It's no secret that the clock growth in modern x86 CPU has significantly slowed down of late. Even new cores and process technologies are of no help to manufacturers. So, it's a sad picture for sensation-mongers. But every cloud has a silver lining: instead, Intel managed to do the impossible again and launched two desktop processors, but with 2 MB L2 Cache. One of them is not even eXtreme Edition, but a regular Pentium 4 with Processor Number 660. Both of them are based on the new revision of Prescott core, equipped with 2 MB L2 Cache: sort of a record for desktop CPU. The reason for this record lies on the surface actually: one can say that the further build-up of frequencies in Prescott core is "officially cancelled" (Intel removed the 4 GHz processor from its roadmap), but nevertheless performance should grow somehow. Well, it doesn't really matter for us what it will take it to grow, does it? Let's see whether the innovation has an effect.

Besides, both new processors support EM64T, a counterpart of AMD64, which was previously known as x86-64. Again this is something new for desktop CPUs from Intel – EM64T had been previously used only in Xeon Nocona for servers. However, despite the catchy enthusiasm of AMD concerning 64 bit calculations on desktop platforms and that even Intel joined this initiative, the major "helmsman" Microsoft (currently known as "Senior Program Architect") is not in a hurry to cheer us with the official release of the 64 bit version of Windows for AMD64/EM64T. That's why there is nothing for us but to analyze the performance aspects of the new processors, which are vital to most users: test execution speed under 32 bit Windows XP and "ditto-bit" software.

A tad later we are going to publish test results of the new processors in SPEC CPU under one of 64-bit Linux versions – specially for those who are looking forward to 64 bit sensations. Those who are interested in architectural features of processor cores are recommended to read Dmitri Besedin's article. The senior developer of RightMark Memory Analyzer traditionally researches new cores with his benchmark package and with his ultimatistic meticulous nature manages to fish out a couple of interesting issues. Thus, leaving the future to the future and fine points to professional researchers, let's start the procedure, which still appears the most informative despite its jaded character: performance tests in real applications. By the way, to avoid extreme traditionalism, we have recently considerably changed the list of tests – a separate article is devoted to this issue. Thus, the tests below are carried out by the new method. Besides, we have even slightly modified it relative to the described in the article (let's hope – for the better).

Testbed configurations

• Processors
  • Intel Pentium 4 eXtreme Edition 3.73 (266x14) GHz, LGA775, 2 MB L2
  • Intel Pentium 4 660 (200x18 GHz), LGA775, 2 MB L2
  • Intel Pentium 4 eXtreme Edition 3.43 (266x13) GHz, LGA775, 512 KB L2, 2 MB L3
  • Intel Pentium 5,560 (200x18 GHz), LGA775, 1 MB L2
  • AMD Athlon 64 FX-55 (2.6 GHz, Socket 939)
• Motherboards
  • ASUS P5AD2-E Premium (i925XE chipset, Socket 775)
  • Engineering sample of a motherboard on ATI Xpress 200P (RX480, Socket 939)
  • Albatron K8X890 Pro (VIA K8T890 chipset, Socket 939)
• Memory
  • 2x512 MB PC3200(DDR400) DDR SDRAM DIMM Corsair, 2-2-2-5
  • 2x512 MB PC2-4300 (DDR2-533) DDR2 SDRAM DIMM Corsair, 4-4-4-11
• ATI Radeon X800 (PCI Express x16)
• Samsung SP1614C (SATA) HDD, 7200 rpm, 8 MB Cache
• Windows XP Professional SP2, DirectX 9.0c

A small comment on the testbed configurations: some of you may have already noticed that there are two motherboards listed for the AMD Athlon 64 FX-55 based testbed. As chipsets for Socket 939 with PCI Express support are new, we decided to be on the safe side and run some tests on two motherboards to be able to compare the results. No significant differences have been detected, that's why you will see only one column for this processor on the diagrams.

Test results

"Semisynthetics"

CPU RightMark (RMCPU 2004B)

In fact, CPU RM comprises two tests. The solver module calculates interaction of solids and the render module displays this interaction on screen. Besides, the render module can use up to 16 processors, as it's multi-threaded, while the solver module uses only one processor irregardless of the number of CPUs in the system. As we have already ascertained in practice before, the correlation of CPU performance of various architectures in solver and render modules is not always the same.

