by Ilya Gavrichenkov
10/18/2004 | 09:00 PM
Today AMD is announcing two new processors for their highly promising Socket 939 platform: Athlon 64 4000+ and Athlon 64 FX-55 models. These processors will probably remain the fastest products AMD ever makes with its 130nm tech process. At the same time, without loud speeches, the company is also starting mass shipments of Socket 939 processors of the Athlon 64 family made with 90nm technology. The models on the new 90nm cores are going to hit straight at the heart of the market, their ratings ranging from 3000+ to 3500+.
<%BANNER[article]%>This all means just one thing: AMD is poised for an all-out invasion into the desktop CPU marketplace. There are all the necessary prerequisites for that, by the way. First, the company has got ready a stable platform (Socket 939) which is going to live more than just a year. Second, all leading chipset makers have mass-produced chipsets for that platform, and more advanced Socket 939 mainboards with support of the PCI Express bus and based on chipsets from VIA, NVIDIA and ATI are going to emerge in the market one of these days. Third, the Socket 939 CPU series currently includes high-performance processors, announced by AMD today, as well as relatively inexpensive models on the new 90nm core. Thus, AMD’s current market standing looks very firm.
I’d like to emphasize the fact that AMD went through the year 2004 without a single hitch. Introducing a number of new CPU models, the company has pulled the rating of the topmost Athlon 64 processor from 3200+ up to 4000+. Intel, the archrival, on the contrary had to toil through various kinds of problems that started since the unsuccessful release of the 90nm processor core called Prescott. Due to the high power consumption and heat dissipation of the core, the frequency of the Pentium 4 processor has only grown from 3.2 to 3.6GHz in 2004. Well, Intel is intending to launch a Pentium 3.8GHz before the end of the year, but this processor is very likely to become the last Prescott-core product. At least, Intel has abandoned its plans to release processors with frequencies of 4GHz and higher in the next year, saying they will focus on other ways to boost the performance instead.
All these things conspire to make the Athlon 64 platform a most appealing choice today. Its appeal is not only in AMD’s success in bringing ever faster processors to market, but also in the rich capabilities of these processors. I hope no one needs to be reminded that the Athlon 64 processor has long been supporting 64-bit extensions and featuring Cool’n’Quiet and Enhanced Virus Protection technologies. Intel’s desktop CPUs have only recently acquired some analogs to these innovations. So, AMD should be in no worry as to its current market position.
But I’ve got rather off topic. The review you’re now reading will be dedicated to the new high-performance CPUs from AMD: Athlon 64 4000+ and Athlon 64 FX-55. AMD’s new processors manufactured with 90nm tech process and belonging to the middle sector of the market will be covered in our upcoming reviews.
A transition of top-end members of the Athlon 64 family to 90nm tech process was expected to coincide with the release of the Athlon 64 4000+, but this never happened. AMD has decided to implement the new tech process in middle-range products first, and the new Athlon 64 4000+ is still based on the old 130nm core. The 2.4GHz frequency of the Athlon 64 3800+ was considered a limit for AMD’s CPUs manufactured with 130nm+SOI technology, so the higher rating of the Athlon 64 4000+ was arrived at by other performance-improving ways. In other words, the clock rate of the Athlon 64 4000+ and of the 3800+ is absolutely the same – 2.4GHz, while the higher rating is the result of a larger L2 cache (1MB against the Athlon 64 3800+’s 512KB). I’d say AMD returned to where it had started: the L2 cache of the new processor has the same size as in the first Athlon 64 processors in the Socket 754 package.
