by Ilya Gavrichenkov
05/22/2006 | 09:00 PM
The coming summer promises to be a truly hot season. From the meteorological prospective this forecast may turn out wrong, if some unexpected cyclones kick in. However the situation in the processor market has long been determined and goes as planned. Both leading processor manufacturers, AMD and Intel, selected summer time for the renewal of their high-performance line-ups. In the middle of this summer Intel is going to introduce processors with principally new Core micro-architecture, while AMD will be pushing the new AM2 platforms with DDR2 SDRAM support throughout the entire summer season.
Although everyone impatiently awaits the new Intel Core 2 Duo solutions that are also known as Conroe, AMD once again managed to get ahead of their competitor, just like all those previous years. Their new Socket AM2 solutions will start shipping in mass quantities on June 1 already. Therefore, today we are going to introduce to you the new AMD processors in all detail, and those who are dying to find out more about the Core 2 Duo solutions from Intel, will have to wait until the launch date.
Despite the upcoming announcement of pretty promising Intel CPUs, Socket AM2 platform from AMD attracted a lot of attention. AMD has been postponing the introduction of DDR2 SDRAM until the very end, because the K8 processor micro-architecture including the integrated memory controller benefits mostly from the low memory latency rather than its high bandwidth. And the currently existing DDR2 SDRAM cannot boast very low latencies just yet. Nevertheless, the speed of today’s DDR2 SDRAM has grown up enough to ensure that Athlon 64 processors will benefit from shifting to the new memory type. Although the very first tests of the engineering samples of the new AMD platform didn’t really reveal any significant performance improvement from using the DDR2 memory, now we are talking about the mass mainboards and CPUs, which is a totally different story. And this is the major intrigue of our today’s test session: most AMD fans want to believe that the new Socket AM2 processors will prove worthy rivals to the upcoming Intel Core 2 Duo.
Moreover, the new AMD CPUs will acquire a new core revision that can not only support new type of memory but has also undergone some minor modifications to make Athlon 64 processors even more attractive to the users. Of course, once the new Intel processors with Core micro-architecture hit the streets, AMD may lose some of its adherents. However, it is still too early to make any final conclusions especially since some of the improvements and modifications of the K8 core may turn out pretty useful in certain cases.
So, let’s meet the new AMD processors for Socket AM2 in person now and try to forecast what the future prepares for them.
AMD designed an updated core with K8 micro-architecture for their Socket AM2 platform. This core received the name of Revision F. So, all dual-core and single-core AMD processors with the integrated memory controller supporting DDR2 SDRAM will so far be based only on this core revision.
The major innovation introduced by AMD is certainly the DDR2 memory support. AMD simply replaced the integrated memory controller with a new one, since Athlon 64 micro-architecture allows making changes like that easily. At the same time, the new memory controller of the Athlon 64 processor family is not backward compatible with the regular DDR SDRAM. It means that from now on DDR memory can be regarded as an outdated solution. These days the two leading processor developers, AMD and Intel are quite unanimous in terms of the memory support: their platforms require DDR2 SDRAM. Of course, this should affect the DDR2 prices, so that we would expect the DDR2 SDRAM prices to drop below the DDR prices for the modules of the same capacity fairly soon.
Returning to the DDR2 SDRAM support by the core Revision F memory controller I have to stress that it officially supports DDR2 with up to 800MHz frequency. In other words, AMD managed to introduce DDR2-800 SDRAM support in its platforms ahead of Intel. Of course, the new AMD CPUs are also compatible with the slower DDR2 memory working at 667MHz and 533MHz frequencies. However, since K8 micro-architecture will benefit much more from the low memory latencies, you need DDR2-800 to ensure that you get the highest performance gain possible.
Note that the memory controller of the new core has traditionally acquired a slightly longer list of supported DDR2 frequency dividers than the official specification claims. As a result, some mainboards will even allow using Socket AM2 Athlon 64 processors with DDR2-1067 SDRAM without clock frequency generator overclocking. However AMD doesn’t officially claim support of DDR2 memory faster than DDR2-800 yet.