Quite a habitual distribution of results depending on a task, we have already come across it before many times: Intel is leading in rendering (due to its multi-processing module among other things), AMD – in calculations. The "old" Pentium 4 eXtreme Edition on Gallatin core (Northwood + L3) is slightly off the rail. CPU RightMark is obviously not its best test: it's the last in calculations, and it has nearly lost to Athlon 64 FX-55 in rendering (traditional demesne of Intel). You can notice almost no effect of the large cache in this test: Pentium 4 660 is almost on a par with Pentium 4 560.

Archivers

7-Zip can use several processors, so it gets some advantage from Hyper-Threading (in some of our previous articles we researched this issue and found out that this advantage may be quite noticeable). So the advantage of all Intel processors over Athlon 64 FX-55 looks appropriate here. You may notice that it's not too large, though.

The situation with WinRAR is absolutely different: the advantage of Athlon 64 FX-55 over two out of four Intel processors is rather large, and only two models (both of them – eXtreme Edition) almost catch up with the leader. It's not difficult to guess that they profit from two features: the large cache (2 MB L3 cache in P4 XE 3.43 GHz and 2 MB L2 Cache in P4 XE 3.73 GHz) as well as the support for FSB with the 1066 MHz equivalent effective bandwidth. It's extremely significant that P4 XE 3.43 GHz outscored a higher clocked Pentium 4 660, though the latter also has a large L2 Cache (but its bus is worse – "just" 800 MHz).

3D Graphics

In this case we do not mean the 3D Graphics in general, but "3D modeling". It's represented by three most popular programs: 3ds max, Maya, and Lightwave.

SPEC apc for 3ds max 6

The test from SPEC is a script for 3ds max 6. It includes rendering as well, but as the built-in 3ds max renderer is of little interest from the point of view of professional use, we take into account only Interactive test, that is the imitation of user operations in this software package.

The interactive test demonstrates an evident advantage of the top AMD processor. It's significant that the situation is contrary in rendering, as demonstrated by our previous tests. Pentium 4 "rules" there. It's a funny distribution of roles: it turns out that it's more convenient to work in this program on Athlon 64, but final renders are better on Pentium 4. However, professional studios can just afford the simple but expensive solution: to use advantages of both systems, they can equip designers' workstations with AMD processors and use another Intel based system for rendering.

Maya 6

In this test we switched to SPECapc for Maya for Version 6 (until that moment we would run SPEC apc for Maya 5 under Maya 6). From the point of view of the test process organization, the SPEC test for Maya is similar to the 3ds max test – it's also a script that imitates various user actions. But in this case there is no final render test.

It's funny that the CPU test has two pronounced leaders – top processors from Intel and AMD, but in total score the situation gets more even. In our opinion it indicates the strong influence of a 3D accelerator in Maya 6. It's also interesting that the CPU test is obviously predisposed to Prescott core in combination with large cache (so called Prescott-2M core), while in total score the first two places are taken by Intel processors with the fastest bus. One of them is based on the older Gallatin core (it's actually the Northwood core with L3 Cache).

Lightwave 3D 8

Unfortunately, we didn't manage to find an interactive test for Lightwave 8, so in this case we'll have to content ourselves with rendering a relatively complex scene. By the way, here it is.

In our opinion it's one of the clever scene examples for Lightwave 3D, which corresponds to usual operations in this 3D software package. This scene was found in one of Lightwave forums and can be downloaded for free.

The large cache in new Intel processors did not allow them a quantitative leap, but provided a qualitative leap instead: before this time (you can judge by Pentium 4 560 results) Intel had no "regular" (not eXtreme Edition) processor, which would outscore the "extreme" processor from AMD. Now it has it: Pentium 4 660. However, even despite the large cache, Lightwave 8 still does not like Prescott core (including the 2M modification). In this test it obviously succeeds "in number" (megahertz, cache) rather than "by dint". The proof of this fact is the result of still second-to-none Pentium 4 eXtreme Edition 3.43 GHz on old Gallatin core.

