This evaluation of current desktop processors utilises over 60 benchmark tests including office and multimedia software, 3D games, Internet applications, video rendering and compression. We have used benchmarks that are relevant to a range of market sectors in order to get a balanced view of CPU performance. However, in an ideal world, you should also run your own mission-critical applications on any processor that you're considering.
Of course, performance is only one aspect of a processor purchase decision. For example, the Athlon 64's support for the NX (No Execute) feature safeguards it from certain virus attacks, and could be reason enough to choose an AMD processor. And if you're after a quiet PC, then AMD's chips have clear advantages over Intel's latest 'Prescott' Pentium 4. The power consumption of the Athlon 64 is lower than that of the Pentium 4 thanks to AMD's use of Silicon-on-Insulator (SOI) technology. The Athlon's 64's 64-bit capability is also a potential advantage, although this feature remains largely unused because of the missing operating system support -- 64-bit Windows XP has now been delayed until 2005.
|Athlon 64 3800+ (Nforce 3)||91W||172W|
|Athlon 64 3800+ (KT800 Pro)||82W||162W|
|Pentium 4 560 (925X)||155W||258W|
Note: The power consumption figures quoted in the table above refer to a complete system with otherwise identical components.
As a baseline for this comparison, we used Windows XP SP1 with all Hotfixes up to the point of testing. The standard resolution was 1,600 by 1,200 pixels, although the 3D gaming tests were also carried out at 1,280 by 1,024 and 1,024 by 768 pixels.
In the 3D gaming tests, image quality was boosted by turning on anti-aliasing (4x AA) and anisotropic filtering (8xAF). Most users of modern graphics cards like the Nvidia Geforce 6800GT -- as used for this test -- will activate these settings, so their effect on performance is worth considering.
For this performance comparison, we selected Intel and AMD's current fastest desktop processors. Intel's Pentium 4 560 running at 3.6GHz and the 3.4GHz Pentium 4 550 feature the new Prescott core and 1MB of on-board Level 2 cache. Both P4 processors are housed in the latest D925XCV motherboard from Intel.
The Athlon 64 3800+ running at 2.4GHz and the 2.2GHz 3500+ are housed in MSI's Nforce3-equipped K8N Neo2 Platinum motherboard. Both systems have 1GB of main memory: DDR400-RAM is used in the Athlon system, while the Intel board is fitted with the new DDR2-533 RAM.
|Athlon 64 3800+||2.4GHz||939||$643|
|Athlon 64 3500+||2.2GHz||939||$346|
|Pentium 4 560 (925X)||3.6GHz||775||$637|
|Pentium 4 550 (925X)||3.4GHz||775||$417|
|Motherboard||MSI K8N Neo2 Platinum||Intel D925XCV|
|Chipset||Nvidia Nforce3 250GB||Intel 925X|
|Graphics card||Nvidia Geforce 6800GT (AGP)||Nvidia Geforce 6800GT (PCI-E)|
|Hard disk||Maxtor SATA Maxline III 250GB||Maxtor SATA Maxline III 250GB|
|Memory||2x Corsair CM512-3200LL (CL225)||2x Micron DDR533 (CL444)|
|OS||Windows XP SP1 with current Hotfixes||Windows XP SP1 with current Hotfixes|
Memory & cache tests
The memory and cache tests determine how quickly data is transferred between the CPU and the memory subsystem. This does not indicate how a system will perform overall, but it's an important component of a balanced evaluation.
The results of the ScienceMark 2 test are interesting -- especially if you use software that's sensitive to long memory access times. Memory bandwidth is the first performance criterion -- the total output of a memory subsystem depends on the bandwidth and the latency. The latter is measured in bus cycles, with lower values better. With bandwidth, which is measured in megabytes per second (MB/s), a higher value is better.
Unfortunately, ScienceMark 2 can't measure bandwidth of the L1 and L2 cache for the Prescott Pentium 4. However, results for the Athlon 64 show that, thanks to the integration of the memory controller, cache performance increases with clock frequency. Both with latency and bandwidth, the Athlon 64 delivers better performance than the Pentium 4. To a large extent, this underpins the very good overall performance of the two AMD processors.