As a result, AMD had to refuse using the cheaper NewCastle core with a smaller die size for the Athlon 64 4000+. AMD has been very proud of this core when the transition to the Socket 939 platform was made, but the 4000+ model, like the Athlon 64 FX-53, is based on the well-known ClawHammer core. The arrival of a Socket 939 Athlon 64 with 1 megabyte of L2 cache doesn’t necessarily mean that there will be junior member of the family equipped the same amount of cache memory. The use of such a large cache in the Athlon 64 4000+ must have been an emergency measure: AMD can’t yet achieve an acceptable yield of chips capable of working at 2.6GHz with neither of its tech processes – older 130nm or new 90nm. Thus, we’ve got a topmost Athlon 64 for the Socket 939 platform that has a twice larger L2 cache than any of its kin. I suppose, however, that such processors will be replaced by another Athlon 64 4000+, on the 90nm Winchester core, with a smaller L2 cache but higher clock rate. The question is how quickly AMD will be mastering the new tech process.
The formal characteristics of the Athlon 64 4000+ processor look as follows:
Here’s what diagnostics utilities are reporting about this processor:
Thus, the Athlon 64 4000+ is an exact copy of the Athlon 64 FX-53, which is out of production since today. In other words, the Athlon 64 4000+ is a remarked Athlon 64 FX-53 and differs from the previous model, the Athlon 64 3800+, with its larger L2 cache only.
Released in June, the Athlon 64 FX-53 processor was targeting a rather narrow consumer group – extreme gamers. It has only one point of difference compared to the Athlon 64 3800+, the topmost model of the Athlon 64 family up to now – a larger L2 cache. This was supposed to continue further – a larger L2 cache remaining a prerogative of Athlon 64 FX series processors. But the just-announced Athlon 64 4000+ with 1 megabyte of L2 cache has brought about some confusion. The fact is AMD currently doesn’t have a core a CPU with more than 1MB of cache could be made out of. So, the company had to invent something different for the extreme gaming community.
Well, they just couldn’t be too choosy: all the capabilities of modern cores of the K8 architecture have long been realized in the Athlon 64 family. There was only one way out – a higher clock rate. As I mentioned above, the 2.4GHz clock rate of the Athlon 64 4000+ and Athlon 64 FX-53 is the limit of AMD’s capabilities both with old 130nm and the fresh 90nm tech process. But don’t forget that the Athlon 64 FX series includes premium products: such processors are expensive and are manufactured in limited quantities. When making them, the manufacturer can employ some methods that wouldn’t be economically profitable if employed for mass production of ordinary Athlon 64 CPUs. That’s how they found the solution: they just used some elements of their own 90nm technology to enhance the frequency potential of 130nm cores.
Particularly, AMD employed the so-called strained silicon in the 130nm process used to make Athlon 64 FX-55 dies (look here for details on this technology). Roughly speaking, they use this strained silicon in some critical spots of the chip, which limit the frequency growth the most (the 90nm tech process implies using such silicon everywhere in the chip).
This technology is rather expensive, but it allowed AMD to build something which had seemed impossible before: a processor with the K8 architecture, on the 130nm core and clocked at 2.6GHz. Thus, the difference between the Athlon 64 FX-55 and the Athlon 64 4000 is in 200MHz of the core clock rate. The rest of the formal characteristics of the Athlon 64 FX-55 are almost identical to those of the Athlon 64 4000+:
Well, you may have noted another discrepancy in the spec above – a higher heat dissipation of the Athlon 64 FX-55. Yes, the increase of the clock rate above the previously intended limit without a transition to a more advanced manufacturing technology has pushed the power consumption and heat dissipation parameters out of the earlier set limits. The typical heat dissipation of the Athlon 64 FX-55 is 15 watts higher than that of the Athlon 64 4000+.
Considering the specifics of the Athlon 64 FX-55, our diagnostics utilities can report nothing sensational to us:
I should confess that the release of the 130nm Athlon 64 FX-55 processor is not as big a surprise as the arrival of the Athlon 64 4000+ on the old ClawHammer core. The transition of the Athlon 64 FX series to 90nm tech process had only been planned starting from the next model (supposedly, from the FX-57). So, I don’t find anything extraordinary in the characteristics of the Athlon FX-55.
I haven’t yet met Athlon 64 processors with a thermal design power of above 89W, although the higher heat dissipation of the Athlon 64 FX-55 doesn’t shock me at all. It was a year ago that AMD warned mainboard manufacturers that support of a new thermal design power and higher CPU supply currents would be necessary for compatibility with future products of ratings 4000+ and higher and FX-55 and higher. Back then this warning was understood to concern future CPUs on 90nm cores.
