AMD has led the 64-bit desktop processor field up to now with the Athlon 64 -- although there is, as yet, no shipping 64-bit Windows operating system, which is necessary to make the most of a 64-bit CPU. However, at the beginning of February Microsoft made Release Candidate 2 of Windows XP Professional x64 Edition available to testers, and the final version is due to ship at the beginning of April.
With its 6xx-series Pentium 4 processors, Intel has finally jumped on the desktop 64-bit bandwagon. Intel's EM64T architecture can implement Windows' x86-64 mode, and is also compatible with the AMD64 architecture. This means that you don't need different Windows operating systems to take advantage of 64-bit processing.
Intel's new processors offer some other innovations besides the 64-bit EM64T architecture. For example, all 6xx-series CPUs deliver 2MB of Level 2 cache -- double the amount on the 5xx-series chips. There's also improved power management in the shape of Enhanced Intel Speedstep Technology (EIST) and protection from malware infection via memory overflow thanks to the eXecute Disable (XD) bit.
The 6xx-series Pentium 4 chips have an 800MHz frontside bus (FSB) and are designed for use in standard desktop PCs. Intel is also offering a new Pentium 4 Extreme Edition clocked at 3.73GHz with a 1,066MHz FSB. This expensive ($999), high-end chip is only likely to appeal to a limited market -- mainly power users and gamers.
According to Intel, the 6xx-series Pentium 4s will replace the older line in the medium term. However, it may be some months before you can no longer buy 5xx-series chips.
|6xx-series Pentium 4 processors|
|Processor ||Clock speed ||Level 2 cache ||EM64T / XD / EIST ||Price|
|Pentium 4 660 ||3.6GHz ||2MB ||yes / yes / yes ||$605|
|Pentium 4 650 ||3.4GHz ||2MB ||yes / yes / yes ||$401|
|Pentium 4 640 ||3.2GHz ||2MB ||yes / yes / yes ||$273|
|Pentium 4 630 ||3.0GHz ||2MB ||yes / yes / yes ||$224|
Protection from worms and trojans
The new Intel chips include technology designed to prevent dangerous code -- contained in email attachments, for example -- being executed via buffer overflows. This is the eXecute Disable (XD) bit, which makes its first appearance on a mainstream Intel desktop CPU: until now, Intel has only implemented the XD bit in the server-class Itanium processor, and a limited number of Pentium 4 variants (those with the suffix 'J'). The Athlon 64, Athlon 64 FX and Opteron processors also offer this feature, which AMD calls No eXecute (NX).
Windows XP has supported the XD/NX feature since Service Pack 2 (Microsoft calls it 'Data Execution Prevention', or DEP). With DEP, the operating system provides a measure of inbuilt protection from worms and trojans, reducing the need to constantly install new security patches. Microsoft estimates that memory overflows lead to 50 per cent of all virus attacks; systems with XD or NX chips are therefore better protected against such attacks than those lacking this feature.
Improved power management
The new 6xx-series Pentium 4 processors benefit from improved power management in the shape of Enhanced Intel Speedstep Technology (EIST). The addition of this feature, along with improvements in the fabrication process, results in lower power consumption.
EIST provides similar functionality to AMD's Cool 'n Quiet technology in its Athlon 64 processors. If the system is in Idle mode, the voltage and clock frequency are reduced. As the CPU load increases, the voltage and clock frequency are gradually stepped up. The power consumption figures below demonstrate the effect of this technology. The Pentium 4 560 system without EIST (Intel D925XCV motherboard, 2 x 512MB Micron DDR2 RAM, MSI GeForce 6600GT, Maxtor Maxline III 250GB) uses 153 Watts in Idle mode, and this is reduced by 40 Watts when the EIST-equipped Pentium 4 660 is used instead. That's all the more impressive when you consider that Intel's new chip contains more transistors thanks to its 2MB of Level 2 cache. Compared to AMD's Athlon 64 4000+ (Asus A8N-SLI Deluxe), the Pentium 4 600 has closed the power consumption gap significantly, and is now only ten per cent behind.
|Power consumption: Pentium 4 versus Athlon 64|
|CPU ||Pentium 4 660 |
|Athlon 64 4000+ ||Pentium 4 560|
|Maximum* ||230W ||203W ||267W|
|Minimum ||113W ||103W ||153W|
* Maximum power consumption is measured using the CPU Stability Test 6.0 utility. If there is additional load on the graphics card, the system's maximum power consumption will be higher still. Note: the power consumption figures refer to complete systems with otherwise identical specifications.
