In compliance with a fashionable trend to assemble system components into a platform for this or that target audience (sometimes giving it a loud marketing label), such "system" offers should be reviewed from a consumer point of view for all components as a unit. That's why first of all we shall sum up what new features are offered for the new processor socket.

DDR2 memory support

As is known, DDR2 technology theoretically possesses three indisputable advantages — it's more economical, it allows twice as much memory size on a single module, and it's cheaper to manufacture (however, the last fact hadn't been reflected in market prices for a long time). When ripened (memory itself and its infrastructure), it promised to excel the best DDR implementations in all performance parameters, that is to demonstrate higher bandwidth and lower access latencies.

The difference in memory power consumption in absolute values is evidently insufficient for taking DDR2 benefits for desktop computers seriously. And memory volumes of over 4GB still concern only the server segment. Besides, 2-GB DDR2 modules are currently too expensive and available only as DDR2-400. Thus, performance gain remains the only relevant point. But our tests demonstrated that the expected memory bandwidth values remained unattainable, though AMD engineers managed to cope with latencies.

We can only establish a fact that neither Intel nor AMD managed to reveal the full potential of this memory type (three years on the market). In the first case, the problem is in a relatively high latency of the external memory controller and limited FSB bandwidth. In the second case — architectural limitations of the processor and, probably, of the current revision of the DDR2 controller. Thus, in future, Intel will have to raise FSB clock to 1600 MHz and AMD will have to make some changes in the CPU architecture to utilize resources of dual-channel DDR2-800.

Can AMD do without DDR2 support now? From the technical point of view, yes, it can. At least, it can easily delay it until the 65-nm K8 core or even until K8L processors and its top models (based on the current core) will still not be limited by memory performance. On the other hand, it has been delayed for too long already. And now support for DDR2 as well as DDR memory is objectively necessary to successfully fill the market with processors from Fab36. Moreover, it may be even more important for the budget sector, where attraction of a platform depends even less on performance nuances. But DDR2 price drops may become a significant factor.

Support for AMD Virtualization Technology

Previously known as Pacifica, the virtualization technology is supported by all processors for Socket AM2, except Semprons. Its technical implementation should be reviewed in a separate article, as the process of virtualization (that is, administrating access of programs, including OS, to CPU, memory, and peripheral resources) cunningly uses the integrated memory controller and Hyper-Transport bus.

No doubts, this technology is very useful in the server segment. What concerns desktop PCs, we cannot expect its wide usage at least until MS Vista. Running several operating systems on a desktop computer and switching between them in real time, or even restarting them independently of each other - it all sounds impressive, but such desktop environment is rarely necessary. In future, virtualization may increase resistivity of a computer to cracking and virus attacks. How will it be implemented exactly? We'll see. It's up to software.

Processors

The series has been supplemented with a couple of top Athlon 64 5000+ and FX-62 models, which should consolidate the advantage of the new platform over the old one using an old all-purpose trick — due to higher clocks. There is also a new Athlon 64 X2 4000+ model, which was skipped for Socket 939. It's addressed to those users, who want to buy a dual-core processor with maximum cache size for a relatively democratic price. Before this time, the line of processors with 2MB cache had started from the 4400+ model, except for server models. Single-core Athlon 64 models for Socket AM2 come solely with 512 KB cache and give ground to the expanding line of Semprons. Mass-production models have become more economic — all Athlon 64 X2 processors officially consume no more than 89 W, while 4600+ and 4800+ for Socket 939 officially consume 110 W. The limit for single-core models is lowered from 89 W to 65 W.

And finally, the main addition to this family is Energy Efficient processors. They differ from "usual" models with the same rating solely by reduced heat release. It's planned to print their TDP on the box to distinguish these models.

As we have already heard about the intention to launch processors with reduced power consumption, we were mostly interested in final prices. They are higher only by 5% for dual-core models with TDP of 65 W, which clearly means high percentage of such processors in the overall production. Models with 35W TDP are officially intended for computers in compact cases and media centers with high requirements to a noise level. Considering that the line of 35W models includes a dual-core model and that this level of power consumption is reached with the 90nm process technology, AMD engineers have all reasons to be proud.