Office & multimedia benchmarks
Business and Content Creation Winstone are two benchmarks that carry out typical application-based tasks like a converting video files or delivering a PowerPoint presentation. The following applications are used in each test:
Content Creation Winstone 2004
Adobe Photoshop 7.0.1
Adobe Premiere 6.50
Macromedia Director MX 9.0
Macromedia Dreamweaver MX 6.1
Microsoft Windows Media Encoder 9.0
Newtek Lightwave 7.5b
Steinberg Wavelab 4.0f
Business Winstone 2004
5 Microsoft Office XP applications (Access, Excel, FrontPage, PowerPoint and Word)
Microsoft Project 2002
Norton AntiVirus 2003
The benchmark offers two options. The first is a standard a test without multitasking: several applications are held in memory, but only the foreground application is active. This corresponds, for example, to the work routine of a Web designer, who first creates graphic elements in Photoshop and then uses Dreamweaver to integrate this into the Web site.
The two Athlon 64 processors keep their Intel counterparts at bay in these tests, with a performance advantage of between 10 and 16 per cent. Things look completely different in the multitasking tests, which are described on the next page.
Business Winstone 2004 also has a multitasking mode: here, different scenarios are simulated in which applications are active in both the foreground and the background.
In the first scenario, files are copied in the background, while in the foreground Outlook and the Internet Explorer process certain tasks. In the second scenario, a script kicks off processes with Word and Excel in the foreground while WinZip files are compressed in the background. In the third section, an antivirus test with Norton AntiVirus is run in the background, while in the foreground all the components of Microsoft's Office suite are busy with tasks.
In the multitasking test things are clearly better for Intel, where its HyperThreading (HT) technology comes into play. Altogether, the Intel processors complete the tests up to 9 per cent faster than their AMD counterparts. In two of three tests the Pentium 4 chips are in front; the AMD chips are only faster in scenario 2 (Word, Excel and WinZip).
Internet benchmarks: iBench 5.0
Many users feel that performance differences between processors are not significant when it comes to surfing the Web. But this is only the case if the Internet is accessed via a slow modem connection; even then, significant differences can emerge if content (Java applications, PDFs) must be downloaded before execution. If a fast DSL connection is used, processor speed becomes a significant factor.
Intel's Pentium 4 only has the upper hand in the XML test, with a 20 per cent advantage. There are no significant differences between Intel's and AMD's chips in the Java VM test. In the remaining five tests, the Athlon 64 beats the Pentium 4 by between 19 and 66 per cent.
We would like to report further benchmark results in the Internet performance section, but unfortunately BAPCo's WebMark 2004 refused to work with our test systems. It's possible that with a pure XP installation without service packs and Hotfixes is required; the use of older versions of Flash players and Sun's Java VM could also be a problem. Also, iBench 5.0 accepts the most current plug-ins, while WebMark only supports with certain versions. We are working with BAPCo to resolve these problems, and will deliver the performance results in due course.
Gaming performance: AquaMark 3
The AquaMark benchmark is currently the most complex 3D gaming test available. This can be seen from the fact that even without anti-aliasing and anisotropic filtering, there are few differences between processors.
Obviously the test routines are so complex that the performance of the graphics card – even a GeForce 6800 GT -- is the limiting factor. The AquaMark test is a good indicator of the CPU performance with current 3D games but should not be used as the only decision criterion. Therefore in the following pages we report further tests with current and older 3D games.
Gaming performance: Colin McRae Rally 4
With Codemasters' rally simulation Colin McRae Rally 4 there are -- as with the AquaMark test -- no significant performance differences between Intel's and AMD's processors. As before, it's clear that the graphics card is the limiting factor, masking any differences between the CPUs.
Gaming performance: F1 Challenge 99-02
Electronic Arts' Formula 1 simulation F1 Challenge 99-02 is an established title, but it continues to enjoy a good reputation with 3D gamers. Third-party updates for the new Formula 1 season are already available from designer Ralph Hummerich. At the moment, since Electronic Arts has no plans for an update, this is the most up-to-date version of the Formula 1 simulation. For this test we used the latest Ralph Hummerich update and drove a circuit around Imola. Performance was evaluated using the FRAPS tool and the stored Replay. By contrast with the AquaMark and Colin McRae Rally 4 tests, there are significant performance differences between the CPUs in this test. Without anti-aliasing or anisotropic filtering, the Athlon 3800+ is up to 12 per cent faster than the Pentium 4 560. With the image quality features turned on, the AMD chip's lead ranges between 6 and 13 per cent.