Besides the DDR2 SDRAM support, core Revision F can boast a few additional features. For example, Athlon 64 processors for Socket AM2 support Pacifica virtualization technology. This is AMD’s response to Intel’s VT that has been first introduced in Presler based CPUs.
Another important achievement resulting from the transition to AMD’s new Revision F core is the lowering of processors’ power consumption. Although AMD continues manufacturing the CPUs with the 90nm technology (using SOI and DSL technologies), Socket AM2 solutions boast lower heat dissipation and power consumption than their Socket 939 counterparts. By moving the Athlon 64 X2 processor family to the new core AMD has formally lowered the maximum heat dissipation by 19% (from 110W to 89W). As for the single-core Athlon 64 processors, the Revision F core allowed to reduce their maximum heat dissipation by 30%: from 89W to 62W.
Making CPUs more economical is actually as important as switching to new DDR2 memory. Especially, since the performance-per-watt concept is gaining more and more attention and is being actively promoted by the CPU developers as a major metrics for product quality evaluation.
However, lower heat dissipation of mass AMD processors is not the last thing yet. As a matter of fact, Socket AM2 platform and the use of Revision F core for the new AMD processors allowed introducing the whole new family of more Energy Efficient processors. AMD is going to offer two types of Energy Efficient CPUs: with 65W and 35W peak heat dissipation levels. Of course, the processors with 65W maximum heat dissipation will compete with Conroe from the thermal and electrical point of view, while the 35W solutions will target at small, quiet and economical systems. AMD is not going to use any specific manufacturing techniques to produce Energy Efficient CPUs. They will simply sort out dies from the Revision F batch that meet the low thermal requirements.
AMD processors will migrate to AM2 Socket massively. The company will launch dual-core Athlon 64 X2 as well as single-core Athlon 64 and budget Sempron processors at the same time. Therefore, there will be a few modifications of the Revision F cores in the market. All the possible modifications and their specifications are given in the table below:
And this is what the Athlon 64 X2 core Revision F looks like:
Note that even though this core acquired DDR2 SDRAM support, it doesn’t have any ultimate enhancements from the micro-architectural standpoint. Since the first Athlon 64 processors came out, AMD persistently avoids making any changes to the core decoders or execution units. In other words, we are still witnessing the extensive development of the K8 architecture by means of slight additional modifications. And it used to be more than enough to ensure that they compete successfully with Intel. But today the situation is different. The Intel Core 2 Duo processors coming out this summer feature principally new micro-architecture that allows processing up to 4 commands per clock cycle. And AMD CPUs will have really hard times competing with them, keeping in mind that they cannot boast the same theoretical peak performance. From this standpoint, we are a little bit disappointed with the core Revision F despite all the innovations. Frankly speaking, we wish they had done more on the micro-architectural level. However, looks like not this time.
Let’s take a closer look at the new Socket AM2 platform and its features other than DDR2 SDRAM support.
First of all, I would like to say that Socket AM2 is formally none other but a 940-pin processor socket. However, the new Socket AM2 processors are absolutely incompatible with the previous generation Socket 940 and Socket 939 connectors: neither from the logical nor from the electrical prospective. To prevent the user from installing the processors into the old incompatible mainboards, the engineers changed the pin-out of the new processors: the pin layout is completely different and the new CPUs will simply not fit into the old sockets.
The major advantage of the new Socket AM2 solution is that from now on AMD will have a unified platform for both: expensive dual-core and single-core budget processors. The same Socket AM2 mainboards will work with both: Athlon 64 X2 as well as Athlon 64 and Sempron CPUs.
However, the introduction of the new processor socket doesn’t bring the life cycle of the older socket types to an end. AMD promised to continue supporting and shipping Socket 939 solution until the user interest to this platform exists.
