It is over a year already that we are watching a really exciting battle between the Intel Pentium 4 and AMD Athlon families for the right to be called the fastest desktop solutions. Both, AMD and Intel, which have been pushing their architectures into the market, for a long time couldn't offer any indisputably best solution to beat the competitor from all viewpoints. Since the launching of the first Pentium 4 processors, there has been no leader in the competition between it and Athlon: the core clock frequencies grew higher and the leadership migrated from one to another all the time. Neither great computing capacity of the Athlon processor, nor NetBurst architecture of the Pentium 4, which allowed it to work at incredibly high core clock frequencies, could grant indisputable advantage to one of the competing CPU families.

This situation has slightly changed in the middle of autumn, when the renewed Athlon XP family came out. These CPUs based on the new 0.18micron Palomino core boasted some minor architectural improvements, such as the implementation of SSE instructions support, TBL enlargement and introduction of Data Prefetch Logic, however, they started dashing ahead of the eldest Pentium 4 processors. Moreover, the core frequency of the Athlon XP processors grew much faster during the last couple of months than that of the Pentium 4 CPU family, which froze at 2GHz. As a result, in the end of 2001 Athlon 1900+ won all the laurels of the fastest desktop CPU.

However, Intel was preparing a worthy response to AMD's launching of the Athlon XP family. This response was a new 0.13micron Northwood core, which from now on will be used in the latest Pentium 4 processors. AMD in its turn also didn't keep its hands in pockets and reacted to the announcement of the new Pentium 4 (Northwood) working at 2.2GHz with the launching of the next Athlon XP 2000+ processor working at 1.67GHz. So, today we will dwell on these two latest solutions from the Pentium 4 and Athlon XP families.

To begin with, let's take a closer look at our today's heroes.

Intel Pentium 4 (Northwood)

New Northwood core to be used in the upcoming Pentium 4 processors doesn't differ too much from the predecessor from the architectural point of view. That is why you can consult our article called "Intel Pentium 4 1.4GHz Review. Part 1: Processor Architecture and Platform Overview", if you are looking for more details on the architecture peculiarities of the Willamette core. But what forced Intel to develop a new core? The reasons are quite standard, actually. 2GHz frequency appeared a kind of top for the 0.18micron manufacturing technology, which is used for Pentium 4 (Willamette) production. Therefore, Intel had to resort to more progressive manufacturing methods in order to be able to continue this CPU family.

Here we should note that the seeming low heat dissipation of Pentium 4 processors on Willamette core is a pretty illusory thing. Although the coolers used with Pentium 4 processors, even with those working at 2GHz remain not that hot doesn't mean anything, in fact. The matter is that Pentium 4 core heats not very evenly, so that some parts of it dissipate much more heat the other parts. Therefore, it appeared completely impossible to continue increasing the clock frequency of Willamette, even though the general temperature wasn't that high.

As usual, a more advanced manufacturing technology helps to solve this problem. In the middle of last year already Intel had at its disposal some production lines for 0.13micron processors. In particular, Intel used this technology to produce Pentium III processors based on Tualatin core. At first, they were going to introduce finer manufacturing process for Pentium 4 CPUs in autumn. However, the company had some problems with purchasing the necessary equipment, so that the first 0.13micon Pentium 4 CPUs came out only now.

Well, let's try to figure out what the differences between the new 0.13micron Pentium 4 (Northwood) and 0.18micron Pentium 4 (Willamette) are. First of all, we would like to point out that the cores manufactured with the new finer technology are smaller in size. So, Intel got the chance to add some more transistors implementing larger L2 cache. This way, the new Northwood acquired a twice as large 512KB L2 cache (unlike 256KB by Willamette). But even despite it, the size of Northwood die remained smaller than Willamette: only 145sq.mm by Northwood against 214sq.mm by Willamette.

Besides, the use 0.13micron technology allowed reducing the Vcore of Pentium 4 (Northwood) from 1.75V to 1.5V, which certainly influenced the heat dissipation. If the maximum heat dissipation of the older Pentium 4 processors working at 2GHz made 69W, then the new Pentium 4 Northwood working at the same clock frequency dissipate only 41W of heat.