To evaluate the performance of Intel's top-end 6xx-series Pentium 4 (the 660 model running at 3.6GHz), we have compared it to AMD's Athlon 64 4000+. Both processors cost just over $600.
The Pentium 4 test system is built around Intel's 925XCV motherboard, which is equipped with 1GB of 533MHz DDR2 RAM. For the Athlon 64 system, we used the Asus A8N-SLI motherboard, again with 1GB of (DDR400) RAM. The other specifications are identical, comprising the MSI GeForce 6600GT graphics card and Maxtor's Maxline III 250GB SATA hard drive.
Benchmarking is based primarily around current programs that might be used in a typical digital home PC. Available 64-bit applications are also used to test performance. Some tests are run in parallel, in order to evaluate the processors' multitasking ability. Last year's top 3D games are also represented in the shape of Doom3, Far Cry and Half Life 2.
Most tests were carried out under Windows XP Professional x64 Edition RC2 (Build 1433). However, the production of an HTML picture gallery with Jalbum was done under 32-bit Windows: this application requires a Java Virtual Machine, which is not yet available for the 64-bit platform. We contacted numerous vendors about 64-bit Windows applications, with limited success. We do have some 64-bit applications in this performance test, but most are still beta versions.
|Povray 3.6 (32- & 64-bit) ||raytracing|
|Panorama Factory (32- & 64-bit) ||panoramic photo production|
|FLAC 1.2 (32-bit), Monkey's Audio 3.99 (32-bit) ||lossless compression/decompression of audio data (APE and FLAC)|
|Nero Recode (32-bit) ||DVD re-quantising/transcoding|
|TMPG Express 3.0 (32-bit) ||video compression|
|Jalbum 5.1 (32-bit) ||HTML picture gallery production|
|Blobby Dancer Demo (32- & 64-bit) ||3D demo|
|Doom3, Far Cry, Half Life 2 ||3D games|
|Media Player 10 ||HDTV video (1080i) playback|
Benchmarks: audio data, multitasking
Lossless compression of audio data is mainly of interest to hi-fi enthusiasts, who require the highest quality audio streams. This form of audio compression, which is becoming ever more popular in file sharing networks, can be handled by two programs: Monkey's Audio and FLAC.
Appropriate plug-ins are available for media players such as Winamp. APE files are provided with Monkey's Audio 3.99. FLAC files are compressed using the open source FLAC 1.1.1a utility. This program also has some optimisations for the Pentium 4 platform.
For the benchmark test, a compressed audio file (APE: Pink Floyd / Dark Side of the Moon, FLAC: J.S. Bach / Complete Cantatas) is first decompressed and then compressed again.
With Monkey's Audio 3.99, the Athlon 64 4000+ has the edge, and the picture looks similar with FLAC. However, the Pentium 4 profits from the optimised FLAC version 1.1.2, which is based on the Intel Math Library. This shortens the time for decompressing from 32.8 seconds to 23.2 seconds. When it comes to compression, however, the Athlon 64 takes 32 seconds less than the Pentium 4. In this mode, FLAC obviously has no Intel optimisation.