Socket AM2 also sets new requirements to the mainboards in terms of the CPU peak power consumption and heat dissipation. Although we have already said that new core Revision F processors can boast lower power consumption, the platform can now accommodate even more powerful processors than before. The peak current supported by the platform has now been raised from 80A we received on Socket 939 mainboards to 95A. As a result, this platform can work with processors consuming up to 125W of power, while the maximum power consumption of the Socket 939 CPUs was no higher than 110W.
Besides the powerful processor voltage regulator circuitry, new Socket AM2 mainboards have also acquired new cooler retention mechanism. Now the retention bracket pressed against the mainboard should be fastened with 4 screws instead of previously used 2. However, the locking slits on the bracket are exactly the same as in the previous case.
It means that Socket AM2 mainboards can allow installing old cooling solutions that used the default retention bracket. Those cooling systems that were fastened directly to the Socket 939 mainboard PCB will not fit onto the new platform without appropriate retention modifications.
The table below contains the complete list of Socket AM2 processors that should be available in retail on June 1, 2006:
Note that the relation between the frequency, L2 cache memory size and performance rating of the Socket AM2 processors is the same as that of the Socket 939 processors. On the one hand, it will help the users to better find their way among the new processor characteristics, but on the other it is an indirect indication that AMD doesn’t expect any dramatic performance increase from the transition to the new core revision.
I would like to draw your attention to the fact that only dual-core processors are claimed to support the today’s fastest DDR2-800 SDRAM. As for the single-core CPUs, the official spec claims that they should only work with DDR2-667 memory. This is quite logical, since dual-core processors are in much higher need for memory bandwidth at least because system RAM deals with cache coherency issues for both cores.
Socket AM2 processor family got significantly larger thanks to the introduction of Energy Efficient processors with 65W and 35W thermal envelopes. These CPUs do not support the same high frequencies as their “fully-fledged” counterparts and cost a little bit more. However they promise to be extremely attractive in a number of applications such as small quiet systems, for instance. In other words, we do not expect Energy Efficient processors to get very widely spread just yet.
Nevertheless, you should remember that you can fairly easily identify new processors with lower thermal envelope. The third latter in the processor marking of the regular CPUs is “A”. The 65W solutions will have an “O” in the third position, while the most economical ones – the 35W ones – will be marked with a “D”.
Unfortunately, the transition to Socket AM2 form-factor is unlikely to make dual-core AMD processors more popular. Even though the transition helps enlarge the product range in the dual-core segment, it will not lead to any price drop there. All Athlon 64 X2 CPUs will continue shipping for over $300 a piece, which will hardly help them win more market. Especially, since Intel has thrown quite a bit of inexpensive dual-core processors into the market getting ready for the new Core micro-architecture based solutions. For example, the youngest dual-core processor from Intel has already dropped below the $150 bar. So, from this prospective looks like Intel is the primary force for further dual-core processors promotion.
To test the performance of the new Socket AM2 platform AMD provided us with two new CPUs: Athlon 64 FX-62 and Athlon 64 X2 5000+. The first one is a dual-core processor targeted primarily for hardcore gamers willing to get the maximum performance at any rate (from the financial standpoint). The second CPU is a top dual-core solution in the Athlon 64 X2 processor family.
Athlon 64 FX-62 boasts the highest actual clock frequency out of all the previous and newest models. It works at 2.8GHz. Moreover, it has even caught up with the single-core Athlon 64 FX-57 processor in terms of clock speed! However, it had to pay its price for this high frequency rate: the heat dissipation of the newcomer is 125W, which can be regarded as a kind of a record here. There are no other processors in the AMD product range that are as hot as this baby…
The CPU-Z diagnostic utility reports the following info about Athlon 64 FX-62.
Note that the nominal Vcore of the Athlon 64 FX-62 processor is 1.35V-1.4V, which is higher than that of any other dual-core CPU from the Athlon 64 X2 family.
