A lot of water has flown under the bridge since the epoch of 486 processors, when both: Intel and AMD existed side by side and could be hardly called competitors, because they were using the same technologies. Now it looks as if AMD has for the first time got the chance to defeat Intel. Why so? The answer is simple as that: a new seventh generation CPU from AMD appeared in the market. It is Athlon.
But we would like to turn back to history before we start speaking about this newcomer. The first processor developed solely by AMD was K5 launched in 1996. Quite few people remember it now, and to tell the truth, there is not so much to remember, actually. As usual, AMD was too late with its new chip, which moreover fell significantly behind in clock frequency and performance that is why AMD failed to win the users' hearts and acknowledgement.
After this failure AMD purchased another independent x86 processors developer - NexGen, which you hardly remember now. This company possessed one of the most progressive technologies in those times and manufactured small amounts of processors without the arithmetic coprocessor. Having made use of these developments, AMD designed a new generation of its CPUs - K6. In terms of integer operations these processors surpassed all the existing analogues from Intel, however, the floating point unit still left much to be desired. And since the action games, which were gaining more and more popularity and public love were simply based on the FPU, AMD K6 suffered another great failure.
But AMD didn't want to give in. So, they offered an alternative solution: instead of the coprocessor they suggested using a specially designed set of SIMD-instructions - 3DNow!. This is how AMD K6-2 was born. Besides the usual K6 core there also appeared another single precision FPU. It could simultaneously carry out calculations of the same type with four pairs of operands and due to this fact AMD K6-2 performed quite well in applications optimized directly for 3DNow!. However, there appeared to be only one directly optimized application - Quake2 that's why most of the users remained disappointed.
Then AMD integrated into the core of its K6-2 processor L2 cache working at the CPU frequency. This helped to save high performance: the final product - K6-III had every opportunity to compete with its rivals, however Intel's price policy nearly crushed this initiative. Celeron was much cheaper though the performance it provided turned out rather satisfactory.
As a result we've got what we have now. Trying to beat each other in the price war, Intel and AMD keep on struggling. That is why Intel Celeron goes almost for the manufacturing price if not lower, and the High-End processor market is conquered by Intel's another product - Pentium III. The only chance remaining for exhausted and impoverished AMD is to strengthen its positions in the High-End processor market. But it should resort to the weapon other than the price, which is undoubtedly Intel's strong point, because it can drop the prices much lower than AMD. This weapon should be high performance. And AMD decided to start the battle with a new processor - Athlon.
If we tried to regard AMD Athlon's architecture superficially, we would describe its main parameters as follows:
- Chip manufactured with 0.25 micron technology;
- New generation core known as Argon with 22 million transistors;
- Works in special mainboards equipped with Slot A;
- Uses highly productive system bus Alpha EV6 licensed from DEC;
- 128KB L1 cache (64KB for instructions and 64KB for data);
- 512KB L2 cache working at half the core frequency. Placed outside the processor core but in the processor cartridge;
- Power voltage - 1.6V;
- 3DNow! SIMD-instructions set is enlarged with a number of additional commands, which make 45 altogether;
- The following versions are available: 500, 550, 600, 650MHz and 700MHz.
However, you should bear in mind that when AMD's greatest hope, the Dresdener Fab30, starts manufacturing 0.18 micron Athlons in the beginning of the year 2000, some of the parameters given above may get changed.
Well, Athlon is not so simple at it might seem at first sight. In fact, these several lines imply a great lot of different architectural innovations, which we will consider a little bit later. However, even those simple features of AMD Athlon are just impressive. For example, you could hardly miss that Athlon's maximum clock frequency is noticeably higher (Intel Pentium III has 600MHz and besides, it works at higher core voltage - up to 2.05V.) as well as its L1 cache, which is only 32KB by Intel Pentium III.
Before we go deep into details of the CPU we are reviewing, let's take a closer look at EV6 bus used by AMD, and namely at the main differences between this bus and Intel's GTL+. The similarity of the appearances is often deceptive. Although Slot A on AMD Athlon mainboards looks almost the same as Slot 1 turned by 180 degrees, all the bus protocols of Intel Pentium III and AMD Athlon are different and all the contacts are responsible for absolutely different things. Moreover, even the number of signals involved is different: Athlon makes use of about a half of 242 contacts while Pentium III resorts to a quarter. The exterior similarity was created by AMD on purpose, because they wanted to make it easier for mainboard manufacturers to avoid buying some special connectors to install into Slot A mainboards. And that's all.
