AMD Athlon 64 FX-53 CPU Review

AMD has finally managed to release a CPU with 2.4GHz clock frequency. Due to the new CG core revision we are glad to welcome Athlon 64 FX-53. Read all you wanted to know about the newcomer: a lot of benchmarks, overclocking results and a few exciting tests of the Windows XP 64-Bit Edition for 64-Bit Extended Systems.

by Ilya Gavrichenkov
03/18/2004 | 11:00 AM

Not so long ago we wrote about the whole bunch of new processors announced by Intel. Among the newcomers there were CPUs based on the new 90nm Prescott core, Northwood based CPU sped up to 3.4GHz and the second representative of the Pentium 4 Extreme Edition family working at 3.4GHz. This particular processor from Intel, Pentium 4 XE 3.4GHz selling for around $1000 and targeted for the extreme gaming market deserved the title of the fastest desktop processor.

 

Of course, a few allowances have been made before this processor actually got the title, because it failed to outperform AMD Athlon 64 FX-51 in the whole number of applications (mostly games). But nevertheless, AMD’s positions in the high-performance desktop market were shaken a little bit. AMD was surely unhappy with this state of things that is why they decided to strike back and to prove once and for all their superiority in high-performance CPU market. Today, on March 18, the first day of CeBIT international computer show, AMD is proud to announce their high-performance enthusiast solution – AMD Athlon FX-53.

Closer Look

So, the new Athlon 64 FX-53 processor continues the AMD processor family for gaming enthusiasts, which started with the Athlon 64 FX-51 solution. The new processor will replace the predecessor. Starting today, Athlon 64 FX-51 stopped shipping. So, the price of the new Athlon 64 FX-53 will be the same, i.e. $733, which is much lower than the pricing of the alternative Intel’s solutions – Pentium 4 Extreme Edition CPUs.

There is no architectural difference between the current and the previous models of Athlon 64 FX. The new CPU should be used in the same mainboards equipped with a Socket 940, uses the same 800MHz HyperTransport bus and features the same dual-channel memory controller supporting Registered DDR SDRAM. For this reason we will not dwell on the architecture and peculiar features of the new CPU. All necessary info has already been reported to you in our AMD Athlon 64 FX-51 CPU Review.

I would like to particularly draw your attention to the distinguishing features of the new Athlon 64 F-53, different from the predecessor. First of all I would like to point out that this CPU acquired a higher model number due to its higher core clock frequency. While Athlon 64 FX-51 worked at 2.2GHz, the newcomer supports 2.4GHz. This is a very important thing because Athlon 64 FX-53 appears the first AMD processor, which managed to overcome the 2.2GHz frequency, which has long been not only technological but also a psychological barrier for AMD in a certain way. In fact, we have finally got evidence proving that the use of SOI (Silicon-on-Insulator) technology for Athlon 64 production finally started bearing some fruits. It is totally due to this particular technology that AMD managed to take off the 2.2GHz core clock, which was the maximum for the previous Athlon XP processor family manufactured with the same 0.13micron process but without SOI technology involved.

I have to say however that AMD had to work real hard to conquer 2.4GHz frequency. In order to produce CPUs with these advanced characteristics they had to develop a new Athlon 64 die revision, which would be more scalable in terms of frequency. This revision is marked as CG and is recognized as such in Athlon 64 FX-53 by identification utilities.

Besides better frequency scalability, the new CG revision should also boast a few other innovations. Among them are better compatibility with various memory modules and more flexible Cool’n’Quiet technology. However, we will not see these improvements in the new Athlon 64 FX-53. Since this CPU is designed for Socket 940 mainboards, Cool’n’Quiet technology is disabled and the memory controller works just the same way as that of the Athlon FX-51 requiring registered memory modules. In order to really feel the advantages of the new die revision, we will have to wait for the Socket 754 Athlon 64 3700+ processor scheduled for April. Moreover, the transition of Athlon 64 FX-53 to CG core revision doesn’t change anything in the thermal and electric characteristics of the CPU despite the increase core clock frequency.