All this indicates very clearly that the frequency potential of the 90nm cores with K8 micro-architecture has been almost completely exhausted. However, our overclocking experiments suggest that you can reach even higher speeds if you disregard growing power consumption and the corresponding consequences.
So, if we raised the Vcore to 1.5V then our test sample worked just fine at 3,075MHz obtained as 15x205MHz (Athlon 64 FX processors feature unlocked clock frequency multiplier).
During this experiment the processor was cooled with an ordinary AVC air-cooler (Z7U7414002).
I have to admit that successful overclocking of the Athlon 64 FX-62 over 3.0GHz without any extreme cooling solutions is a pretty impressive result. All FX processors usually allowed no more than a 200MHz frequency increase with the regular air cooling used. So, AMD will easily be able to increase the nominal clock rates of its processors beyond 3.0GHz. The only thing that may become an obstacle in this case is extremely high power consumption and heat dissipation of the CPU. Our Athlon 64 FX-62 test sample overclocked to 3,075MHz and working under full load consumed 192W of power! And this is a way beyond the requirements that AMD has set for the new Socket AM2 platform.
The second processor we tested today, Athlon 64 X2 5000+, works at the nominal frequency of 2.6GHz but features smaller L2 cache memory than FX-62. Each of its cores has 512KB of L2 cache memory.
The CPU-Z utility detects this processor as follows:
Note that all dual-core processors from Athlon 64 X2 family, including the 5000+ model, have lower Vcore lying between 1.3V and 1.35V. As a result, these CPUs can fit into the 89W thermal envelope.
If we compare the results from our practical electrical measurements for Socket AM2 CPUs, we will get a very interesting picture. As always, we used a special S&M utility to measure the maximum power consumption (you can download this utility here ). We measured the current that goes through the CPU power circuitry. So, the numbers given below do not take into account the efficiency of the CPU voltage regulator laid out on the mainboard.
We have already got used to the fact that high heat dissipation is one of the features of processors based on NetBurst micro-architecture. So the numbers on the diagram can have a shocking effect on you. But we cannot deny the facts. The top-of-the-line AMD processor, Athlon 64 FX-62, boasts slightly higher heat dissipation and power consumption than the top dual-core CPU from Intel – Pentium Extreme Edition 965 based on Presler core revision C1. The top models of the mainstream dual-core processor families from AMD and Intel – Athlon 64 X2 5000+ and Pentium D 960 - are now showing about the same heat dissipation, too. So, AMD’s top processor models can no longer retain the title of the most economical CPUs. The latest Intel processors based on the newest Presler core revision are not any worse from this prospective. So, it is not for nothing that the Socket AM2 platform eases the electric current and heat dissipation requirements.
But let’s get back to Athlon 64 X2 5000+, and namely to its overclocking potential. Since the maximum value for the clock frequency multiplier is locked, we can only overclock this processor by raising the clock generator frequency. But this is not an issue and we can still achieve high results. Having increased its Vcore to 1.5V, we got our CPU running stably at 2.99GHz.
The obtained overclocking results for both new Socket AM2 processors used with the regular air-cooler show that the frequency potential of the core Revision F has become slightly higher than the previous AMD processors had. So, the new Socket AM2 platform may be of some interest to overclocking fans, no doubt about that.
Since the chipsets are connected to the K8 micro-architecture through HyperTransport bus and the memory controller is integrated into the CPU, the transition of Athlon 64 processor family to the new socket type and new DDR2 SDRAM doesn’t really require any special chipsets. All the chipsets that have been previously used in Socket 939 mainboards can also be used for the new Socket AM2 mainboards.
However, it didn’t really matter for Nvidia – the today’s primary chipset supplier for AMD platforms. They launched a new family of chipsets designed specifically for the new AMD platform. The new Nvidia nForce chipset family (including nForce 590, nForce 570 and nForce 550) are positioned as “specially designed to support the new AMD platform”. However, there is nothing new in these chipsets from the CPU support standpoint. They have simply acquired new wider range of features. And the simultaneous launch of the new AMD processors and Nvidia chipsets is none other but a good marketing move.