In reality, despite the fact that EV6 works at 100MHz, the data is transferred at both edges of the signal, unlike GTL+, that is why the real data transfer rate makes 200MHz. If we keep in mind that EV6 is 76bits wide, 8 of which are used for ECC (Error Code Correct), then the data transfer rate appears 64bits x 200MHz = 1.6GB/sec. Just if you forgot: the bandwidth of GTL+ working at 100MHz is twice as low and makes only 800MB/sec. If the frequency of GTL+ gets higher up to 133MHz the bandwidth grows only to 1.06GB/sec. Frankly speaking, the bandwidth appears quite significant in both cases - with GTL+ and EV6. However, the modern PC100 memory only can deprive it of up to 800MB/sec, and AGP working at 2x regime - up to 528MB/sec. Not to mention the PCI and some other less "hungry" things. In other words GTL+ turns out unable to cope with the amount of the data being transferred. As for EV6, it hardly gives us any causes for concern and hence this bus is considered to be more promising and advantageous.
Besides, GTL+ is not the only bus, which allows increasing its working frequency from 100 to 133MHz. EV6 is also planned to finally reach 133(266)MHz and then even 200(400)MHz. However, these plans may remain nothing but plans forever. Since EV6 will requires more lines it can appear more complicated to make it work on the mainboards especially at higher frequencies. But on the other hand, if AMD succeeds, the system bus bandwidth may reach 2.1 and 3.2GB/sec correspondingly, which will allow using highly productive 266MHz DDR SDRAM in Athlon based systems.
Here is another interesting peculiarity of EV6. It supports multiprocessor systems. And now AMD really feels like entering this market next year. Unlike GTL+, EV6 provides point-to-point connection between the processors and the chipset, which helps to figure out the bus bandwidth for each CPU. Theoretically up to 14 processors can be connected to it. As for Intel systems the number of allowed processors is restricted because the overall bandwidth of GTL+ is equally distributed among all the CPUs. That is why EV6 bus is a very promising solution for multiprocessor systems.
Before passing over directly to AMD Athlon at work we consider it worthy to touch upon the problem of L1 and L2 caches.
As for L1 cache in AMD Athlon, its 128KB L1 cache is 4 times bigger than L1 cache by Intel Pentium III, which contributes not only to Athlon's good performance but also to its efficient work at higher frequencies. In particular, the small L1 cache is one of the main problems of Intel Katmai architecture, which seems unable to achieve higher processor performance by means of simple clock frequency increase. The small L1 cache starts bogging down at the working frequencies close to 1GHz. AMD Athlon is deprived of this drawback.
As for L2 cache, AMD is again up to the mark. First, the core integrated tag for L2 cache supports its size from 512KB to 16MB. As is known, the external Tag-RAM of Pentium III supports only 512MB L2 cache. Besides, Athlon allows using different dividers for L2 cache speed: 1:1, 1:2, 2:3 and 1:3. This really rich variety of dividers allows AMD to remain independent from the suppliers of particular SRAM, especially when it comes to faster models manufacturing.
Due to the possibility to vary the size and the working speed of L2 cache, AMD plans to introduce four Athlon families intended for different market sectors.
Architecture. General Issues
Well, we have finally come to the story about Athlon's work. Like Intel processors with the core inherited from Pentium Pro, Athlon has internal RISC-architecture. It means that all CISC-operations processed by the CPU are at first presented as a number of simple RISC-operations and only then the CPU calculating units start processing them. On the one hand, it looks as if AMD engineers were striving for additional hardships, but... this solution appeared really worth it. The processor can simultaneously carry out a few RISC-operations because they are comparatively simple, which helps to easily predict all the branches and hence to increase the performance due to higher parallelism. In other words, the manufacturer, who manages to make the most "parallel" processor, can make its device perform cooler with less effort spent. And AMD seems to have been guided by this particular principle when they developed their Athlon.
However, before operating parallel instructions streams, the processor needs to receive them from somewhere else. AMD, as well as Intel, used a special instructions decoder, which transforms the code received by the CPU. The decoder of AMD Athlon can decompose into RISC-parts up to three incoming CISC-commands at the same time. Intel's latest CPUs can also process up to three instructions, however, if Athlon doesn't care what instructions it decomposes, Pentium III requires two simple and only one complex instruction. As a result, Athlon appears omnivorous and can digest up to three instructions per time step in any case, while some parts of the Pentium III decoder may stay uninvolved because of the non-optimized code.