The first processors on the new Northwood core will work at 2Ghz and 2.2GHz core clock. In order to distinguish between Pentium 4 (Northwood) 2GHz and Pentium 4 (Willamette) 2GHz, Intel decided to call its newly born baby Pentium 4 2AGHz. Note that later on Intel is planning to use the new processor core not only for its fastest CPU models, but will move it little by little into the processors working at the frequencies below 2GHz. In the beginning of April there will arrive two more Pentium 4 (Northwood) processors: 2.26Ghz and 2.4GHz. They will be using a new 133MHz FSB, i.e. Intel will start implementing a new Quad Pumped Bus working at 533MH instead of the today's 400MHz bus.

In order to prove that the new Northwood core doesn't have any other architectural improvements besides the increased L2 cache telling on the processor performance, we decided to run some synthetic benchmarks. We measured the performance of Pentium 4 2GHz on Willamette core and Pentium4 2GHz and 2.2GHz on Northwood core. For our tests we selected some benchmarks from SiSoft Sandra 2001 set. The algorithms used in this package to test different parts of the CPU are so simple that the results depend neither on the memory subsystem, nor on the processor cache size.

As you can notice, the performance rates of the CPUs built on different cores and working at the same core clock coincide. It means that Northwood doesn't have any improved calculating units compared with Willamette.

As for the larger L2 cache, its advantages can be traced with the help of a benchmark measuring the memory subsystem performance, namely with cachemem test:

Both graphs, for the read and write speed, show clearly that the L2 cache of Northwood processor increased up to 512KB. Besides that, no other changes touched the memory subsystem performance. If you look at the first graph, for instance, you will be able to see that the part of the L1 cache intended for data remained unchanged and equals 8KB. Although we used to say that it is exactly the small L1 cache that limits the performance Pentium 4 CPU in many cases. For example, their competitors, Athlon XP processors, have 64KB L1 cache for data, which is 8 times bigger.

However, it appeared that increased L2 cache is not the only difference between Northwood and Willamette core. Let's take a closer look at the CPU features returned by the processor itself. For this purpose we will resort to wcpuid utility:

Northwood


Willamette

As you can see, the only difference between the flags of the Willamette and Northwood processors is the support of On-chip APIC hardware implemented in the new core. It implies that Pentium 4 based on Northwood core supports SMP configurations. Pentium 4 CPUs do not officially support dual-processor configurations, however, it looks as if Intel decided to save some money and trouble another time. The matter is that Pentium 4 (Northwood) and Xeon (Prestonia) turn one and the same processor designed in different form-factors. Pentium 4 (Northwood) is designed for Socket478 while Xeon (Prestonia) - for Socket603. Intel believes that this will prevent the new Pentium 4 CPUs from being used in dual-processor systems, since all dual-processor mainboards go equipped with Socket603. However, if it goes on like that, some mainboard manufacturer, which doesn't work too closely with Intel, will very soon release some dual-Socket478 mainboards.

All CPUs based on Northwood core will be intended for Socket478 mainboards. The launching of the new core has automatically sentenced Socket423 to death: there will be no new 0.13micron CPUs coming for it. Also take note that all the already selling Socket478 mainboards will support the new Pentium 4 processors. The only thing old mainboard owners may need is an immediate BIOS update.

AMD Athlon XP 2000+

Although the story about new Intel processors appeared pretty long, we won't be able to say so much about the new AMD Athlon XP. New Athlon XP with 2000+ rating is built on the same architecture as its predecessors, i.e. on the 0.18micron Palomino core, which we have already discussed in great detail. That is why if you want to learn more about 0.18micron Palomino and about the performance rating introduced by AMD, please, see our AMD Athlon XP Processor Family Review.

Here we will point out just the basic features of Athlon XP 2000+ differing it from the predecessors. In the first place, it is the increased core clock frequency. Athlon XP 2000+ works at 1.67GHz and with the 133MHz FSB uses 12.5x multiplier, which is the maximum possible multiplier today. This processor dissipates the 70W of heat (at the most).

Athlon XP 2000+ is the last CPU of the Athlon XP family based on 0.18micron Palomino core. The next processor of this family due in the end of Q1 2002 will acquire a new Thoroughbred core produced with 0.13micron technology. However, as is known so far, Thoroughbred will not differ architecturally from the predecessor that is why we don't expect the processor performance to grow for the reasons other than the frequency growth.

The interesting thing is that the new Athlon XP 2000+ looks just the same as the previous slower models. Even though some Athlon XP 1900+ processors were built on the green processor board, the sample of Athlon XP 2000+, which we had at hand, was painted in usual brown color.