Chipsets

As Socket AM2 processors do not require anything from a chipset for their support, the new motherboards will be based on models, which we have already seen in the previous platform. Nevertheless, replacement of a CPU socket is the best time for launching new chipsets. It's much easier to attract manufacturers (and customers) with a new product, when they have to think about creating a new design (purchasing) new motherboards anyway. So there appeared several new models.

ATI has timed the final release of its long-awaited SB600 southbridge to the launch of Socket AM2. It should help it overcome dependence on shipments of third-party southbridges (from ULi). The new series uses old Radeon 1100, 1600 and 3200 as northbridges. They cover the entire range — from integrated video to systems with CrossFire support and a couple of full-speed PCI Express x16. A reference motherboard (top model) has already appeared in our testlab. It will be tested in the nearest future.

NVIDIA has launched a series of discrete chipsets nForce 500, consisting of four models. The ranking system inside the series is inherited from nForce4. You can read the details on this chipset in a separate article. In this review we shall examine two MSI motherboards, based on two middle-end models — 570 SLI and 570 Ultra. Integrated video from NVIDIA on Socket AM2 is still based on well-known GeForce 6100 and 6150 together with southbridges nForce 410 and 430.

Performance

Testbed configurations:

• Socket AM2
  • CPU: AMD Athlon 64 X2 4000+
  • Memory: 2 x Corsair XMS2 PRO PC2-6400 (DDR2-800, 5-5-5-12, we used more aggressive timings to "equalize" the testbeds, they are already available as a standard scheme to elite DDR2 modules)
• Socket 939
  • CPU: AMD Athlon 64 FX-60 2.0 GHz
  • Memory: 2 x Corsair XMS3200 TwinX (DDR400, 2-2-2-5)
• HDD: Seagate Barracuda 7200.7 (SATA, 7200 rpm)
• Video card: XFX GeForce 7800 GTX, 256 MB DDR
• Power supply unit: Chieftec GPS-400AA
• OS: Windows XP SP2

We decided to take the ECS KN2 SLI Extreme motherboard on NVIDIA nForce4 SLI X16 for our Socket 939 comparison - it had demonstrated "reference" performance in our tests.

If before we could predict the results of AMD motherboard comparison as testbeds utilized the de facto standard DDR400 with 2-2-2-5 timings, today the values compared become a real mess. The DDR2 clock rate can actually be set in a wide range, but it doesn't affect the performance directly, as usually the clock rate increase leads to increased timings as well, and the advantage of higher throughput is still not used by today systems. Obviously, some memory modes will be better than other, but... for the that specific test only. So, we find ourselves on the threshold of long and interesting performance battles in which AMD64 platform hasn't participated for quite some time.

Speaking of specific test results, we must note, firstly, that both MSI motherboards performed almost identically (we couldn't have expected more considering the same BIOS firmware). In general, in archiving test DDR2-800 (4-3-4-5) systems perform along the older reference, but DDR2-667 (3-3-3-5) systems look a bit worse. Video encoding doesn't differentiate competitors, but still the DDR400 system performed a bit worse. Games reduce univocacy even more: Unreal Tournament 2004 prefers DDR2 (only DDR2-667 with 3-3-3-5 timings), and transition to DDR2-800 along with timings change puts such configurations on the last place. Doom3 instead willingly raises fps alongside memory clock rate increase, so here DDR400 corresponds only to its "rating" of 400. Most likely, this can be explained by different dependence of games on throughput and memory latency. Again, there's much more to test.

Bottom line

On the whole, the difference in test results between the new platform and the old one is expectedly low (under equal conditions). So there is absolutely no point in upgrading your system with a Socket 939 processor (which still satisfies your requirements) just for the sake of DDR2 memory.

But what should you prefer in a new computer? Perhaps, while the new components are filling the market, their prices will be higher than those for similar models for Socket 939. So the budget choice is crystal clear — just sum up the costs (both platforms are equipollent in other respects.)

It's quite another matter, if you want to buy a low-noise computer, especially in a tight case, and do not want to sacrifice performance. In this case, new Energy Efficient models should certainly be included into your choice list.

Write a comment below. No registration needed!