Before going into the appropriate calculating unit, the incoming RISC-instructions stream is kept for a while in a small buffer called Instruction Control Unit, which is intended for 72 instructions by AMD Athlon and only for 20 ones by Intel Pentium III. Having made this buffer somewhat larger AMD tried to avoid idle standing of the instructions decoder in case the Instruction Control Unit gets overloaded.
Another thing, which is worth mentioning, is a branch prediction table, which is four times as large - 2048 cells - as that of Pentium III and which stores the previous results of the logic operations. With the help of this data the CPU forecasts the possible outcome in case the operations are repeated. This algorithm allows AMD Athlon to make correct branch predictions in 95% cases, which is far not bad, if we compare it with the similar feature of Intel Pentium III - only 90%.
And now let's take a look at Athlon when it comes to calculations.
AMD processors have always been just perfect in operating integers. Since AMD K6 Intel processors were always falling behind with their integer operations. Nevertheless, AMD decided to give up all the previous developments here.
Due to three pipelined Integer Execution Units AMD Athlon can carry out three integer instructions simultaneously. As for Pentium III, it's maximum is two instructions at the same time.
We would like to say a few words particularly about the pipelines. Today the optimal pipeline depth for the progressive processors is 9 stages. If this number gets bigger the instructions are operated faster, because the pipeline rate is determined by the work of its slowest stage. However, in case the pipeline is too big an ordinary branch prediction error can turn all the previously made work with those instructions, which have already entered the pipeline, absolutely vain. If it happens, the pipeline needs to be cleared and the whole thing starts anew.
That's why the integer pipeline depth by AMD Athlon is 10 stages, which is very close to the optimum. Unfortunately, we have nothing to encourage Intel's fans with, because the pipeline of Pentium III has 12-17 stages depending on the type of the instruction in process.
Taking into account the facts we mentioned above, the following results achieved in ZD CPUMark 99 (integer operations performance) look quite logic:
Wow! Cool! The first benchmark is over and the first great success of AMD Athlon is hardly deniable: it managed to surpass Intel Pentium III by well over 20%! Now we know the value of a large cache plus better organized pipelines. However, it is still too early to celebrate the victory, because the performance of most today's applications and especially games doesn't so greatly depend on the integer operations, as it does depend on the real numbers. Therefore, let's take a look at Athlon's new pipelined FPU.
Floating Point Operations
With a sinking heart we turn to the built into Athlon FPU block. As all of you probably remember, floating point operations used to be a real Achilles' heel for all the previous AMD processors. The main problem was the non-pipelined FPU block in K6, K6-2 and K6-III. As a result, the general performance was unbelievably low though AMD FPU managed to spend fewer time steps than Intel's processors on most floating point operations, because the FPU couldn't start working on the next operation until the previous one was completed. However, AMD didn't feel like changing anything in their FPU in those times and called to the developers for replacing it with 3DNow!.
But it looks as if AMD has learned its lesson from the previous experience. The arithmetical coprocessor in Athlon can boast a 15 stage deep pipeline in contrast to 25 stages by Pentium III. You should also keep in mind the thing we have already mentioned above that the longer pipeline doesn't necessarily mean higher performance. Besides, Intel Pentium III has a very serious drawback, which is, of course, absent by AMD Athlon: such operations as FMUL and FDIV are non-pipelined.
Athlon's FPU includes three blocks: the first is intended for simple operations such as addition, the second - for complex operations such as multiplication and the third one - for data operations. Due to this division of labour Athlon can carry out two floating point operations simultaneously. Well, Intel Pentium III cannot boast the same thing, actually! In its case the instructions are completed in sequence - one by one.
Only working at 650MHz Pentium III can manage to catch up with a new super FPU of AMD Athlon. This situation strikes as very unusual but seems to be extremely promising especially for 3D games. At last AMD processor will be able to give it hot there!
At first sight there seem to be no changes made to Athlon's MMX-operations engine compared to its predecessor AMD K6-III. However, it is not quite right. Although MMX instructions are used in a limited number of applications, AMD still added another few items to this set, which also appeared in the MMX block of Intel Pentium III processor. Among those instructions added were the operations for obtaining the average, minimum, maximum and some extraordinary data transfer operations.
