Not so long ago AMD Athlon 1.2GHz and Intel Pentium 4 1.5GHz were the eldest desktop CPUs. Things changed very quickly, when on March 22 AMD announced another 1.33GHz model and on April 23 Intel talked back with its Pentium 4 clocked at 1.7GHz. Now this couple will be at the top of these both desktop families. It should be pointed out that Pentium III clan is no longer treated as high-performance processors. Having stuck at the bar of 1GHz, Pentium III became unable to compete with elder Athlon and Pentium 4 CPUs. The problem is unlikely to be settled by switching Pentium III core over to the 0.13micron technology, as soon as Intel decided not to provide this CPU family with a faster bus and larger L2 cache. This way, presently there are only two rivals left. As we saw, it was impossible to make out, which of the two CPUs was the real leader - Intel Pentium 4 1.5GHz or AMD Athlon 1.2GHz. AMD's child dominated in office applications, and Pentium 4 took its revenge in 3D games. For more details concerning the tests of these processors you may consult our review of AMD Athlon 1.2GHz with 266MHz FSB.

So, it will be especially interesting to compare the eldest models of these processor families today. Moreover, both Athlon with 266MHz FSB and DDR SDRAM and Pentium 4 are gradually getting down from the Olympus to our everyday life becoming more and more affordable for general public. Another issue that arouses our interest towards the newcomers is that they are the latest representatives of their families based on Thunderbird core from AMD and Willamette from Intel.

Neither Palomino Nor Northwood Yet

As is known, a grave predicament to an unlimited clock speed increase is the CPU's heat dissipation that grows as the clocking gets higher. In the long run it may lead to the CPU physical damage. Some High-End gamer OEMs take advantage of this phenomenon and create exorbitantly overclocked systems in cases equipped with real freezers. But this way is unsuitable for CPU manufacturers. They have to make processors fit for mass systems, i.e. in systems with common air cooling. Consequently, the developers are forced to improve their CPU cores, shifting to finer manufacturing technologies and constantly caring about lower heat dissipation.

As a perfect illustration to this idea we would like to remind you of Intel's calling back Pentium III 1.13GHz processors. Attempting to catch up with AMD CPUs, Intel tried too hard to overclock its Coppermine core used in Pentium III and contrived to launch a CPU working at 1.13GHz. But systems based on the newly-announced solution turned out unstable because of extreme overheating. So, it became clear that 1GHz is Coppermine's ceiling, which cannot be exceeded without major architecture redesigning. A new core for Pentium III, Tualatin, which will be made with 0.13micron technology, can theoretically work at frequencies approaching 2GHz. However, by the moment when Intel is able to start its mass production (no sooner than in Q3), the company's main product will be Pentium 4. This way, for obvious marketing reasons, there will be only two versions of Tualatin - 1.13 and 1.2GHz.

Similar to Coppermine, top clock speed for the current Pentium 4 core, Willamette, is 2GHz, and for Athlon Thunderbird it equals to 1.4GHz. So, both CPU manufacturers, AMD and Intel, will soon have to switch over to new processor cores.

Now let us have a look at the companies' roadmaps for the near future:

As you can see at the above given chart, Intel is planning to introduce the new Northwood core starting from-2GHz and up. This core will be produced with 0.13micron technology, allowing Intel to hit 6GHz without any further remakes. Furthermore, Northwood will undergo some improvements aimed at increasing the performance. Desktop CPUs based on this core are expected to have a two times bigger L2 cache than Willamette has: it will be equal to 512KB. At the same time, intending to enlarge the L2 cache in Northwood, Intel is more concerned not about performance improvement, but about the possibility to compensate the performance drop, which will be inevitable in the Pentium 4 systems with PC133 SDRAM and Brookdale based mainboards. In RDRAM systems a larger L2 cache is also unlikely to cause a notable performance increase. The bandwidth of dual-channel Rambus, which is used on mainboards based on i850 core logic, is high enough that's why the users of these systems intended to solve typical tasks can easily do without 512KB L2 cache. It would be much more fruitful to increase the size of Pentium 4 L1 cache. Yet there is no information on Intel's intentions in this direction.