In addition to what has just been said, I would only like to remind you that Athlon 64 FX-53 is the last desktop processor from AMD for Socket 940 form-factor. In May the entire Athlon 64 FX family will move to Socket 939 and will acquire faster HyperTransport bus with 1GHz frequency and a dual-channel memory controller, which will not demand registered memory modules. So Socket 940 will only be used by Opteron processor family intended for servers and workstations.

Overclocking

Since the major key feature of the new Athlon 64 FX-53 CG core revision is higher frequency potential, we thought it would be interesting to check how well this processor can overclock. As you remember, the CPUs based on the previous C0 die revision didn’t go beyond 2.4GHz with air cooling solutions involved. In other words, the new Athlon 64 FX-53 couldn’t be based on the old core revision. The new core should definitely show better absolute results during overclocking. The only question is how impressive these results will be. According to AMD’s current plans, the top core frequency for 0.13micron Athlon 64 processors is 2.4GHz. AMD is going to release faster Athlon 64 and Athlon 64 FX processor models with the prospective 90nm San Diego and Winchester cores. However, we haven’t yet tried the new CG core in practice. Maybe AMD’s pragmatic expectations will not prove true and the new CG core will be able to work at higher frequencies. All in all, an experiment is necessary.

Before we pass over to the results obtained during AMD Athlon 64 FX-53 overclocking, we have to stress that since AMD positions this CPU as an enthusiastic solution, it is free from all artificial limitations that could hinder successful overclocking. Unlike Athlon 64, the clock frequency multiplier of the new Athlon 64 FX has no upper limit. That is why those hardware enthusiasts who decide to invest in the new Athlon 64 FX get an opportunity to perform easy overclocking of their CPU without changing the working frequencies for the memory, HyperTransport, AGP or PCI.

So, we overclocked our AMD Athlon 64 FX-53 on ASUS SK8V mainboard. For cooling purposes we used Thermaltake Silent Boost K8 (A1838) cooler. To increase the overclocking efficiency, we raised the Vcore by 10% from the nominal 1.5V to 1.65V.

During overclocking we first of all increased the processor clock frequency multiplier. It turned out that our processor can work fine with the maximum multiplier equal to 13x. Setting higher multipliers led to the system’s inability too boot up. When we reached 2.6GHz obtained as 13x200MHz, the system worked absolutely fine. For even better result we decided to slightly increase the FSB frequency. However, this is where we faced the first problems pretty soon: the top frequency of the clock generator when our Athlon 64 FX-53 worked stably equaled only 202MHz.

As we see, the maximum frequency we managed to reach during Athlon 64 FX-53 overclocking was 2632MHz. in this mode the CPU successfully passed all stability tests. In other words we managed to overclock the CPU just a tiny bit beyond 2.6GHz. So, it is no secret any more that the new CG processor revision will not suit for mass products with higher core clock. Since Athlon 64 FX-53 is the top processor model on the CG core, its frequency doesn’t grow too much. So, overclockers will be able to squeeze 10% extra frequency out of the new AMD Athlon 64 FX-53 at the most without any extreme cooling involved.

Testbed and Methods

The major goal of this test session was to find out how fast the new AMD Athlon 64 FX-53 actually is compared with the competing products. Athlon 64 Fx-53 is positioned as a direct competitor to Pentium 4 Extreme Edition 3.4GHz. However, in this review we decided to compare the new AMD CPU with the top-end processor models from all Pentium 4 families: on Northwood, Gallatin and Prescott cores. I would like to point out that even though Pentium 4 with 3.4GHz core clock frequency and Prescott core is not yet available in stores, we will have it also participating in our tests. The thing is that Intel is about to start shipping these processors in the next few days and we were lucky to get our hands on one of these pieces. Also we will include the results for the outdating Athlon 64 FX-51 CPU and the top model of the Athlon 64 family with 3400+ rating.

As a result, our testbeds got configured in the following way:

Note:

Performance

3D Games

Pentium 4 XE and Athlon 64 FX are positioned as solutions for extreme gamers. That is why we decided to start our test session with the gaming tests.