The road map has been corrected, however, and the Athlon 64 4000+ with the Athlon 64 FX-55 come out on old 130nm cores and the former even fits into the older requirements to power consumption and heat dissipation. The latter, however, does not, and AMD applies new power specifications to it. Fortunately, these specs were formulated quite a long time ago, and all hardware manufacturers should have got ready to them. Let’s see if they really are and what problems an owner of an Athlon 64 FX-55 may expect to see due to the processor’s unusual thermal design power.
Mainboards first. AMD worded the new power and cooling requirements of upcoming Socket 939 processors about a year ago, long before the platform itself appeared, so the overwhelming majority of existing Socket 939 mainboards can work all right with the Athlon 64 FX-55. It means that the CPU power circuitry of such mainboards can output the necessary 80amp current. That said, there still can be some pitfalls. Particularly, the MOSFETs in the power circuitry may become too hot when you are using such a power-hungry processor. For example, the temperature of the MOSFETs on our ASUS A8V Deluxe mainboard was as high as 70-80°C during our tests of the Athlon 64 FX-55. This is certainly no good for them, and we’d recommend you to consider additional cooling of the CPU power circuit.
Second, the higher heat dissipation requires an appropriate CPU cooler, too. Many coolers for Socket 754/939/940 were developed when there was no info about the new TDP of Athlon 64 processors. It means you should pay attention to the cooler’s compatibility with the particular processor. For example, the Thermaltake Silent Boost K8 cooler we have been using long and successfully in our labs failed with the Athlon 64 FX-55 processor. With this cooler on, the Athlon 64 4000+ was 64°C hot, which is acceptable, while the temperature of the Athlon FX-55 under the same conditions was up to 83°C and that’s hardly a safe operational mode.
That’s why AMD sent us a more advanced cooler on heat pipes (made by AVC), specially for testing the Athlon 64 FX-55:
The use of the Athlon 64 FX-55 has also implications concerning the power-supply unit, especially if you’re planning to overclock this processor. You can get an estimate of the required wattage, basing on the averaged data from the following table:
Component | Maximum power consumption | Used power lines |
Athlon 64 FX-55 | 105W | +12V |
Mainboard | 20-30W | +3.3V, 5V, 12V |
Memory | 30-40W for each 512MB | +3.3V |
Expansion cards | 5-10W each | +5V |
Graphics cards | 60-80W for high-end cards | +5V and/or +12V |
Cooler | 2-4W | +12V |
HDD | 5-20W | +5V, +12V |
Floppy drive | 5W | +5V, +12V |
Optical drive | 10-20W | +5V, +12V |
Thus, a simple computer system with an Athlon 64 FX-55 processor, a powerful graphics card, a single hard disk drive and one optical drive will require a 350W PSU at the very least. And this PSU must provide a current of at least 18amp on the +12v power rail.
Then, you must keep in mind the fact that the power consumed by the processor grows dramatically at overclocking. It is described by a simple formula:
Power=a*c*F*V2,
where a is a coefficient that is proportional to the average CPU load, c is a coefficient that depends on the CPU architecture, F is the frequency, and V is the processor core voltage. So, while the Athlon 64 FX-55 clocked at its regular 2.6GHz frequency and with the normal 1.5v voltage consumes up to 104W under the maximum load, it will consume up to 140W if you overclock it to 2.8GHz and increase its voltage to 1.65v. The overclocked system with an Athlon 64 FX-55 must use a high-quality 400W PSU at least in this case.
Well, this guarantees nothing, too. It’s always better to have a margin of safety for the PSU. I ground this recommendation on my own experience, by the way, as our 400W Zalman ZM400B-APS PSU, which has been working in our testbed for about a year, successfully handling the Prescott and the Pentium 4 Extreme Edition, crashed during our tests. It crashed during our overclocking tests, to be exact, so this confirms the formula above and its consequences.