As far as the architectural peculiarities are concerned, AMD Athlon has two MMX-blocks that's why both: Athlon and Pentium III, allow carrying out two MMX-instructions simultaneously. However, the MMX blocks in Athlon possess higher latency, than those in Pentium III, which should theoretically make this CPU perform slower in MMX applications. In order to find out the actual speed of MMX we resorted to the good old Intel Media Benchmark:
Strange as it might seem but despite this theoretical defeat, the practical performance of Athlon still keeps it up to the mark. It is most likely to be due to a larger cache or better instructions decoder that Athlon made such good progress. However, we would like to stress once again that only few applications make real use of MMX that is why even if Athlon proved worse than Pentium III in terms of MMX it would be hardly noticeable.
3DNow! block of AMD Athlon underwent serious changes. Though its architecture remained unchanged - two pipelines operate the instructions working with 64-bit registers, which contain pairs of single precision real numbers, the set acquired 24 new items. These new operations should allow not only increasing the data processing rate but also involving 3DNow! Technology for such tasks as sound and video recognition and Internet :-) Besides, there also appeared new instructions for processing data stored in the cache, which was similar to SSE. The enhanced 3DNow! support is already integrated into Windows 98 SE and DirectX 6.2.
So, now the 3DNow! set includes 45 instructions in contrast to 71 SSE instructions from Intel. And these new instructions are very likely to contribute to greater effect made by 3DNow!. In order to prove this statement AMD distributed an additional DLL for the well-known benchmark 3DMark 99 MAX, which makes use of all the CPU's newest features.
Especially to estimate the processors performance in 3D games, 3DMark 99 MAX offers CPU 3DMark index, which calculates all 3D scenes but does not display them on the screen. So, we get the result depending only on the CPU performance in terms of 3D graphics processing and on the system memory bandwidth. We examined the performance of AMD Athlon with the help of this benchmark for all three possibilities: with the old 3DNow!, with the new Enhanced 3DNow! and without this technology. For a better comparison we offer you the results shown by Intel Pentium III with the enabled SSE-optimization and without it.
As we can see, AMD Athlon proved very powerful according to this test. Even without the new 3DNow! instructions it retains its leading position far ahead of Intel processors. This fact lets us believe that Athlon will provide remarkable performance in 3D-games. And now we are going to pay attention to a more urgent question: the real efficiency of different sets of SIMD-instructions: SSE against 3DNow!. In this respect we will take a look at the performance gain in 3DMark 99 MAX with the corresponding optimization enabled:
Well, the results are very illustrative. If the performance gain provided by the older 3DNow! set composed of 21 instruction was almost similar to that provided by SSE, then the new 45-instruction SIMD set from AMD manages to leave its rival far behind. Frankly speaking, these results were achieved in a synthetic test and the situation in real application will undoubtedly depend on the developers. However, as for us, we can only insistently ask them to optimize their products for both: SSE and 3DNow!, especially keeping in mind that with the launching of AMD Athlon both sets turned out pretty close to each other in terms of the features and possibilities provided.
After we have considered the interior of AMD Athlon from various points of view and tested some of its parts with the synthetic benchmarks, it's high time we started trying it in real situations. And while all AMD fans imagine their quick victory over all other competitors, we suggest describing the test system.
The system based on AMD Athlon was configured as follows:
- AMD Athlon 600MHz CPU;
- MSI MS-6167 mainboard;
- Diamond Viper V770 Ultra graphics card;
- Creative Sound Blaster Live! sound card;
- IBM DJNA 372200 HDD;
- 128MB SEC PC-100 SDRAM.
The system based on Intel Pentium III was configured as follows:
- Intel Pentium III 550 and 600MHz CPUs. The data concerning Intel Pentium III 650MHz was obtained due to its overclocking up to 6,5?100MHz;
- ABIT BE6 mainboard;
- Diamond Viper V770 Ultra graphics card;
- Creative Sound Blaster Live! sound card;
- IBM DJNA 372200 HDD;
- 128MB SEC PC-100 SDRAM.
All the tests were run under MS Windows98 SE. The resolution in all 3D-tests was set to 800x600x16.
Abiding by our old tradition let's begin with the performance of AMD Athlon in office applications.