Gaming performance: Unreal Tournament 2003
Of all the 3D gaming tests, the one using Unreal Tournament 2003 reveals the biggest performance differences between AMD's and Intel's processors.
This is because the Botmatch test stresses the performance of the CPU more than that of the graphics card. AMD's processors out-perform their Intel counterparts by up to 25 per cent. Even with anti-aliasing (4xAA) and anisotropic filtering (8xAF) enabled, the advantage remains approximately the same.
Gaming performance: Comanche 4
AMD's chips are clearly faster than their Intel counterparts in the Comanche 4 test, achieving a maximum lead of 20 per cent. At the highest resolution of 1600 by 1200 pixels with anti-aliasing and anisotropic filtering enabled, the advantage drops to around six per cent.
Workstation benchmarks: 3D Studio Max 6
3D-rendering is not necessarily a suitable job for a desktop PC -- most users of such applications will work on dual-processor workstations. As a result, the performance results presented here are not 100 per cent representative. However, if you're thinking of buying a single-CPU workstation (which numerous PC manufacturers offer), these results are perfectly relevant. Although the Pentium 4 is usually behind the Athlon 64 in the 3D gaming tests, the Intel chip puts up a better showing with 3D Studio Max. The software is not only SSE2-optimised, but also supports Intel's HyperThreading technology. The results speak for themselves: although the Athlon 64 keeps up with the Pentium 4 with less complex scenes such as 3dsmax, the Intel processors are up to 31 per cent quicker in the other tests.
Workstation benchmarks: Lightwave 3D 8
The Pentium 4 is also the winner in the second workstation test, using Lightwave 3D 8. However, the Intel CPU's lead is not as marked here as it was with 3D Studio Max -- usually only a few percentage points. The Radiosity Reflective scene is rendered 18 per cent faster by the Pentium 4 than the Athlon 64. Clearly, it's necessary to examine several scenes in order to get a balanced performance assessment in this test.
Video performance: Video Studio 8
When processing a video with Ulead's Video Studio 8, the Athlon 64 comes out on top: AMD's chips have a lead of between five and eight per cent over Intel's Pentium 4 in this test.
Video performance: TMPEG / PowerDVD Copy
Compared with Video Studio 8, an editing/processing suite, TMPEG Enc 3.0 Express is essentially limited to video conversion and encoding.
TMPEG Enc 3.0 Express is one of the few applications to support all of the Pentium 4's special features. Besides SSE2 and HyperThreading, it also supports the 13 new instructions introduced with the Prescott core. As a result, the Athlon 64 is beaten hands-down: AMD's chips cannot keep up with the competition in this test, lagging behind by up to 36 per cent.
In the test with the DVD copy software PowerDVD Copy, the picture changes: here, the Athlon 64 is between three and six per cent faster than the Pentium 4.
Video performance: Power Director 3
The final video test, using Power Director 3, delivers variable results. When producing MPEG1 videos, the Athlon 64 processors are up to 10 per cent faster. In the other two tests, the Intel processors take the lead.
So, after performing many tests, which is the fastest CPU -- AMD's Athlon 64 or Intel's Pentium 4? The answer depends on which software you use: there is no clear winner here.
The Athlon 64 has an advantage with mainstream office and Internet applications, and 3D games. Meanwhile, the Pentium 4 can always score with optimised software -- particularly 3D and video rendering. The multitasking tests also show an advantage for Intel's Pentium 4.
The Athlon 64 has a clear advantage in terms of cost. Although AMD's and Intel's fastest chips are comparably priced, system costs are higher for Pentium 4-based PCs with DDR2 memory. The price difference between Athlon 3500+ and Pentium 4 550 systems is even larger, because the AMD chip costs $71 less than its Intel counterpart.
Post-purchase costs should not be ignored, either. A Pentium 4-based system uses 60 to 70 Watts more power when idling and up to 89 per cent more power relative to an Athlon 64 PC. For a household with a few computers, this may not add up to a significant cost saving, but for larger enterprises with several thousand PCs the sums will look completely different.
The Athlon also supports the NX feature, which offers protection from certain viruses in conjunction with Windows XP Service Pack 2. Intel has announced that the Pentium 4 is to support the NX feature later this year. The Athlon chips' 64-bit capability is currently no advantage, as 64-bit Windows XP isn't due to ship until 2005.