|Monkey's Audio 3.99|
|Pink Floyd / Dark Side of the Moon ||Pentium 4 660 ||Athlon 64 4000+|
|Decompress (seconds) ||60.1 ||52.8|
|Compress (seconds) ||47.2 ||42.2|
|J. S. Bach / Complete Cantatas ||Pentium 4 660 ||Athlon 64 4000+|
|Decompress (seconds) ||32.8 ||32.1|
|Compress (seconds) ||143 ||115|
|FLAC 1.1.2 Optimised (Intel Math Library)|
|J. S. Bach / Complete Cantatas ||Pentium 4 660 ||Athlon 64 4000+|
|Decompress (seconds) ||23.2 ||30.9|
|Compress (seconds) ||143 ||115|
When the two audio files are processed separately, the compression results favour the Athlon 64, which takes a total of 157.2 seconds (APE: 42.2s, FLAC: 115s) compared to the Pentium 4's 190.2 seconds (APE: 47.2s, FLAC: 143s). However, things look different if the compression takes place simultaneously. Then the Pentium 4 benefits from its HyperThreading (HT) functionality and overhauls the Athlon 64. Intel's chip takes 163 seconds with simultaneous processing compared to 190 seconds with separate processing. For the Athlon 64, the pattern is reversed: 157.2 seconds with separate processing extends to 170 seconds with simultaneous processing.
|Monkey's Audio / FLAC ||Pentium 4 660 ||Athlon 64 4000+|
|Monkey's Audio compression (seconds) ||74 ||99|
|Total time including FLAC compression (seconds) ||163 ||170|
Benchmarks: video data
When it comes to video compression, the Pentium 4 660 is ahead of its AMD rival. Most applications in this area offer HT support, so the advantage of using the Intel architecture is hardly surprising.
TMPEG Enc Xpress 3.0 supports the Pentium 4's SSE3 instructions, which have yet to be incorporated into the Athlon 64 (this is slated for the next revision). Compression of the SVCD-format video file is therefore faster with the Pentium 4.
With Nero Recode you can copy a non-copy-protected 8.7GB DVD onto standard 4.7GB DVD-R media. To do this, the DVD data is compressed, a process that takes the Pentium 4 660 15:05 minutes in high-quality mode. The Athlon 64 4000+ takes over 20 minutes to complete the task.
|Video ||Pentium 4 660 ||Athlon 64 4000+|
|Nero Recode (minutes) ||15:05 ||20:12|
|TMPG Xpress 3.0 (minutes) ||03:44 ||04:02|
In the gaming tests, which were carried out at 1,280 by 1,024 resolution, the Athlon 64 is the better performer overall.
Nvidia's Blobby Dancer demo, which is available in 64-bit as well as standard 32-bit versions, provides an interesting result. The 64-bit version runs about 20 per cent faster under both processors.
Also interesting are the results from Quake II. This very popular first-person shooter from the 90s has gone open source and is available as a .NET version. There is a beta version of Microsoft .NET Framework 2.0 for Windows XP Professional x64 Edition RC2, which allows this game to be played under the 64-bit OS. The results show two things. First, the Pentium 4 delivers better performance with the .NET version than the Athlon 64 (a similar pattern is observed with the normal version of the game). Second, in contrast to the Blobby Dancer demo, Quake II runs slower under the 64-bit OS – so 64-bit programs are not always faster than their 32-bit counterparts.
|Game ||Pentium 4 660 ||Athlon 64 4000+|
|Far Cry 1.3 (fps) ||43.1 ||44.3|
|Doom3 (fps) ||60.1 ||62.1|
|Half Life 2 (fps) ||39.4 ||40.1|
|Blobby Dancer 64-bit (fps) ||42.4 ||42.3|
|Blobby Dancer 32-bit (fps) ||34.8 ||35.0|
|Quake II .NET 64-bit (fps) ||28.6 ||23.1|
|Quake II .NET 32-bit (fps) ||30.2 ||27.8|
64-bit benchmarks: photo processing, raytracing
In the final benchmark section we examine 64-bit applications alongside the 32-bit Jalbum 5.1 program. Panorama Factory is a popular tool for producing panoramic photos and, like the open source raytracing program Povray, is available in both 32-bit and 64-bit versions.