Testbed and Methods

As we have already told you before, our today's test are devoted to the performance of the latest processor models of the year 2002 from the Pentium 4 and Athlon XP families. At the same time we paid attention to the advantages of the new Northwood core of Pentium 4 CPUs and tried to compare it with the good old Willamette, which has been used in Pentium 4 processors all this time. That is why we tested two new Intel Pentium 4 processors on Northwood core working at 2Ghz and 2.2GHz, Pentium 4 2GHz based on the Willamette core and two AMD Athlon XP CPUs: 1900+ and 2000+ working at 1.6GHz and 1.67GHz correspondingly.

To test our processors we assembled the most popular and high-performance platforms. Pentium 4 processors were tested with an i850 based mainboard and RDRAM, while for Athlon XP CPUs we took VIA KT266A based mainboard with DDR SDRAM. The table below shows the details of the testbed configurations:

All tests were run in MS Windows XP Professional.

We carried out somewhat more tests this time, as we had many new processors to look at. Therefore, we decided to split the tests in several groups for your convenience. So, let's get started.

Performance

Office and Content Creation Applications

Athlon XP processor family has always been very successful in this type of tasks. The thing is that the typical business applications included into this benchmark work with relatively small data packs that is why the L1 and L2 caches size matters a lot here. 128KB L1 cache of Athlon XP CPU as well as exclusive 256KB L2 allows Athlon XP to easily outperform Pentium4 (Willamette) with 8KB L1 cache and 256KB L2 cache. However, the new Northwood core with larger 512KB L2 cache allows Pentium 4 based on it to start a really cut-throat competition with Athlon XP CPU. As a result, the performance of the 2GHz Northwood turns almost equal to that of Athlon XP 1900+. As for Athlon XP 2000+, it manages to leave Pentium 4 2.2GHz behind, though it appears only 1% faster than the rival.

Here we used a new version of Content Creation Winstone. Unlike the previous Content Creation Winstone 2001, it contains newer applications. Content Creation Winstone 2002 includes Adobe Photoshop 6.0.1, Adobe Premiere 6.0, Macromedia Director 8.5, Macromedia Dreamweaver UltraDev 4, Netscape Navigator 6/6.01, Sonic Foundry Sound Forge 5.0c (build 184). Besides, this benchmark set includes also Microsoft Windows Media Encoder 7.01.00.3055, which was absent by the previous benchmark version.

We have already described the major problems about the Windows Media Encoder. Because of some developers' mistake, this application doesn't use SSE instructions supported by Athlon XP processors. Therefore, Athlon XP doesn't prove as good here as in the previous benchmark, for instance. Moreover, we would like to point out that NetBurst architecture is more suitable for processing large packs of streaming data. Bearing in mind all these things, the excellent results shown by Pentium 4 appear pretty natural.

In SYSmark 2001 the eldest Athlon XP processor beat its competitor from the Intel Pentium 4 family based on Willamette core. However, the CPUs based on the new Northwood core and featuring larger 512KB L2 cache outperform Athlon XP 2000+. For instance, Pentium 4 2.0AGHz turns 3% faster than Athlon XP 2000+.

Here Pentium 4 processors showed their real best. The reasons for this victory are again the same as in Content Creation Winstone 2002. Internet Content Creation part of SYSmark 2001 includes Windows Media Encoder 7.0, which doesn't use SSE instructions implemented in Athlon XP processors because of some bug. That is why AMD babies lag behind Pentium 4 processors in this test.

As for the office part of SYSmark 2001, the situation is totally different here. Elder Athlon XP CPUs get faster than even one Northwood based Pentium 4 processor (the one working at 2GHz). Athlon XP 2000= performs almost as good as Pentium 4 2.2GHz.

To make the office coverage complete, we measured how fast the systems could archive the data packs of big size (directories with installed Unreal Tournament) in the "heaviest" mode for the system CPU that is with the maximum compression. So, the shortest time on the diagram stands for the best performance.

Note that the processor cache size has a tremendous influence on the performance in WinZIP. Nevertheless, even a large cache didn't let Pentium 4 Northwood outperform Athlon XP 2000+.

During the encoding of DVD graphics data stream into DivX MPEG-4 format, the situation repeats: the doubling of Pentium 4 L2 cache grants a nearly 5% performance increase. This fact allows both Northwood CPUs beat the eldest Athlon XP models.

Pentium 4 Processor Application Launcher v.2.1

When Intel announced its first Pentium 4 processor over a year ago, they also released a special benchmark set called Pentium 4 Processor Application Launcher v.2.1, which could brilliantly reveal all the advantages of NetBurst architecture. Today we would like to check how larger L2 cache can tell on the performance of the Pentium 4 processor in this benchmark. This test measures the CPU performance in a number of real applications, which either process large serial data streams or use SSE2 SIMD instructions.