Although you will still need to upgrade your mainboard is you wish to switch to a new Socket AM2 platform. Therefore, new chipsets from Nvidia appear just in the right place at the right time, because those users who decide to upgrade will most certainly want to get a better and newer set of features for their buck. This is exactly the target user group for the new solutions based on Nvidia nForce5xx series core logic.
The new Nvidia nForce chipset family includes 4 solutions targeted for different price groups.
All these chipsets are based on the same element base created around nForce 570. This is the chipset that should be regarded as a reference point for all the other members of the family: nForce 590 and 550.
Nvidia nForce 570 SLI chipset is a single-chip solution that can be called a continuation of the nForce4 SLI family.
This core logic set supports SLI mode but works only as PCI Express x8 + PCI Express x8. Nvidia nForce 570 Ultra is a similar solution with that only difference that it doesn’t have SLI support.
The most experienced gaming fans will get at their disposal a new Nvidia nForce 590 SLI that can support PSI Express x16 + PCI Express x16 SLI mode. In this case there is a special microchip that allows implementing the support of the second PCI Express x16 graphics card slot. This chip is connected to the processor and MCP via the 16bit wide bidirectional HyperTransport bus working at 1GHz frequency.
As for the budget solution from Nvidia, the nForce 550 chipset, it is none other but the same nForce 570 Ultra with slightly cut-down features.
The formal specifications of the new nForce chipset family are given in the table below:
If we take a closer look at the specifications of the new Nvidia chipsets for Socket AM2 platform, we will see that they are not very much different from the previous generation nForce4 chipsets. In fact, there are three major improvements:
Even though this is a relatively short list of improvements, Nvidia presents the new chipset family as a tremendous step forward. They not only create huge marketing stir around them but also accompany the introduction with the additional software features and tools.
Without going too deep into details this time, we would like to mention the major technologies the new Nvidia chipsets deliver to the AM2 platform:
Moreover, Nvidia is going to accompany the nForce 590/570/550 based mainboards with the new nTune 5.0 utility that has now acquired more advanced system monitoring and fine tuning options. One of the first mainboards based on Nvidia nForce 590 SLI chipset appeared ASUS M2N32-SLI Deluxe, which was used in our today’s test session.
We used the following equipment to assemble test platforms for our AMD Socket AM2 processor test session:
The tests were performed with the mainboard BIOS setup for maximum performance.
Before we pass over to the performance tests of the new AMD Socket AM2 platform, we decided to take a closer look at the DDR2 SDRAM and the advantages it should bring to the Athlon 64 platform in terms of performance. It is no secret for anyone here that AMD platforms are very sensitive to the memory subsystem latency. And even though the transition from DDR to DDR2 SDRAM promises a significant increase of the memory bandwidth, it doesn’t really reduce the latencies.
To obtain some practical results that could allows us make some conclusions about the performance improvement from the transition to DDR2 SDRAM, we assembled two identical systems with DDR and DDR2 memory and compared the performance with different memory timings and memory bus frequencies. We used the following CPUs for these tests: Athlon 64 FX-60 for Socket 939 and Athlon 64 FX-62 for Socket AM2 with the core clock dropped down to 2.6GHz. Note that we installed 512MB memory modules for these tests, so the total amount of RAM in the test systems equaled 1GB.
First of all let’s take a look at the synthetic benchmarks results showing memory bandwidth and latency:
The practical results prove our theoretical suppositions. DDR2 SDRAM boasts higher bandwidth than the regular DDR memory, and as the memory frequency increases, so does the bandwidth. However, the situation with latencies is totally different. Only DDR2-800 SDRAM with rather aggressive timings (for this frequency) 4-4-4 can compete with DDR400 SDRAM working with minimal timing settings of 2-2-2. DDR2-667 with the minimum possible timings of 3-3-3 can only reach the same practical latency as DDR400 with 2.5-3-3 timings and cannot compete with fast DDR SDRAM. As for DDR2-533 SDRAM, this memory is definitely worse than any DDR400 SDRAM in terms of latency.