To produce an HTML picture gallery with Jalbum, we processed 151 images. As well as applying special effects, the pictures were converted to a different resolution. The Pentium 4 and Athlon 64 took almost exactly the same time to complete this process.
|HTML picture gallery ||Pentium 4 660 ||Athlon 64 4000+|
|Jalbum 5.1 (151 images) ||2:11 min ||2:10 min|
The Povray raytracing test runs 27 percent faster in 64-bit mode on the Athlon 64 than in 32-bit mode. With the Pentium 4, surprisingly, the pattern is the exact opposite: here, the 32-bit version is significantly faster. However, this program still is in the beta stage, and we should be careful not to draw premature conclusions.
|Povray ||Pentium 4 660 ||Athlon 64 4000+|
|Povray-Benchmark 64-bit (PPS) ||83 ||116|
|Povray-Benchmark 32-bit (PPS) ||103 ||91|
The Panorama Factory test demonstrates the advantages of the 64-bit architecture, with both processors showing improvements over the 32-bit version. The Pentium 4 speeds up the panorama production process by around 25 per cent, while the Athlon 64 acceleration is nearly 30 per cent.
|Panorama Factory ||Pentium 4 660 ||Athlon 64 4000+|
|Panorama Factory 64-bit (seconds) ||59.5 ||55.2|
|Panorama Factory 32-bit (seconds) ||80.1 ||78.5|
With the launch of the Pentium 4 6xx series, Intel now supports 64-bit computing on the desktop. However, this won't be a huge deal for the majority of users right away. Windows XP Professional x64 Edition will ship within a few weeks, but the lack of key device drivers -- for example for digital TV cards -- will deter many from upgrading for a while yet. Also, the number of available 64-bit Windows applications can be counted on the fingers of one hand; again, this will not be enough to tempt many upgraders.
So it's just as well that the new Intel processors have more to offer than 64-bit support. Improved power management and protection from memory overflow (via the XD bit) will prove valuable. Enhanced Intel Speedstep Technology (EIST) allows for the dynamic adjustment of voltage and clock speed, which reduces average power consumption significantly. Even when playing a HDTV video, the Pentium 4 660 remains at its lowest clock speed of 2.8GHz, with reduced voltage. As a result, the cooling system can be simpler than it was when the Prescott CPU was introduced over a year ago. Very quiet systems can now be built around high-performance Pentium 4 processors -- a particularly important factor if the PC is to function as the basis for the 'digital home' and take its place in the living room.
As far as performance is concerned, the 3.6GHz Pentium 4 660 can hold its own against its main competitor, the Athlon 64 4000+. The Intel chip performs particularly well if several tasks are running at the same time; under these circumstances, the Pentium 4 can outpace its AMD rival even if the latter is quicker at performing the tasks on their own. Thanks to HyperThreading (HT), the Pentium 4 distributes processing tasks across two virtual cores, resulting in more efficient utilisation of CPU resources. Such scenarios are found ever more frequently in the real world. For example, no-one should venture onto the Internet without firewall, antivirus and anti-spyware protection. These services are constantly active and need appropriate resources. Likewise, operations such as data encryption or hard disk defragmentation can load the processor, while the user compresses streaming video or audio data. Under such usage patterns, the advantage of HT is particularly apparent.
With the 6xx-series Pentium 4, Intel has ensured that important functionality such as protection from memory overflow, power management and 64-bit support on the desktop is no longer an AMD domain. And with HT support and SSE3 instructions, the new Intel processors offer additional benefits. One change will annoy Intel, however: on the installation CD for Windows XP Professional x64 Edition, the most important operating system files are no longer in the 'i386' folder; Intel systems must load the installation files now from the 'AMD64' folder. Although that might hurt the industry leader, Intel may draw comfort from the fact that it has already sold more 64-bit chips than the inventor of the x86-64 architecture, AMD.
|CPU ||Pentium 4 660 ||Athlon 64 4000+|
|64-bit support ||yes, EM64T ||yes, AMD64|
|Power management ||yes, EIST ||yes, Cool 'n Quiet|
|Memory overflow protection ||yes, XD bit ||yes, NX bit|
|HyperThreading ||yes ||no|
|SSE3 ||yes ||no|