Bitmap graphics
and prepress operations

The main test in this section is a script for Adobe Photoshop CS (8), developed in our testlab. it includes the most frequent operations: Blur and Sharpen filters, RGB -> CMYK -> Lab conversions, lighting effects, image rotation, resize, transform type operations. These actions are performed over a real photograph taken by a digital camera. At multiple requests from our readers we have also included an Adobe Acrobat Distiller test into this section – PS into PDF conversion. It uses several real iXBT.com magazine articles.

Adobe Photoshop CS (8)

The picture is motley but explainable: the old Intel core and AMD architecture are out of favour, Prescott is good. The new modification of this core with 2 MB cache is little faster than the old one – the further increase of its size seems not critical to Photoshop.

Adobe Acrobat 6 Distiller

Adobe Acrobat Distiller gives implicit preference to Intel architecture – NetBurst, both to its older modification represented by P4 XE 3.43 MHz and new processors with Prescott[-2M] core. On the whole it's a demonstrative example of an application, where the leading role is played by frequency. Let's compare the ratios: Pentium 4 560 is 46% faster at this task that Athlon 64 FX-55. The frequency ratio is 3600/2600, P4 560 frequency is 38% higher. Yes, the architecture actually matters here, but the figures are still more or less comparable...

CAD/CAM

SolidWorks 2003

According to the SPECapc tradition, the test script imitates user operations and outputs four results upon completion: total score, performance of the graphics system, input/output system, and processor. It should be noted that SPECapc for SolidWorks 2003 retains the performance rating in points, but the best result here is the lowest one.

Architectural preferences of SolidWorks 2003 are clear: a processor from AMD has won with a great breakaway from other contenders. The other results are less clear, but let's try to get down to the solid bed-rock. So, a large cache seems not very important in itself: the advantage of Pentium 4 660 over Pentium 4 560 is not very large. P3 XE 3.43 is also not impressive: despite the bus, it is defeated by Pentium 4 660. Let's assume that it's due to the old core – it's quite logical, isn't it? Then it turns out that NetBurst can profit only from the combination of three components: a large cache, Prescott-like core, and a fast 1066 MHz FSB. After that there is nothing but frequency growth...

Media encoding

This section comprises everything connected with encoding video and audio data, that is classic WAV -> MP3 conversion as well as video compression with popular codecs.

Encoding audio

Good old LAME... Due to the great number of presets and their votaries, we have taken the compromise tack: encoding with maximum possible quality: (320 kbps CBR, q=0) and VBR encoding from 160 to 320 kbps with "high" ("-q 2" option, or simply "-h") quality. Then we calculate the geometric mean of the results obtained.

Introduction of the second test without the Q=0 option ("unloved" by AMD processors) has not changed the situation cardinally, and the reason is simple: all processors came to finish approximately at the same time with the options "-b 160 -B 320 -m j -q 2 -V 0": the worst result (Pentium 4 560) is equal to minus one minute and six seconds, and the best one (Athlon 64 FX-55, Pentium 4 eXtreme Edition 3.73 GHz, Pentium 4 660)... is one minute sharp. Thus, one can say that equal processors from both manufacturers offer practically no difference in MP3 encoding with LAME in Q>0 modes. When Q=0, Intel Pentium 4 processors are victorious. By the way, note that the best result in Pentium 4 processors is demonstrated by the old core.

Encoding video (MPEG4)

The results on the diagram below are the fruit of a compromise between our desire to cover as many popular codecs as possible and the reluctance to crowd the article with diagrams. You can see the average encoding time for a test file with three codecs: DivX, XviD, and Windows Media Video 9. To calm down those who fear the effect of the "average temperature of patients in a hospital", we'll be more precise: the rating distribution among separate codecs and on the summary diagram is absolutely the same.

We have tested Athlon 64 FX-55 for the first time and here is the surprise: even though video encoding is a traditional strong point of Intel processors, the first place is taken by the top CPU from AMD. It's easy to notice though that the advantage is very small. In fact all processors coped with this task quite well. So in this case we'd better prefer the cheapest processor but not the fastest: the performance difference is just a couple of percents...

Encoding video (MPEG2)

By multiple readers' requests we now use Canopus encoder in this test – ProCoder 2 (we had been using Mainconcept MPEG Encoder 1.4 before that).