The results of SiSoftware Sandra 2007 correspond to the results we obtained in Sciencemark 2.0. In fact, it is evident already that the performance of your Socket AM2 system will improve only if you have DDR2-800 SDRAM or fast DDR2-667 SDRAM with 3-3-3 timings settings. The performance gain in all other cases is pretty questionable and will primarily depend on the type of applications running.
Now that we have tested the memory subsystem parameters let’s move over to the performance in complex benchmarks.
SuperPi test aggravates what we have just mentioned. It is true, only if the Socket Am2 system is equipped with DDR2-800 SDRAM it can outperform a Socket 939 system with DDR400 SDRAM set to 2-2-2 timings.
Some applications appear pretty independent of the memory subsystem performance. Nevertheless, even here we can see that DDR2 SDRAM is quite inefficient compared with fast DDR400 SDRAM.
The performance of WinRAR archiving utility depends a lot on the memory subsystem performance. In this case we see that the application is very sensitive to growing memory bandwidth. However, even in this case DDR2-800 with aggressive 4-4-4 timings can show slightly better results than a Socket 939 platform with 2-2-2 latencies.
The same is true for gaming performance. Even the slowest DDR400 SDRAM proves more efficient than some types of DDR2 memory.
So, now we can answer the question we asked in the beginning of this review chapter: the transition to DDR2 SDRAM didn’t bring any significant performance improvement. Of course, the use of new memory standard can be quite handy for the future. The development of DDR SDRAM is over, both – JEDEC and memory makers – are now focusing on faster memory standards based on DDR2. Therefore, AMD has definitely made the right choice. The company waited until DDR2-800 SDRAM became widely available and performed the transition without losing any of the platform speed. By the way, a significant advantage of the DDR2 memory over the regular DDR SDRAM is the broad availability of high-capacity modules, which should be especially valuable when Windows Vista comes out.
First of all we decided to test the performance of our processors in popular synthetic benchmarks.
Here I have to say that we didn’t reveal anything new. As we have already shown above, DDR2 SDRAM provides a small performance gain. Therefore, the new Athlon 64 FX-62 owes its high results to high clock speed of 2.8GHz in the first place. As for Athlon 64 X2 5000+, it is sometimes even slower than Athlon 64 FX-60, because this processor features half the L2 cache of the latter despite the identical clock speed. However, if the benchmark is not sensitive to the cache memory size, Athlon 64 X2 5000+ can outperform any Socket 939 processors thanks to its DDR2-800 SDRAM.
We tested the general performance in office and digital content creation applications with the help of SYSMark 2004 SE benchmarking suite, which also uses multi-threading very actively.
During digital content processing AMD processors show much better results than Intel competitors. As for the new Socket AM2 platform, it doesn’t offer us any surprised this time.
The size of L2 cache memory is very important for office applications. Therefore, Athlon 64 X2 4800+ for Socket AM2 systems leaves the newcomer with 5000+ performance rating behind. I would also like to stress the high results of the Intel Pentium D 960 processor. As you can see from the chart, it only yields to AMD FX processors that cost much more.
When we tested audio and video encoding speed with DivX, iTunes and Windows Media codecs, the new Socket AM2 platform showed very tangible advantage. Steaming video encoding is very sensitive to memory bandwidth increase. So, Socket AM2 processors cope with these tasks about 2-4% faster than their Socket 939 counterparts.
Apple Quicktime doesn’t take the new platform with as much enthusiasm. In this application Socket AM2 Athlon 64 4800+ is even slower than its Socket 939 fellow. However, the performance different is not dramatic even when we work with streaming data.