Due to the higher clock frequency, which is now equal to 2.4GHz, Athlon 64 FX-53 processor managed to outperform Pentium 4 XE 3.4GHz in popular 3DMark gaming benchmarks.

Unfortunately, Athlon 64 FX-53 processor failed to outperform Intel Pentium 4 Extreme Edition 3.4GHz in Aquamark3 test, which knows to use Hyper-Threading technology very efficiently.

The gamers will definitely be happy with the gaming performance of AMD Athlon 64 FX-53. Almost everywhere except only Quake3, the new AMD processor shows excellent results even against the background of the 1000-dollar competitor from Intel.

By the way, we also added the results shown by top AMD and Intel processors in the new Unreal Tournament 2004 game. Here they are:

AMD fans will definitely be pleased. Just like in the previous version of Unreal Tournament, Athlon 64 and Athlon 64 FX processors are beyond any competition

Office and Digital Content Creation Applications

As usual we are beginning this section with the Winstone tests:

Athlon 64 FX-53 proves its leadership once again and outperforms Pentium 4 XE 3.4GHz, which is the today’s fastest Intel CPU.

Multi-Threaded Tasks

The new Business Winstone 2004 test package allows estimating the performance of the tested systems under multi-threaded workload created by a few simultaneously working applications. Since the participating Intel processors support Hyper-Threading technology, which improved the CPU performance during several streams processing, we couldn’t disregard this opportunity.

In this benchmark we use the simple file copy operation as a background workload. At the same time, we measure the systems performance in Microsoft Outlook and Internet Explorer. As we see, this applications combo is no serious workload even for AMD processors, which do not support Hyper-Threading technology.

Now let’s check how the CPUs will cope with a more serious multi-threaded workload:

In this case we have a more complicated task running in the background: it is the working WinZIP archiving utility. At the same time the benchmark emulates the work in Word and Excel. Strange as it might seem, but AMD processors, which do not support Hyper-Threading managed to show even better results here.

This is the hardest test where we have Norton AntiVirus running in the background and the whole bunch of office applications including Microsoft Excel, Microsoft Project, Microsoft Access, Microsoft PowerPoint, Microsoft FrontPage and WinZip. And this is exactly the case when the absence of any analogues to Intel’s Hyper-Threading technology results into a performance drop by AMD CPUs. Even Athlon 64 FX-53 is considerably behind Intel CPUs here.

SYSmark 2004

In this review we decided to return to SYSmark benchmarks. As you remember, we refused to use the last SYSmark 2002 version, because we discovered that it was optimized for NetBurst architecture. It happened because Intel was one of the developers of this benchmarking set and AMD was not. The new recently released version of this test package, SYSmark 2004, was developed by the new BAPCO committee with all parties participating, including Intel and AMD.

This list of companies working on the SYSmark 2004 test makes us absolutely sure about its objectiveness.

The ideology behind this test implies measuring the time it takes the tested platforms to react when solving various practical tasks. We are going to dwell on the type of tasks and the results obtained from the top-end AMD and Intel processors there.

Benchmark description: 3ds max 5.1 set renders the image into a .BMP file, while the user is busy working on web-pages in Dreamweaver MX. Then the user renders 3D animation in vector graphics format.

The major contribution to the overall result in this test is the performance of the testing participants in 3ds max (95.5%).

Here we model the user’s work in Premiere 6.5, where he creates a video movie from a few other movies in raw-format and a few separate audio tracks. While waiting for the operation to be completed, the user modifies an image in Photoshop 7.01 and saves it to the hard disk drive. When the video is created, the user edits it and adds some special effects in After Effects 5.5.

The applications contribute to the overall performance index in the following way: 43.3% - Adobe Photoshop 7.01, 39.1% - Premiere 6.5, 17.6% - AfterEffects 5.5.

The user extracts files from a zip-archive and at the same time uses Flash MX to open an exported 3D vector graphics file. The user modifies the file by inserting other images and optimizes it for faster animation. The end-movie with all special effects applied is compressed by Windows media Encoder 9 for further Internet translation. Then the created web-site is put together in Dreamweaver MX, while the system is being checked for viruses with VirusScan 7.0.

The major influence on the final result is made by Windows media Encoder 9 (56%), VirusScan 7.0 (30.4%) and Flash MX (9.8%).

Here the test models the user’s actions when he received a message in Outlook 2002 with a set of zip-archives. While the files are being scanned for viruses with VirusScan 7.0, the user looks through his e-mails and makes a few notes in the calendar. Then the user browses through the corporate web-site and a few other documents with the help of Internet Explorer 6.0.

The biggest contribution to the overall performance index results from VirusScan 7.0 (80.8%) and Outlook 2002 (15.4%).

In this benchmark the hypothetical user edits a text in Word 2002 and uses Dragon NaturallySpeaking 6 to convert an audio-file into text document. The end-document is converted into a .PDF-file with Acrobat 5.0.5. Then the user creates a PowerPoint presentation basing on the prepared document in Power Point 2002.

The percentage of applications contribution to the end index looks as follows: Word 2002 - 10.4%, PowerPoint 2002 - 16.7%, Dragon NaturallySpeaking 6.0 - 34.6% and Acrobat 5.0.5 - 38.4%.

The user opens a database in Access 2002 and created a number of requests. The documents are packed with WinZIP 8.1 archiving utility. The results of the requests are exported into Excel 2002 and then are used for a diagram.

The major part of the performance index is determined by Excel 2002 - 76.6% and Access 2002 - 19.8%.

And now take a look at the usual performance index diagrams:

As we see, Intel’s solutions perform faster in this benchmark set. However, this is not a surprise. If you take a closer look at the benchmarks descriptions you will see that Pentium 4 processors had to be faster due to their efficient Hyper-Threading technology. Actual workload modeled in SYSmark 2004 implies that the user will not keep his hands in pockets while waiting for one application to finish the task and will shift to some other type of work in the meanwhile, which is actually quite a common thing.

Archiving Utilities

Intel processors manage to defeat AMD Athlon 64 FX-53 only in one case: during data compression in 7zip utility. No wonder as this archiving tool supports multi-threading. However, as you can see Hyper-Threading doesn’t help Intel during data decompression.

Audio and Video Encoding

The diagrams below deliver the results in seconds. That is why the lower is the value, the faster is the CPU.

Audio and video streams encoding is one of the trumps of Intel’s NetBurst architecture. AMD CPUs have always been behind their Intel rivals in applications of this kind. However, higher clock frequency of the new Athlon 64 FX-53 ensures a significant performance gain so that AMD newcomer manages to almost catch up with Pentium 4 3.4E on Prescott core, which is sometimes slower than the predecessors on Northwood core working at the same clock frequency because of its longer pipeline.

Scientific Calculations

No comments are necessary here. We all have known for a long time now that AMD processors have always been faster with all sorts of scientific calculations.

We also used a new beta-version of the ScienceMark 2.0 test package to measure the performance of our CPUs. This test is devoted to math1ematical modeling tasks and the systems performance in them. Note that the diagrams below show the results in second, that is why the lower value stands for faster performance.

The increase in AMD Athlon 64 FX-53 core clock to 2.4GHz allows this CPU to show the maximum performance possible in all tests of the ScienceMark 2.0 package.

3ds max and Cinema4D

To test the processors performance during final rendering we used the new sixth version of 3ds max package. Moreover, we measured not only a single frame rendering speed, but also the animation rendering speed with the simultaneous saving of the avi-file. The two diagrams below show the performance of the CPUs in second, therefore the lower is the value, the faster the CPU actually is.

Note that the higher working frequency of the new Athlon 64 FX-53 helps this CPU a lot in 3ds max tests. Although the absence of Hyper-Threading support by Athlon 64 processors, doesn’t allow us to call them suitable for systems used for final rendering needs. The results obtained in Cinema4D prove this point once again.

However, the processing of OpenGL context in professional applications is an indisputable success of the new Athlon 64 FX-53, as it managed to even outperform Pentium 4 XE 3.4GHz.

Adobe Photoshop

Adobe Photoshop CS 8.0 is a very popular graphics editor, which is very widely used for 2D graphics editing. That is why we decided to pay special attention to the performance of our processors in this test. For our needs we used a slightly modified PSBench 7 with a 100MB image to be processed. The diagram below contains all results in seconds that is why the lower value denotes higher performance.