Summarizing this section of the review, I want to say that Socket 939 processors with the new TDP (right now there’s only one such processor – the Athlon 64 FX-55) pose strict requirements to the hardware parts you use in your computer system.
The Athlon 64 FX-55 and 4000+ are top-end and also the last processors AMD is making with 0.13-micron tech process, so we can’t expect them to be overclocking marvels. Considering the fact that the Athlon 64 4000+ is a remarked Athlon 64 FX-53 – and we have already investigated the overclockability of the latter well enough, it is clear that this processor will be rather bad at overclocking and will hardly reach a frequency much above 2.6GHz. The Athlon 64 FX-55, on the contrary, is a “dark horse” with its specially adjusted 0.13-micron tech process with SOI and strained silicon, so I would refrain from making any suppositions about its overclockability, especially since its regular clock rate is 2.6GHz already.
I checked out these theories and assumptions in our standard testbed (the ASUS A8V Deluxe mainboard), using an ordinary air cooler from AVC described earlier. For a higher efficiency of overclocking, I increased the CPU voltage by 10%: from 1.5v to 1.65v.
The Athlon 64 4000+ comes first. This processor works at a frequency of 2.4GHz, and its maximum multiplier is fixed and equals 12x. To overclock this CPU you need to increase the frequency of the clock generator. Having experimented for a while, I found that the maximum frequency of the clock generator the processor was stable at was 220MHz. The frequency of the CPU itself was 10% above the norm in this case, or 2.64GHz.
So, the overclocking potential of the Athlon 64 4000+ is the same as with the Athlon 64 FX-53 we tested before. Using air cooling you can speed these processors up to about 2.6GHz. By the way, this is yet another confirmation to the fact that it’s practically impossible to produce 130nm CPUs clocked at frequencies of over 2.4GHz in mass quantities.
Next goes the more interesting sample, the Athlon 64 FX-55, which is manufactured using special tricks like spot application of strained silicon. Thanks to that, its overclockability should be higher than that of the Athlon 64 4000+. Moreover, the Athlon 64 FX-55 is intended for non-compromise enthusiasts and its multiplier is not locked, so you can overclock it by simply increasing the multiplier in the BIOS Setup.
I chose that overclocking method, too, and increased the multiplier to 14x, keeping the system stable. Trying to get some more megahertz by increasing the frequency of the clock generator, I found the system losing its stability, so the highest clock rate of the Athlon 64 FX-55 without any special cooling methods was 2.8GHz in my tests.
So, the Athlon 64 FX-55 has had a smaller relative frequency gain than the Athlon 64 4000+ (7.7% only), but it reached a 150MHz higher maximum frequency. This result suggests a better overclockability of the Athlon 64 FX-55, which is manufactured with a special 130nm technology with SOI and strained silicon, in comparison with the Athlon 4000+. Owners of the Athlon 64 FX-55 will be able to achieve higher frequencies, and this may become another argument in favor of the “premium” nature of this series.
Now, I want to add a few words on the thermal behavior of the new processors, both at their normal frequency and at overclocking. Using the same cooler from AVC, I got the following CPU temperature data (Cool’n’Quiet was disabled during the tests):
To heat the CPU up in this test I used a specialized S&M utility, which is evidently the best program for testing the thermal behavior of AMD’s processors and checking their stability, for today. This utility can also warm up the memory modules and test them for any errors.
We are going to use this program in the future, too. It is free for download from here.
As for the results proper, the temperatures of the two tested processors don’t practically differ, so a larger cache doesn’t greatly increase the heat dissipation, if at all. Overclocking has a much harder effect on the temperature. For example, the 2.6GHz Athlon 64 FX-55 is very hot, just like the overclocked Athlon 64 4000+. The temperature of the overclocked Athlon 64 FX-55 looks just frightening, and quite naturally so: the higher heat dissipation during overclocking is due to the frequency growth as well as to the increased core voltage.