Canon eJay MP3 Plus 1.3 is an application intended for work with audio files. It measures how fast a wav-file can be encoded into mp3 format with the help of Jay MP3 Plus Encoder. This coder supports SSE and 3DNow! SIMD instructions.

As you can see from the diagram, Pentium 4 processors cope with this task just beautifully. Moreover, the larger L2 cache by Pentium 4 has a really great positive effect on the processor performance. New Pentium 4 processors on Northwood core leave Athlon XP competitors very far behind.

This benchmark counts the fps in a demo of Incoming Forces 3D game developed by Rage Software. The game uses the latest generation visualization technology including high-detail landscapes and complex animated objects built on NURBS and 3D-fractals. All calculations in this game involve SSE and SSE2 instructions sets.

The results shown by the entire Pentium 4 family are more than impressive: elder Athlon XP processors without SSE2 support lag far behind. Moreover, Pentium 4 2.2GHz even failed this test, as there occurred "division by 0" error when getting the fps rate.

Here we measured the speed of MPEG-2 encoding with a codec from Ligos. As you can notice, the support of SSE2 instructions used by this codec, Pentium 4 processors again manage to become the leaders. It is also worth mentioning that the effect made by the increased L2 cache is no longer that significant here, so that the performance of Pentium 4 2.0GHz is nearly the same as that of Pentium 4 2.2GHz.

MagniTrax is a special 2D and 3D image viewer, which the user can manage by simply moving his head, which is followed by a USB camera for instance. For testing purposes the application follows the eyes of some hypothetical user, which determine the positioning of the 2D image on the screen.

As we have already pointed out several times, the calculations are performed best of all by Athlon XP processors. The coming of the new Northwood core didn't change this state of things and Pentium 4 family keeps falling behind Athlon XP in tasks of the kind.

Dragon NaturallySpeaking Preferred 4.0 is one of the best speech recognition applications. This benchmark measures how fast different processors can transform the wav-files into the regular text format.

Although Athlon XP manages to surpass the old Pentium 4 2GHz on Willamette core in this test, Pentium 4 Northwood with twice as large L2 cache still wins. The eldest Pentium 4 2.2GHz, for instance, outperforms Athlon XP 2000+ by 12%.

The tests ran in the popular graphics editor package called Adobe Premiere 5.1 showed the complete superiority of Pentium 4 architecture. As for the performance difference between Willamette and Northwood cores working at the same core clock, it made 5% in favour of Northwood.

The next benchmark involved one more video editor aka Ulead VideoStudio 4.0. It measures the time needed to create one MPEG-2 movie from four initial ones by using different effects. Like the previous benchmarks this one uses MPEG-2 codec from Ligos optimized for SSE2.

As far as the obtained results are concerned, Pentium 4 processors appear indisputable leaders, especially since Athlon XP processors do not support SSE2 instructions. The increased L2 cache of the new Pentium 4 CPUs doesn't contribute to the overall performance boost here.

Well, we see once again the already familiar picture. Pentium 4 CPUs appear much fast in Windows Media Encoder 7.0 than their AMD rivals.

3D Games

This gaming test is one of the most favourable benchmarks for Intel Pentium 4 processors, which manage to show really great result in it. And although Athlon XP 2000+ has nearly caught up with Pentium 4 2.0GHz, Intel's timely announcement of the new Northwood based CPUs increased the performance gap between Pentium 4 and Athlon XP quite significantly. As a result, Athlon XP 2000+ falls behind Pentium 4 2.0AGHz by some 12%, while Pentium 4 2.2Ghz is 18% ahead.

In another game, Unreal Tournament, the situation is totally different. The eldest processors from both families run neck and neck.

The results in DroneZ remind us of what we saw in Quake3, though here the whole thing looks somewhat hypertrophied. Due to though optimization of the game for SSE2 and high bandwidth of the memory bus, Pentium 4 processor family leaves all the competitors far behind. Note that the L2 cache size is very important for DroneZ like for all other gaming applications. In this benchmark, for instance, the increase of the Pentium 4 L2 cache from 256KB to 512KB with the same core clock frequency caused an increase in the fps rate by 16%.