Athlon working at 600MHz performs almost the same as Pentium III 650MHz. No doubt, better integer operations and larger cache help Athlon to surpass Pentium III in business applications at the same working frequency. However, it is not so very important if MS Word or MS Excel work faster than usual. The thing that really needs higher speeds is the world of 3D-games. But before we pass over to it let's check the 3D performance with the help of the synthetic test 3DMark 99 MAX:
Again Athlon failed to disappoint us. Even without any special optimization for the new 3DNow!-instructions this processor maintains the leading position over Intel Pentium III even working at higher clock frequency. But enough for talks, we are passing over to real tests - 3D games. The first game we are going to look at will be the OpenGL game - Quake2, which has a specially optimized for 3DNow! miniport:
Although AMD represented by its K6-2 and K6-III has always been really cool in Quake2 and its 3DNow! technology proved very efficient, nobody could even dare expect Athlon to show a 25% performance gain! Note that it is not a synthetic test but a real application, which also greatly depends on the graphics card used. Besides, you should also bear in mind that our test system was equipped with a graphics card based on nVidia Riva TNT2 chipset. If it were 3dfx Voodoo3, then the performance gain could appear even higher! But you may feel suspicious about such impressive results that is why in order not to stain AMD's reputation and to avoid accusing it of being dishonest we also tried another gaming OpenGL application, which is not yet touched by omnipresent AMD - Quake3 Arena Test:
Well, again we see a 25 percent gain achieved by AMD Athlon in all regimes except High Quality, where the performance is limited by the graphics card and not by the CPU. Therefore high-speed pipelined FPU not only proves reliable but also allows Athlon to become the fastest gaming processor. All the evidence backing up our point is given in the chart below derived for Direct3D-game Expendable:
And here one more time Intel Pentium III 650 fails to catch up with our hero. In conclusion we offer you the results shown in one of the most complicated games for any CPU - in Unreal:
Frankly speaking, there is not so much left we can say about Athlon here. In all possible cases AMD Athlon appears faster than Intel Pentium III.
Well, the mere fact that Athlon always appears ahead of Pentium III speaks for itself. The brilliant arithmetical coprocessor, high quality integer operations and perfect SIMD-instructions support made Athlon a great thing for everyday work and entertainment.
In fact, a comprehensive review of a new processor implies covering such an important issue as overclocking. However, we are not so optimistic about it.
Although Athlon core doesn't have any locked multipliers or fixed frequencies, the nominal frequency is set by means of a few resistors located on the processor board. That is why if you want to change some CPU parameters, you have to take away the cartridge automatically losing your warranty and to expose your processor to great risk when resoldering the pieces.
However, it is not the only possible way of solving this problem. Athlon processor board is equipped with a technological connector, which again can be reached only after you remove the cartridge. Through this connector you can install an additional adapter, which allows changing the multipliers and the bus frequency. The only disappointing thing is the fact that no one knows where we can get adapters like that. Unfortunately, we haven't yet heard anything about any supplies of these devices.
As far as the practical results are concerned, Kryotech manages to overclock AMD Athlon 600 up to 800MHz in its cooling machines, which seems very promising.
In other words, Athlon is quite overclockable but in fact this thing is pretty hard to fulfil. That's why overclocking of this processor is very unlikely to become a highly beloved sport as that of Intel Pentium III or Celeron. Nevertheless, some irresponsible retailer, who hunt easy money and possess all the required skills can easily start remarking Athlon processors. Unfortunately, AMD doesn't offer any ways to protect their products from these harmful actions.
Having finished with the CPUs for Super 7, AMD began pushing through its own Slot A and EV6 system bus. However, it appeared deprived of all Intel's achievements in the field of chipsets and mainboards. Now AMD will have to create the entire required infrastructure, so that we could get not only the processor but also the mainboards for it equipped with Slot A.
And judging by the first results here, AMD succeeded. To begin with the company developed their own logic AMD 750 known as Irongate, and their own mainboard - Fester, which was used by a number of Taiwanese mainboard manufacturers.
As for AMD 750 chipset, it is quite an ordinary thing: its features are very similar to those of i440BX. However, nothing more is actually needed. As we saw, AMD Athlon works perfectly and even leaves all other competing products far behind.
AMD 750 has a traditional architecture and consists of the North Bridge AMD 751 and the South Bridge AMD 756. Via EV6 bus North Bridge provides the contact between the processor and the memory and PCI/AGP buses, and the chipset supports up to 768MB PC100 system memory in three modules at the most, AGP 2x and 6 PCI bus master devices. The duty of the South Bridge, which is responsible for the whole interface with the peripheral devices, is not limited to the traditional functions set. It also supports UltraDMA/66 IDE-devices.
Here we would like to touch upon another very exciting problem - compatibility. We all remember that the previously made mainboards intended for AMD processors required a special miniport with AGP GART driver. This was quite a significant cause for concern for most users because very often this driver refused to work properly with certain graphics cards. The driver is also a must for Athlon's proper work with AGP, however, we have every reason to say for sure that its realization has got significantly better. We failed to come across any modern graphics cards, which proved incorrectly working. And in general in terms of stability and compatibility AMD Athlon caused absolutely no problems in Windows 98 as well as in Windows NT. So, the reliability and stability of Athlon based systems is beyond doubt.