Finishing this temperature-related discussion, I want to remind you that AMD is offering its Cool’n’Quiet technology to reduce the heat dissipation of its processors under small loads. This technology is realized through a CPU driver, which drops the CPU voltage and the multiplier coefficient when necessary. Cool’n’Quiet is supported in both Athlon 64 4000+ and Athlon 64 FX-55 processor models.
The major goal of this test session was to find out how fast the new Socket 939 processors from AMD turned out and to compare their performance with what their predecessors as well as their competitors from Intel showed.
As for the predecessors, the situation here is pretty clear, I should say: the newcomers will not be able to avoid the comparison with Athlon 64 3800+. And as for the competing products, the situation here turned out somewhat tricky. Since AMD started developing its Athlon 64 CPU family pretty actively, and Intel is in hurry to introduce its new Pentium 4 CPUs, the market turned out somewhat misbalanced. The today’s fastest AMD processors selling for about $700 do not have any direct competitors in the market now. The top Pentium 4 model working at 3.6GHz is now selling for a much smaller sum of money: only $417. Therefore, the today’s comparison between the top Athlon 64 solutions and the existing top Pentium 4 processors should be taken quite relatively: as I have already said these solutions belong to different weight categories and sell at too different retail prices to be compared directly.
As for Athlon 64 FX, the situation here is just the opposite: the competitor solution, Intel Pentium 4 Extreme Edition, costs more than the AMD CPU targeted for extreme gaming enthusiasts. This way, when we look at the benchmark results, we should keep in mind that these processors also belong to different price groups.
The testbeds used in our today’s session were configured as follows:
We ran all the tests in MS Windows XP SP2 operation system with the installed DirectX 9.0c pack. The testbeds were set for maximum performance. Note that, we increased the Cycle Time (Tras) timing for Athlon 64 up to 10, because our experience suggests that the memory controller of Athlon 64 processor works more efficiently in this case, than in case this setting equals 5.
The average performance in “general” applications of different types has been revealed with the help of SYSMark2004 test package. This test set is a kind of industry standard, since it has been developed by a committee made up of a few different companies’ reps, including all the parties interested, such as Intel or AMD.
According to the results of SYSMark2004 we see that Athlon 64 platform is on the whole slightly slower than the Intel Pentium 4 based platform. It can be explained by the fact that Pentium 4 supports Hyper-Threading, which is a very efficient thing for a multi-threaded applications modeled by this particular benchmark. However, it would be incorrect to say that Athlon 64 architecture lost the SYSMark2004 test, because Athlon 64 FX-55 with the working frequency raised to 2.6GHz is still faster than the top Pentium 4 and Pentium 4 Extreme Edition models.
Everything I have just said is also proven by the diagram given below showing the average SYSMark2004 performance.
PCMark04 test also showing the average performance of the systems during various tasks solving, again puts Pentium 4 and Pentium 4 Extreme Edition CPUs to the forefront. Even Athlon 64 FX-55 refuses to oppose them.
However, the memory subsystem benchmark from the same test package awards Athlon 64 and Athlon 64 FX from AMD with the winner’s medals. This is not at all surprising, because with the shift to Socket 939 platform these processors acquired a dual-channel integrated memory controller with unprecedentedly low latency.
3DMark 2001 SE is a pretty old but still pretty popular benchmark. Here we see Athlon 64 and Athlon 64 FX processors to be at the head of the race.
In 3DMark03 we do not see such a huge gap as in 3DMark 2001 SE. However, the CPU test from this package still shows that processors with AMD64 architecture are faster than the rivals. Although the overall results of this benchmark do not indicate an indisputable advantage of Athlon 64 processors any more.
The processor test from the latest 3DMark05 shows that the top Pentium 4 processor based on Prescott core defeats Athlon 64 4000+. However, it still fails to outperform Athlon 64 FX-55. As for the total score, the Intel platform appears much faster than AMD platform. It is most likely to be connected with the fact that LGA775 systems use graphics cards with the new PCI Express x16 bus, which boasts considerably higher bandwidth than the good old AGP 8x bus used in the today’s Socket 939 systems.