Before starting to manufacture Northwood 2GHz due in mid Q4, Intel will announce only one more CPU built on the old 0.18micron Willamette core. This processor clocked at 2GHz is scheduled for August launch.

Things stand not so clearly with the new core for AMD Athlon CPUs, which is supposed to work at the frequencies over 1.5GHz. These CPUs were initially promised to appear in the end of 2000, but then the launching date was postponed and is postponed over and over again. According to the latest roadmap update given above, the first Palomino 1.5GHz will come out only in July. However, this AMD's decision was not ungrounded. The previous Thunderbird core looks smart at the frequencies below 1.4GHz, so till recently AMD Athlon happily endured the competition with Pentium 4 that works at a considerably higher core frequency. For this reason it makes no sense for AMD to launch Palomino right away to enable a painless clocking growth till 2GHz for Athlon. That's what we'll prove later in the review.

It is noteworthy that Palomino is a temporary product, too. This core will simply let increase the maximum clock speed of Athlon CPUs without a necessary shift to finer manufacturing technology. AMD will come to it by means of an elementary redesign, which should reduce the processor heat dissipation. That's why it is hard to say now whether Palomino will feature any architectural changes meant to improve the performance or not.

AMD isn't going to position Palomino as a competitor to Northwood. It will make no use anyway. Palomino is manufactured with the old 0.18micron technology, just like Thunderbird, that's why 2GHz appears the top for it. Northwood's real competitor will be Palomino's successor, Thoroughbred, that will be made with the real 0.13micron technology. As a result, AMD will be able to increase the CPU working frequency up to over-2GHz.

Well, this is nothing more than future projects so far. The heroes of our today's review are Athlon 1.33GHz and Pentium 4 1.7GHz. These processors are based on the old cores, Thunderbird and Willamette. So, let's get back to the present.

Pentium 4 and Athlon CPUs Today

Before we draw your attention to the tests and the results obtained, we suggest looking at the key features of processors. We also included some details about the chipsets supporting these CPUs in the table below. We guess it may come in handy for proper understanding of the differences between Athlon and Pentium 4 platforms.

Please, don't treat this chart as anything more than just reference stuff. Pentium 4 and Athlon differ so greatly in architecture that it appears completely wrong to compare them with each other. If you wish to get acquainted with the architectural peculiarities of both these processors, you are welcome to take a look at one of our previous reviews:

• Intel Pentium 4 1.4GHz
• AMD Athlon 600MHz
• AMD Athlon (Thunderbird) 800MHz

The eldest Athlon 1.33GHz uses a 133MHz FSB, therefore it requires the multiplier set at 10x. Alongside with the CPU modification supporting 133MHz FSB, AMD is also offering Athlon 1.3GHz supporting 100MHz FSB. For this CPU you need to set the multiplier at 13x. It's worth mentioning that originally AMD didn't expect a multiplier like that for its processors, so on most mainboards in order to ensure that the CPU settings are correct, you will have to set the clock multiplier equal to 12.5x, AMD has assigned a new value to it. Another reason is that many BIOSes and utilities will define this CPU's FSB frequency incorrectly, showing 1300MHz as 12.5x104MHz. However, you shouldn't worry about this fuss.

As for Pentium 4 1.7GHz, it sticks to the multiplier of 17x. Besides, the new Pentium 4 version demands higher Vcore: Pentium 4 1.7GHz features 1.75V Vcore, while it predecessors worked at 1.7V.

Since both discussed CPUs belong to the latest models, based on Thunderbird and Willamette cores respectively, special attention should be drawn to their extremely high heat dissipation. Thus, Athlon dissipates up to 73W of heat, while Pentium 4 1.7GHz - up to 59W. As a result, proper cooling appears a really urgent matter, especially for Athlon. Here it is not only a good cooler that matters, but also sufficient ventilation inside the case.