In 3DMark2001 the performance difference between the eldest Pentium 4 and Athlon XP models appears not that great. However, Pentium 4 2.2GHz still outpaces Athlon XP 2000+. At the same time, Athlon XP 1900+ performs almost as fast as Pentium 4 2.0AGHz.

When we disable hardware T&L the situation changes a bit in favour of the new Pentium 4 processors on Northwood core, which get slightly ahead of the Athlon XP competitors.

Scientific and Professional OpenGL Applications

It is quite natural that Athlon XP family shows the best of its potential in the applications dealing with math1ematical modeling. In tasks of the kind the FPU performance is of greatest importance, which is an indisputable trump of all Athlon CPUs. Three-pipeline FPU of this processor allows it to rush ahead of Pentium 4. To illustrate the point see the benchmark results: Athlon XP 2000+ proves faster than Pentium 4 2.2GHz by 11%.

Now let's see the detailed report:

As we can see, Athlon XP does not always work faster than Pentium 4 in scientific tasks. For example, in Primodia test based on solving Shredinger equation for all the 61 electrons of Promethium element, all Pentium 4 processors (those on Willamette and those on the new Northwood) outperform the entire Athlon XP family. This benchmark requires higher memory bus bandwidth rather than faster arithmetic co-processor, that is why Pentium 4 architecture turns out quite at home in this benchmark as well as in all other applications operating large data packs intensively.

To check the performance of our testing participants during 3ds max 4 final rendering, we measured the time needed to render the Anisotropic Wheel scene in 800x600 resolution mode. Accordingly, the shortest time stands for the best result. We have already pointed out in our articles that rendering in 3ds max is a purely arithmetical task, and that is where the performance of the arithmetic co-processor matters a lot. The results prove once again that Athlon XP processors are beyond any competition, just like in ScienceMark.

However, in ViewPorts the situation changes drastically. Pentium 4 on the new Northwood core with a larger 512KB L2 cache starts beating elder Athlon XP processors.

And in conclusion we would like offer you a couple of tests from the professional SPECviewperf 6.1.2 package. The results of this benchmark are not influenced by the performance of the graphics card installed in our testbed.

In all the benchmarks Pentium 4 processors built on Northwood core appeared faster than their competitors from AMD. To tell the truth, this is very sad news for AMD, bearing in mind that Athlon XP used to be the leader in DRV-07 and DX-06 tests before Northwood announcement. However, larger L2 cache and higher clock frequencies of the new Pentium 4 CPUs did a great job having made Athlon XP not so attractive for professional OpenGL applications.

Conclusion

If we try to remember, what the first Pentium 4 was supposed to be like before the actual launching, then we will definitely arrive at the conclusion that Willamette turned out quite far from what Intel expected to get. Technological problems made Santa Clara engineers cut down some features of the NetBurst architecture. They deleted L3 cache and a bit of L1 cache intended for storing data. Moreover, they have also imposed some restrictions on the FPU/SSE2 unit. So, Pentium 4 appeared not so fast as Intel had initially planned to make it. Therefore, it is quite logical that competing solutions from AMD, Athlon XP processors, very often showed better performance than Pentium 4.

Now that the new Northwood core came, the situation began changing little by little. We have already complained about the too small cache of Pentium 4 CPUs that is why the increase in L2 cache memory from 256KB to 512KB resulted into a significant performance boost: from 5% to 15%. Besides, the migration to 0.13micron manufacturing technology will allow to keep raising the core clock frequency. So, the latest Pentium 4 processors built on the new core will be not only worthy competitors to Athlon XP, but will also prove much faster than the rivals.

Keeping in mind that AMD also doesn't waste its time, working on faster Athlon XP processors, the situation can be described as follows. In those apps, which use SSE2 instructions set or work with large packs of streaming data (for example, some games, MPEG-2 encoding, content creation applications and some professional tasks), elder Pentium 4 CPUs win the first prize. Athlon XP retains the leading positions only in those applications, which require powerful computing unit (for example, some games or tasks of math1ematical modeling). Moreover, in some office applications elder Pentium 4 and Athlon XP processors show equally good results.

Anyway, the picture we have just drawn for you can get changed any minute. Each CPU manufacturer always has a couple of solutions at hand, which can make their products more competitive. For instance, Intel is planning to provide its Pentium 4 family with faster Quad Pumped Bus working at 533MHz this spring. AMD, in its turn, will shift to a new Thoroughbred core for its Athlon XP family in the end of Q1 2002. This core will be made with 0.13micron technology and will allow increasing the working frequency of the CPUs rapidly.