All the gaming tests we are using unanimously call Athlon 64 and Athlon 64 FX processors the leaders. However, this is actually pretty logical, as Athlon 64 has long been known as a very powerful gaming solution.
During audio files encoding into mp3 format the top Athlon 64 and Pentium 4 processor models as well as Athlon 64 FX and Pentium 4 XE retain the parity with one another.
During video encoding with different codecs, CPUs with NetBurst architecture look much more attractive. In fact, we should have already got used to this.
The efficient memory controller is a very important factor affecting the archiving utilities performance. Therefore, Athlon 64 and Athlon 64 FX demonstrated an impressive victory.
During antivirus check by one of the most widely spread antivirus applications, again the laurels rest with Athlon 64 CPUs.
The higher clock frequency of AMD Athlon 64 FX-55, which has now reached 2.6GHz, allowed this processor based on AMD64 architecture to compete with Pentium 4 in Adobe Photoshop. However, the regular Athlon 64 processors are still falling behind the top Pentium 4 in this popular graphics editor.
Visual C++ is one more application where AMD CPUs show their best.
A similar picture can be observed in a popular Mathematica 5 application. However, as always this is quite an expected situation: calculations have always been a prerogative of AMD processors starting since K7 architecture.
Well, today we can admit that Athlon 64 processors show pretty high results also during 3D rendering. The 8th version of the Lightwave package is better optimized for Athlon 64 processors than the previous version. As a result, the AMD processors perform pretty good here. As for the performance in 3ds max, Athlon 64 processors are still yielding to Pentium 4 during single frames rendering, while during animation movies rendering they perform much faster.
Despite the fact that the new processors from Intel have been slightly delayed AMD still continues the race and keeps announcing the new CPU models. However, it would be still incorrect to say that AMD has finally become the leader in the CPU market. Even Athlon 64 FX-55 doesn’t provide the maximum performance in most applications. Therefore the users can find quite a bit of tasks where either Intel or AMD CPUs are preferable. Thus in games our preferences stay with Athlon 64, in media encoding tasks – with Pentium 4, in computational packages Athlon 64 is the leader, and in general tests Pentium 4 is ahead more frequently. In fact, we could continue this list more and more.
However, Athlon 64 4000+, just like its predecessor, Athlon 64 3800+, is still slightly ahead of the top model in the competitor product family: Pentium 4 560 working at 3.6GHz. However, the price of Athlon 64 4000+ is $729, and Athlon 64 3800+ costs $643, while Pentium 4 560 is available at a much more attractive price: $417. So, it is actually pretty hard to state which of the two processor families would be considered the best buy.
However, in the enthusiastic market the situation is more than clear. Pentium 4 Extreme Edition 3.4 falls steadily behind not only Athlon 64 FX-55, but also Pentium 4 560. Note that its price is $999, which undoubtedly makes this CPU the “worst choice” you can make today. I have not the slightest doubt that enthusiasts will definitely decide on Athlon 64 FX-55, which is now priced at $827.
If we veer away from the price-to-performance ratio for a while and look at the new AMD CPUs from the technological point of view, we will have every right to say that the “swan song” of the 0.13micron manufacturing technology from AMD appeared quite a success. And from this viewpoint Athlon 64 FX-55 is definitely a very attractive and interesting solution, since Athlon 64 4000+ is actually not that new being just an analogue to the Athlon 64 FX-53. Moreover, its rating seems to us a bit too high, because its advantage over Athlon 64 3800+ is only 1-3%. As for Athlon 64 FX-55, it represents a true skill of AMD engineers, who managed to squeeze another 200MHz out of the 0.13micron processor core due to point integration of the stretched silicon. As a result, Athlon 64 FX-55 turned out a unique CPU from the technological point of view. To make it AMD uses very “tricky” approach that is why its frequency potential is higher than that of any other Athlon 64 processor. However, there is nothing free in this world, so besides the high working frequency, the new Athlon 64 FX-55 also “boasts” higher heat dissipation and power consumption. So, get ready for those…