Before we deliver the results of the tests, we find it necessary to comment on the different clock speeds of the compared CPUs. Indeed, watching the long-lasting rivalry of Athlon and Pentium III, as well as Duron and Celeron, we have got accustomed to see competitors with equal clock frequencies. But the advent of Pentium 4 broke the tradition. Intel developed this CPU anew from A to Z, so its brand new scalar architecture is very much different from the regular one. New Intel processor has an ultra-long 20-stage pipeline, hence, on the one hand, it can perform well at rocket-high clock speed. But on the other hand, the average number of instructions it can fulfill per clock is smaller than that by Athlon and even by Pentium III. This way, Athlon will be always slower than Pentium 4 in terms of clock frequency, and Pentium 4 in its turn will always fall behind Athlon in case both CPUs are running at the same core clock frequencies. For this reason AMD positions its Athlon 1.33GHz as a competitor to the elder Pentium 4. Therefore it is correct from the technological as well as from marketing point of view, to compare the elder models of both families, now clocked at 1.33GHz (Athlon) and 1.7GHz (Pentium 4).

Testbed and Methods

The main aim of our investigation was to reveal the fastest PC platform, so first of all we looked at the performance of Intel Pentium 4 1.7GHz and AMD Athlon 1.33GHz in the fastest systems. In the meanwhile there is no alternative for Pentium 4 platform: all you can get for a system with this processor is an i850-based mainboard supporting RDRAM. However, the situation with Athlon is a bit more complicated and it turns out a serious task to select a mainboard for Athlon, since there are several options. Anyway, aiming at getting the fastest Athlon system available, finally we decided on a board based on AMD-760 core logic and supporting DDR SDRAM. Still, we couldn't resist the temptation to see how today's fastest Athlon works in a system with PC133 SDRAM. This made us add another platform built on VIA KT133A that we have already described in one of the previous reviews. As our tests showed, platforms like that are smart enough even in comparison with DDR SDRAM systems.

As long as nowadays AMD offers two types of Athlon CPUs - 1.33GHz version designed for 266MHz FSB and 1.3GHz CPU for older systems with 200MHz, we considered it quite reasonable to add the results shown by Athlon 1.3GHz in a system with VIA KT133 based mainboard and PC133 SDRAM. Also we wanted to show how greatly the performance increased compared to the previous CPU models, we added the results obtained for Athlon 1.2 (200/266MHz FSB) and for Pentium 4 1.5GHz.

Eventually, our testbed comprised four platforms and there were eight processors tested:

For office and gaming applications were run under Windows 98 SE and the professional OpenGL applications and 3D Studio MAX we chose Windows 2000 Professional SP1.

Performance

Now let's get close to the outcome of our experiments. All the results are put together into three groups. In the first one there are the results obtained in office applications, in the second group we have the performance in 3D games and in conclusion we offer you some numbers characterizing the CPUs performance in professional OpenGL applications.

Business Winstone 2001 is specially created to test the performance in typical office applications. These tasks imply intense work with small data packs. So, it is pretty logical that Pentium 4 1.7GHz lags behind even Athlon 1.2GHz in this benchmark. The architecture of Pentium 4 systems is intended for processing large amounts of streaming data, that is why no wonder that Pentium 4 appeared a complete failure here. Working with RDRAM, which latency is bigger than that of SDRAM, and having an eight times smaller L1 cache for data, Pentium 4 1.7GHz lags behind Athlon 1.33GHz (AMD-760 based mainboard) by 10%.

By the by, systems with Pentium 4 and i845 chipset (Brookdale) will be able to show higher performance in business applications thanks to the good old PC133 SDRAM. But the first mainboards based on Brookdale are expected to appear only in August. So far, we should state that using Pentium 4 for office applications looks more like misusing this CPU :-)

In Content Creation Winstone 2001 we can trace the same trend as in the previous case. Nothing to be surprised at, since this test follows the principles implemented in Business Winstone 2001, with that only difference that it takes the performance in content creation applications. So, we have every right to say that this benchmark is memory- and processor-demanding. Another proof of this fact is a considerable performance increase shown by the Athlon based system when we switched from PC133 SDRAM to PC2100 DDR SDRAM. Thus, it's not for this matter that Pentium 4 hangs back again.

Perhaps, the 10% lag of Pentium 4 1.7GHz behind Athlon 1.33GHz is caused by its too small L1 cache. Besides, it may also be the incorrect branch prediction, carried out by the corresponding processor unit. If so, Pentium 4 will have to clear its entire 20-stage pipelines, which will inevitably lead to lower results in Content Creation Winstone 2001.