It's another test that was not to AMD fans' liking, they even accused us of being prejudiced: as if we used Mainconcept MPEG Encoder on purpose – to make Intel processors look better. Let's hope there will be no objections to Canopus ProCoder, especially as they recommended to use this encoder. So, what are the results? Pentium 4 is still better :).

3D visualization

Modern 3D games

We use the 640x480x32 mode with minimum quality settings in all the games. That's why strictly speaking, this section cannot be used for testing real gaming performance. However, it couldn't pretend to this role even if the settings were more realistic, because we use only one video card anyway – and gaming performance depends on this component to no lesser degree than on a processor.

So, why do we need this test? Mostly out of theoretical interest rather than of practical one, actually. Having minimized video card's influence, we analyze processor preferences of game engines. It would have been certainly of little interest, if we had used only one game. But four state-of-the-art games and their averaged results allow to answer the following question with more precision: "Which CPUs are preferred by modern game engines?" Later on we'll probably introduce more games into this test, but we shall still analyze the last diagram only: summary. Just from the point of view described above.

An impressive victory of AMD. And probably the most impressive victory in these tests. Of course, we can recall Adobe Distiller, but it was just one application, and here is the summary diagram of the four tests. We'll risk an assumption that even Half-Life 2 (which is not included into the tests for now) will not be able to change the situation anyway. Thus we can safely recommend AMD processors for modern games. In real gaming modes the CPU advantage will certainly be less evident, because it'll be also influenced by a video card. But if one figure is greater than the other, the latter will never be greater than the former, no matter how you divide them, won't it?

3D modeling packages

It's quite an easy diagram to interpret: the larger a cache is, the better. That's how the groups are rated: two processors with 1 MB L2 are on the bottom, two processors with 2 MB cache are a tad higher (a cache level doesn't seem to matter in this test), 2 MB Cache CPUs with the fast 1066 MHz bus are on top.

Conclusion

First of all we'll compile a small table, which provides the most interesting way (though not indisputable!) to evaluate results of all the tests on the whole. It provides results of all processors in all tests percentagewise, where the worst result is 100% (in various tests it's demonstrated by different CPUs). And at the end of the table the percentage values obtained are calculated into an arithmetical mean. Of course, this rating is very approximate, because it assumes that we are equally interested in all test results. But considering the taste variety of various users this method seems to us at least no less preferable than any other.

So, if we treat the above table as sufficient grounds for conclusions, there will be two main conclusions:

1. In total score the new top processor from Intel snatched the palm from AMD (I can fancy Intel tearing it up by the roots...), moreover: the new "non-extreme" Pentium 4 660 is on a par with Athlon 64 FX-55 and outscores the previous P4 XE from Intel. Looking at this dull picture you understand the vanity of high-sounding titles like "eXtreme Edition" or "FX": they are useless, if in a short time afterwards (six months or earlier) there appears a regular faster processor.
2. The difference between the worst and the best results is... 9%. That is considering the average performance of the reviewed processors, users will hardly feel the performance difference among the systems. As is well known, human sensitivity is much lower than that of a benchmark. There appears a fair question: isn't that too much ado about nothing? The answer to this question is up to you personally.

We can just establish a fact: from the point of view of top CPU performance comparison, everything still comes to the ever-going argument who "the master of the hill" is, that is a formal leader. Everybody seems to know that no manufacturer can offer a solution that will destroy the competitor. Or doesn't want to. By the way, it's not that absurd as it looks at first. There is also possible a combination of these scenarios: one manufacturer deliberately "holds down" the release of its new product, the other one is straining to the top, reaches it... and is soon answered with a long-ready CPU from its competitor. We shall most likely never know which of the three scenarios is real... and frankly speaking we don't want to. There may be a great number of reasons, but the consequence is only one: CPU performance does not grow in leaps, but in small calculated portions. And the change of a formal leader is as inevitable as the change of seasons. Show goes on this way or another! Only the audience is tired. That's why our forecast will be paradoxical: by the time these two leading competitors on the x86 CPU market finally find out who's the boss – it's will be of little interest to users.

However, the above said does not diminish Intel's victory. It's not important who has won, after all: what's important is that the performance bar has been raised again. You may not worry about an industry on the whole, where the bars are raised from time to time: it's on the move...

Article navigation: