, but they are installed, and everything works fine - as in the case of Intel's Ivy Bridge (but with Sandy Bridge of the same 2011 as FM1, already there are some rough edges).

And in terms of hardware configuration, everything is also simple: completely standard (so far) DDR3 memory, conventional drives with SATA600 interface, built-in USB 3.0 support, and PCI and PCIe buses are used for expansion cards - there have been no significant changes on the market. The latter, by the way, allows you to "spur" the gaming performance a little if necessary, simply by adding a discrete graphics card. Of course, it makes no sense to set the expensive one, since the performance of solutions for this platform is still low - the expensive one will not be fully used.

To be fair, if we had tried this experiment in 2011, but with the 2006 system, we would have basically succeeded as well. Problems could arise with memory (due to the transition from DDR2 to DDR3 that occurred at the end of the 2000s), but not with other peripherals. But with a computer of 2001 in 2006, everything would be very difficult ... AGP for video cards, Parallel ATA for drives, already exotic SDRAM or RDRAM memory - why go far: in 2006, we used the x64 version of Windows for testing XP (and after all, Vista came out at the end of the year), and the first processors suitable for its operation appeared only in 2003. In general, just about 2005-2006. processes in the market were quite turbulent. After - one and a half changes in the type of memory (transition from DDR2 to DDR3 and the ongoing process of introducing DDR4), and leapfrog of processor sockets. Other interfaces have evolved already evolutionarily and with the preservation of compatibility. The software more or less stabilized in its requests, which grew only quantitatively (which, taking into account the compatibility of interfaces, was solved), but not qualitatively. And in some areas - and no quantitative changes were observed: the computer on which it was possible to install and comfortably use Vista does not cope well with the "top ten".

In general, there is nothing surprising that systems from five or even ten years ago are still in operation. Interestingly, the performance of processors from 2006 to 2011 grew faster than from 2011 to 2016, so it is, in general, far from it (despite crying and moaning about this in various forums and others like them). It is clear that all those processors are already either slow or very slow - a lot depends on the year. In particular, if we return to our today's heroine, the AMD FM1 platform, then in 2006 it would have been top-end (this is, of course, a hypothetical comparison, but according to earlier tests, processors for FM1 just correspond to the level of the best Core 2 Duo / Quad, and their video part is worthy of comparison with good discrete video cards of that time), in 2011 - budgetary and already only conditionally gaming, but today ... you yourself saw :) However, all investments in themselves such systems have long been recaptured, so what if the performance "does not press" - then why fix something that isn't broken? If something really breaks and / or ceases to suit for other reasons, then when buying a new computer, you can no longer worry about the choice. As you can see, even the integrated graphics of Intel processors have already pulled up to this level, and the new AMD “APUs” are even faster. In terms of processor performance, both those and others also "grew up" - albeit to varying degrees, but still. Thus, whatever you buy to replace the old system with FM1 will be at least as good, but at the same time cheaper. And if you are not limited to the cheapest offers, then definitely better. In general, you can not think about what was, but just buy what you need - as if there was no computer at all. Good news, in general.

Every time we buy an AMD-based computer, we wonder which processor and socket to choose? Especially now that AMD changes them almost every year. Will there be a prospect of replacing the processor in the future and what is the old processor good for? It is also important to know when there is a bunch of old hardware with different performance. And you need to assemble a computer of tolerable performance from all this. This table shows that the range for creativity is decent. Especially for overclockers and gamers, a large amount of iron accumulates. And it makes sense to rummage on the mezzanine and assemble, for example, a computer to the country house, or to a younger brother / sister.

CPU	motherboards
		AM2	AM2+	AM3	AM3+	FM1	FM2	+ - Compatible; - Theoretically compatible, but compatibility in each case must be checked on the website of the motherboard manufacturer; - Absolutely incompatible.
	AM2	+	+	—	—	—	—
	AM2+		+	—	—	—	—
	AM3			+	+	—	—
	AM3+	—	—	—	+	—	—
	FM1	—	—	—	—	+	—
	FM2	—	—	—	—	—	+

It is clear from the table that, unfortunately, contrary to popular belief, FM1 and FM2 sockets are absolutely incompatible. Here you need to choose whether to move a more expensive motherboard and a budget processor, or build a powerful PC, but on the previous socket. In my opinion, the decisions are equivalent. For example, you have purchased a powerful computer on an outgoing socket, it doesn't matter that you will use it for several years. Although if you assemble a PC on a new socket, there is a prospect of installing a more powerful and more economical CPU in a year.

AMD processors use different sockets than Intel-supplied models. Therefore, the choice of the processor itself is so important - it defines a set of additional components, like a motherboard, and at the same time can tie the user to one platform.

This socket is designed for the new group of Phenom II processors, as well as Athlon II, Sempron and Opteron. Entered the market in 2009. The most important change was the introduction of support for the DDR3 memory controller. AM3s are not compatible with older versions, even though the design difference is just the addition of a single pin.

socket AM3+

The AM3+ socket is a new, modified version of its predecessor (AM3), prepared with AMD Bulldozer architecture-based processors in mind. Unlike most Intel stands, the AM3+ is AM3 compatible, which means that processors based on the new socket can be installed in an AM3 socket. The only condition is to update the BIOS if the manufacturer decides to release a new version. However, keep in mind that not every processor will work in an AM3 package. Among the design differences, you can specify the addition of one contact. The new stand also introduces improved power regulation.

The processor families that can be installed in the AM3+ socket are Phenom II, Athlon II, and FX.

In motherboards with socektem, AM3+ also introduced new 990FX and SB950 logic chips, which provide support for octa-core processors. Support for HyperTransport technology version 3.1 has also been added. In addition, the first of them introduces the possibility of combining video cards in SLI.

The FM2 connector is used in motherboards designed to serve desktop APUs based on the Piledriver architecture. These include the Athlon 2 and Athlon X4 processors. There is also a new chipset - A85X. Compared to the FM1 connector, several contacts have been removed, leaving them in the amount of 904.

Socket FM2+

The new socket was created with the new Kaveri platform in mind. Contrary to previous company policy, FM2+ is only partially compatible. This means that owners will not be able to install new processors, however, nothing prevents them from installing an old processor in the FM2+ socket. There is also a new chipset: A88X.