Pentium 4 1.7GHz looks even worse in SYSmark 2000. The crucial difference between this test and the previous two ones is that SYSmark 2000 measures the performance in applications which are carried out successively but not simultaneously, as it happened in both the Winstone 2001. This makes the work of the processor and memory buses even easier than before reducing the workload, which means that Pentium 4 has no chances to show everything it is capable of. The outcome is an almost 15% lagging of Pentium 4 1.7GHz compared with Athlon 1.33GHz.

Let us compare the performance of the CPUs in SYSmark 2000 run in applications of different types:

As you can see from the diagrams, Pentium 4 proves worst of all exactly in office applications. In content creation tasks its performance is not so deplorable. Anyway, we should pinpoint that in most business and content creation applications Athlon 1.2GHz goes abreast with Pentium 4 1.7GHz. It is a sheer breakthrough for Athlon!

However, don't forget that the Willamette core in Pentium 4 is a temporary solution. "Real" Pentium 4 CPUs scheduled for Q4 will have another core, Northwood, but it's hard to predict what it will be like.

Here comes a traditional WinZIP archive test. To check the archiving speed, we measured the time needed for zipping a directory with Unreal Tournament. As you've guessed, shorter time stands for better results.

There are two curious things about this test. First, WinZIP is optimized for Intel's processors. Second, the performance of this archiving utility strongly depends on the fastness of the storage subsystem and ATA/100 controller. Bering in mind these two remarks, you shouldn't be surprised about the results obtained. In particular, we finally managed to see Pentium 4 1.7GHz overtaking Athlon 1.33GHz. But we would like to stress another point. The system built on AMD-760 chipset is beaten not only by VIA KT133A based system, but also by VIA KT133 based one. The matter is that unfortunately the bus master driver by VIA (we used VIA service pack 4.29) yet doesn't recognize VIA686B South Bridge if it is paired with AMD-761 North Bridge. So, on mainboards with AMD-760 chipset the ATA/100 controller is much slower than on other boards, for instance those based on VIA's chipset. In the long run it doesn't let the AMD-760 platform get high results in WinZIP.

Well, let us now pass over to the second round of the CPU battle, where we'll see how things stand in games.

The first benchmark, which could be nominally called a gaming test (3DMark 2001 is based on checking the performance in four game scenes), brings a rough change. Pentium 4 1.7GHz, and Athlon 1.33GHz go more or less evenly, with Pentium 4 a bit ahead. Why so? Firstly, 3DMark 2001 is one of the first applications optimized for SSE2, which is implemented in Pentium 4 CPU. And secondly, this test requires high processor bus bandwidth.

We are not shocked to see the impressive results of Pentium 4 in Quake3 Arena. Something of the kind occurred as we tested Pentium 4 1.5GHz. However, we can't find a suitable explanation for these high results obtained not only in this game but also in all other games based on Quake3 game engine. Possibly, it's that Quake3 is better optimized for SSE than for 3DNow!, or it may be for a greater bandwidth of the CPU and memory buses. Most probably, there is a number of factors telling on the results of this benchmark that's why full explanation can be given only by the developers of this game or even by the compiler developers. Meanwhile, we can only state that Pentium 4 1.7GHz surpasses Athlon 1.33GHz with DDR SDRAM by over 30%.

We would like to add that if we consider the relation between the fps rate and the CPU's core frequency, Pentium 4 will break ahead anyway providing 0.153 fps/MHz compared with Athlon's 0.148 fps/MHz. That's why, even in case the CPUs supported the same clock frequencies, Pentium 4 would be the leader.

As the resolution in Quake3 Arena grows, the results level out. But there is an important note to make. All processors except Athlon 1.2GHz showed almost the same results in systems supporting PC133 SDRAM. It means that Athlon 1.3GHz, Athlon 1.33GHz and both Pentium 4 CPUs have got so fast that they can completely load even such powerful graphics accelerator as GeForce2 Ultra. Since the graphics card becomes the bottleneck of the system, the fastest testing participants showed equal results in the most common resolution: 1024x768x32. This way, if you want to assemble a gaming system based on elder AMD or Intel CPUs, you should make sure that your graphics accelerator is powerful enough. As we can see, the performance of GeForce2 Ultra is not enough to be used in such cases and extreme gamers have nothing else to do but stargaze at GeForce3.