Intel Pentium Extreme Edition processor used to be an undefeated leader in Adobe Photoshop and Adobe Premiere. However the new fast Socket AM2 Athlon 64 FX-62 processor changed the situation completely. Now the newcomer from AMD wins all the laurels in the applications dealing with image and video editing.
Unfortunately, the 2.8GHz frequency of Athlon 64 FX-62 processor is not enough for successful competition with Intel Pentium Extreme Edition 965 during final rendering in 3ds max 7. The thing is that rendering can be split into a few parallel processes easily so that all four virtual cores of the Intel processor get loaded to the full extent. However, the situation is different during final rendering in Maya: the high-end dual-core AMD processors are ahead here.
As for the benefits of DDR2 SDRAM, we can state that there are none, or there is even negative influence. Anyway, final rendering is not the reason for you to switch to the new AMD platform.
Theoretically, DDR2 should provide a pretty tangible performance improvement in games. The fastest DDR2-800 SDRAM may ensure up to 6-7% performance growth. However, we cannot yet state any quality advantage of the new Socket AM2 platform from AMD. At the same time the preliminary results of the promising Conroe processor suggest that it will show a significant performance improvement of the Intel platforms in games when it comes out. In other words, even though AMD processors retain their leading position in gaming applications in the meanwhile, the situation may change dramatically very soon. And AMD fans should be ready to take the blow.
Since it is extremely interesting to compare the performance of the new Socket AM2 platform against the desktop CPUs supporting DDR SDRAM, we decided to add a few more widely spread programs to our list of tests.
The 7-zip archiving tool that supports multi-threading quite efficiently we measured the data compression and decompression speeds.
Optical text recognition speed was measured with the popular ABBYY Finereader 8.0 suite.
We have also tested our platforms in a popular computer algebra tool aka Mathematica. The new version of this application can now take advantage of the potential offered by multi-core processors.
Summing up everything we have said about the new AMD Socket AM2 platform we have to admit that the introduction of DDR2 SDRAM support is a small evolutionary step forward. Our tests showed that the transition to DDR2 SDRAM doesn’t bring in any significant performance gain. Moreover, you have to make sure that your system uses the fastest DDR2 SDRAM with 800MHz frequency and minimal timings if you want to see any performance improvement at all. The widely spread DDR2-667 SDRAM may sometimes provide absolutely no performance gain at all compared with the Socket 939 platforms supporting DDR400 SDRAM with low timings settings.
In conclusions I would like to say that the arrival of the Socket AM2 platform supporting DDR2 SDRAM is still not an ordinary occasion. Even though Socket AM2 systems do not boast any superior advantages over the Socket 939 platform at this time, the positive effect from this transition will be more evident in the future. Of course, DDR2 SDRAM is a much more promising memory standard. Its working frequency and bandwidth increase dynamically, its price goes down faster and it allows building high-capacity memory DIMM modules. As a result, AMD will certainly win from betting on DDR2. And the timing is simply perfect: no one will accuse AMD of unjustified moves neither from the performance nor from the pricing standpoint.
However, there is no real pressure from Intel, AMD’s main competitor, at this time. AMD CPUs retain their leading positions practically in all applications. The increasing clock speed of the top models has definitely contributed to that: Athlon 64 X2 is now running at 2.6GHz and Athlon 64 FX-62 - at 2.8GHz. Of course, there is some concern that this state of things may change when new Intel Core processors come out, but it is still too early to talk about it.
I have to confess that I was left with a little bit of disappointment deep inside now that I have taken a close look at the new core Revision F processors from AMD. The thing is that company engineers have once again done just a few minor modifications and didn’t make any in-depth architectural changes. This approach to processor improvement will sooner or later lead AMD Athlon 64 processor family to a frustrating defeat in the “armament drive”. Unfortunately, we do not have any information about significant changes of the K8 micro-architecture at this time.