By the way, it turned out pretty interesting that Irongate processors can be also used in the mainboards for Alpha processors, because EV6 bus is utilized for Alpha processor as well. In particular, UP1000 mainboard (for Alpha) from API is based exactly on this chip.
Although Irongate looks very up to date now, its cool features will undoubtedly appear insufficient in a while. Very unwillingly AMD continues its development and research and is planning to provide its offspring with AGP 4x and PC133 or Direct RDRAM support. According to AMD, such Taiwanese companies as ALI, VIA and SiS should be involved into the development of Athlon chipsets, though they refused to participate in the promotion campaign of the new processor.
VIA has already announced its Athlon chipset - KX133, which differs from Irongate with PC133 memory and AGP 4x support. KX133 also has a traditional architecture - this is a combination of the North bridge VT8371 and any South bridge taken from VIA's chipsets, for instance VT82C686A. This chipset is very likely to become the most widely spread mass product. It is about to appear in the market in the fourth quarter already and will boast cooler features such as up to 2GB system memory support.
However, if we try to reveal a bit further prospects, then we should expect such things as SMP, faster system bus support, integrated graphics solutions. So, it looks as if we had no reasons to worry about the future of AMD Athlon chipsets.
AMD decided to help the mainboard developers and worked out a reference design called Fester. It was submitted to mainboard manufacturers, which were expected to provide the market with the sufficient amount of mainboards for Athlon by the time it would be launched. As for AMD, they introduced a small number of mainboards intended solely for demonstrating and promoting purposes and not for sale. However, in reality the things turned absolutely different from AMD's expectations.
The matter is that the suggested design had six layers, while almost all the other today's boards have only four layers. This difference caused some difficulties and the mainboards were slightly delayed. Moreover, even the first mainboards by MSI, which appeared in the mass market, had to be called back because of the low stability they guaranteed. Besides, in the situation of chipsets shortage Intel also tried to exert certain influence upon mainboard manufacturers. That's why for now only MSI, Gigabyte, FIC, Asus, Biostar and GVC have already announced their products. And among those available we can see even fewer companies represented.
According to some sources, ASUS and FIC are now carrying out joint research dealing with a four-layer design, however, it is too early yet to demand any real results.
We reviewed only two samples of mainboards designed for AMD Athlon - MSI MS-6167 and Gigabyte GA-7IX. As for those pieces from MSI and Gigabyte they were based on Fester design. MSI turned out very unsophisticated and simply copied the reference design. Gigabyte in its turn handled the matter with greater creativity and even made certain small changes. However, as a result we still have only two mainboards as like as two peas.
Both mainboards are made in ATX-format, based on AMD 750 chipset, support up to 1GHz processors and possess 5 PCI, 2 ISA and 1 AGP and 3 DIMM slots. Of course, they also support UltraDMA/66. System monitoring is carried out by means of W83782D controller. The only visible differences can be detected in BIOS Setup: MS-6167 can boast more memory settings than its competitor, while 7IX appears easier to understand for an unsophisticated user.
In terms of the mainboards performance, they prove almost identical and hence we won't dwell here. It seems to be the best thing if we could wait for the other manufacturers involved in Fester redesigning to launch their mainboards. These boards may turn out not only very speedy ones, but may as well have special options for Athlon overclocking.
Well, it's high time we drew all conclusions. Today our tests prove that AMD Athlon is the fastest x86 processor with progressive architecture. Intel is absolutely unable to withstand AMD's success, because Katmai architecture has almost exhausted its potential while AMD can increase the frequencies of its newest chips without facing any serious hardships. Even Intel's shift to 0.18 micron technology will hardly help solve this problem, since the factory of AMD in Dresden, which will manufacture Athlon processors with the same technology but with copper conductors is nearly open. Besides, Intel won't manage to gain in performance until they introduce some changes to the processor core. That's why Willamette, which is going to come out not so soon is the only hope of this microprocessor giant.
To cut the long story short, AMD has every chance to get the better of Intel and not just to improve its financial state but also to remove Intel from the leading position. It has just to satisfy the growing demand for Athlon and the appropriate mainboards and that's all. However, if the company has insufficient productive capacities the task may appear quite a problem.
To tell the truth, we are not so sure about the situation. It may turn out both: successful and hopeless. That's why we can't do anything but believe that everything will be all right.