In Quake3: Team Arena based on the Quake3 gaming engine but with a more complicated model and textures, the results resemble the formerly obtained ones. Pentium 4 1.7GHz is 19% ahead of Athlon 1.33GHz.

Once again higher resolution brings about closer results.

Unreal Tournament is based on a slightly updated Unreal engine, so any good optimization for SIMD instructions is out of the question. Therefore, of crucial importance are the bandwidths of the CPU and memory buses alongside with the performance of the FPU (the strong point of Athlon CPUs). As a result, the situation is just the opposite to what we saw in Quake3: Athlon 1.33GHz is a little bit faster than Pentium 4 1.7GHz, though the latter has higher clock speed.

At the resolution of 1024x768x32 we can see almost the same thing. According to the game specifics, the major determinant of the fps rate appears not the graphics subsystem performance, but the computing capacity and the bandwidths of the processor and memory buses.

In the old good MDK2, like in Unreal, Athlon 1.33GHz breaks ahead of Pentium 4 1.7GHz, leaving it 5% behind. We would like to stress that in this test Athlon systems based on VIA KT133A with PC133 SDRAM overtake systems based on AMD-760 with DDR SDRAM. What's the matter? It may look strange as long as PC2100 DDR SDRAM seems to provide twice the bandwidth! It turns out that sometimes this doesn't suffice. We have already described this phenomenon in every detail in our article called "VIA KT133A in Action: EPoX EP-8KTA3 Mainboard". Let us remind you that the old MDK2 engine is very sensible to the memory latency. Subsequently, the system with PC2100 DDR SDRAM CAS latency 2.5 loses to its opponent with PC133 SDRAM boasting CAS latency 2.0. That is why the Pentium 4 platform with RDRAM featuring higher latency lags behind Athlon platforms.

Higher resolution reduces the difference, but on the whole things don't change that much.

One of the most up-to-date games, it deals with extremely large textures. That's why, the performance in it depends on the bandwidths of the CPU and memory buses. No wonder that Pentium 4 1.7GHz outpaces AMD Athlon 1.33GHz once again.

Resolution growth doesn't cause any drastic twist in the results again.

To summarize the processors performance in the gaming applications, we need to stress the following aspects. For our tests we used games based on five gaming engines. In three cases the leadership belonged to Pentium 4 1.7GHz, two other rounds were won by Athlon 1.33GHz. On the face of it, no definite conclusions can be made. But if we look, in which games AMD's solution won the laurels, we'll meet the old MDK2 and Unreal Tournament engines. In games based on the more recent engines, Pentium 4 gets its own back. Is that a contingency? Most probably, it is a trend. Modern gaming engines involve effects, which require high memory bus bandwidth and use hardware T&L unloading the processor FPU (remember, it is Athlon's advantage). Moreover, the new set of SIMD instructions from Intel, SSE2, is more flexible and convenient for software developers than 3DNow! . It gives us every reason to expect that Pentium 4 proves faster than its AMD competitor in the forthcoming games.

Now we'll deliver the details of the final part of our testing dedicated to professional OpenGL applications. All these tests were run in Windows 2000.

To start with, we took the professional 3D modeling software, 3D Studio MAX. To estimate the processor performance we checked the time each system required to render Anisotropic Wheel scene at 800x600. So, take note that the smaller is the value (the less time the system needed), the better.

Since in 3D Studio MAX the major workload falls upon the processor FPU and not upon the memory bus, we could say that the results obtained in it look just like "Quake3 upside down". AMD Athlon with a powerful floating point unit easily gets ahead of Pentium 4. Even working at 1.7GHz, Intel's hero fails to keep up with Athlon 1.2GHz.

Other OpenGL benchmarks include tests from SPECviewperf 6.2.1. Their peculiarity is rendering the viewport in real time and visualizing the result by means of OpenGL driver. Their main difference from games is the approach to the image quality. 3D games are focused on the realism of the image provided by pretty high fps rate. When rendering they use a relatively small number of polygons (circa a ten of thousands), but on the other hand, they are overlaid by large textures and a lot of effects are applied to them. An approach like that would never work in professional applications, hence the ultimate rendering is performed by the CPU, while the graphics cards forms just a quick preview in the viewport. That's why there are no effects used in the applications of the kind, and to get a more realistic photographic image a great number of polygons is used (about hundreds of thousands). That will result in much lower fps rates, as you will see on the diagrams below, compared with the fps rates we obtained in games.

Advanced Visualizer is a rendering-simulating test from the software of the same name by Alias/Wavefront, which is designed for 3D modeling, animation and rendering. If you are particularly interested in this test, click here for more details. On the diagram you can see the triumph of Athlon CPU. The reasons are the same as in 3D Studio MAX. The performance of Athlon's FPU is much higher than that of Pentium 4. And since most calculations needed to render the image and to display it on the screen take place right in the FPU, Athlon 1.33GHz outpaces Pentium 4 1.7Hz by 30%.

In DesignReview - software for 3D models investigation (for details click here) - things are totally different. The model used in the test contains almost 400000 vertexes, and all the data about them is stored in the RAM. This time it's the memory bus bandwidth that is of great importance during rendering, so it gives Pentium 4 1.7GHz a good chance to rush forward overtaking Athlon 1.33GHz.

IBM Visualization Data Explorer is meant for visualizing and analyzing scientific data. A more detailed description of this test is available here. By visualizing, this benchmark uses the particles' paths in the vector field. As it has been mentioned earlier, scientific calculations are Athlon's profile, so the results are quite easy to explain.

Lightscape Visualization System by Discreet Logic is a system for interactive light reckoning (you may read about it here). As in the previous test, Athlon retains the leadership. Working in a DDR SDRAM system with high bandwidth, this CPU is again faster than Pentium 4.

This test measures the speed of standard rendering in real time (click here for details). Again it goes about a highly detailed object (comprising about 450000 triangles). So, the performance is first of all determined by the memory bus bandwidth, enabling Pentium 4 to regain its leadership. At the same time its success is partially the result of the CPU's fast ALU working at twice the CPU frequency.

As this rendering test uses a more complicated object than the previous one, the gap between the slower Athlon and Pentium 4 equipped with two-channel RDRAM is even greater. As for the CDRS test by Parametric Technology Corporation, it is a common industrial design software (click here for details).

We are afraid it's too hard to point out a 100% leader in professional applications. It is clear, however, that the performance here is influenced by two matters: the FPU performance and the memory bus bandwidth. And when an application requires a lot of calculations from the CPU, Athlon 1.33GHz proves the best. If high memory bus bandwidth and high-performance ALU are needed, Pentium 4 1.7GHz looks just splendid.

Conclusion

Well, the tests are run. Unfortunately, we wouldn't give any definite recommendations. Each CPU is good only for a certain type of applications. Thus, it's up to you to decide, which processor to chose. For this purpose you had better browse our tests results one more time. As you have learned, Pentium 4 works well with modern games, and Athlon is better for office applications and content creation programs. Speaking about professional OpenGL applications, it all depends on how the load is shared between the processing unit and the memory bus.

It should be stressed that nowadays AMD and Intel retain certain parity, making it impossible to single out the leader. Of course, Pentium 4 is far too pricy yet and demands the unattractive RDRAM, though Athlon too suffers some shortcomings. In particular, its 266MHz versions, especially the elder ones, are not available countrywide, besides DDR SDRAM is still much more expensive than PC133 SDRAM.

It feels like by the time the new processor cores - Palomino and Northwood - do come out, AMD and Intel score evenly. Any further events remain unpredictable. A shift to 0.13micron technology will help Intel enlarge the CPUs' L2 cache and cut their price. AMD in its turn will sponge on the lower power consumption of its on-going core for Athlon to catch up with Pentium 4 in clocking. Apparently, we are up to another fascinating round of the CPU giants' competition, where the winner is bafflingly hard to define.