Here are more detailed results for your reference, which show the performance of our CPUs with different Photoshop CS 8.0 filters. The table also contains times in seconds:

 

Athlon 64 3400+

Athlon 64 FX-51

Athlon 64 FX-53

Pentium 4 3.4

Pentium 4 3.4E

Pentium 4 XE 3.4

Rotate 90

0.2

0.2

0.2

0.2

0.1

0.2

Rotate 9

3.8

3.7

3.4

3.8

3

3.8

Rotate .9

3.5

3.4

3.2

3.6

2.9

3.6

Gaussian Blur 1

1.1

1.1

1

1.3

1

1.2

Gaussian Blur 3.7

3.5

3.5

3.3

2.7

2.3

2.7

Gaussian Blur 85

3.9

3.8

3.6

2.9

3.2

2.8

Unsharp 50/1/0

1.9

1.8

1.7

2

1.5

2

Unsharp 50/3.7/0

4.4

4.3

4

4

2.8

4

Unsharp 50/10/5

4.4

4.3

4

3.2

2.9

3.2

Despeckle

5.3

5.3

4.8

3.1

3.5

3

RGB-CMYK

11.1

11

10.1

6.9

7

6.9

Reduce Size 60%

1

1

0.9

0.7

0.7

0.7

Lens Flare

7.9

7.8

7.3

6.1

6.2

6.1

Color Halftone

10.8

10.8

9.8

14.6

13.5

13.8

NTSC Colors

3.1

3.1

2.8

3.2

3.4

3.1

Accented Edges

20.4

20.5

19.1

22.2

27.9

21.6

Pointillize

41

40.6

37.5

32.8

35

32.1

Water Color

43.5

43.4

40

43

57.6

42.9

Polar Coordinates

14.4

14.4

13.3

10.7

11.4

10.4

Radial Blur

71

72.7

68.7

59.8

48.7

60.9

Lighting Effects

4

4

3.7

3.5

3.4

3.5

Software Development

Besides the traditional set of benchmarks we decided to include one more test, which we have successfully introduced to you in our previous CPU review. Namely, we checked how fast our testing participants would compile the projects in Visual C++ .NET, which is a very popular software development interface. For our measurements we used the source code of the Emule client with the added source code for a few libraries necessary at the compilation stage: crypto51, CxImage, zlibstat. We measured the time in two compilation modes: “Debug” – the version with the debugging info included into the final code, and “Release” – creation of the final product with optimised processing speed and code size. Both diagrams below show the results in seconds, so the smaller value stands for better result.

If you are a software developer or a scientist I would strongly recommend considering new AMD processors as one of the best choices for your needs.

Windows XP 64-Bit Edition for 64-Bit Extended Systems Preview

One of the major topics attracting the public attention when they touch upon AMD Athlon 64 FX and Athlon 64 processors is the discussion of their 64bit capabilities. Unfortunately, there is not much to talk about yet. There are no public operation systems supporting 64bit AMD64 extensions that is why the biggest part of our today’s article will be devoted to the performance of Athlon 64 FX-53 in 32bit Windows XP operation system.

However, the situation is not as hopeless as it might seem at first glance. AMD has already sold quite a bit of processors supporting AMD64 technology. Moreover, Intel has also announced their long-term intention to introduce AMD64-compatible 64bit extensions support in its upcoming desktop CPUs in the far-away future. That is why it is more than evident that AMD64 technology will become a success one day. Sooner or later 64bit operation systems and the whole lot of 64bit software will come to the market. Especially, since software developers have no alternatives left: if the program will require more than 2GB of memory then the only way to solve the problem is to use 64bit processor mode.

But this is the future. Although there is some movement towards the use of AMD64 technology today already. The first remarkable event in this field took place in the beginning of February, when Microsoft Company offered everyone the opportunity to get acquainted with a preliminary version of their Windows XP operation system with AMD64 technology support. This OS is officially called Windows XP 64-Bit Edition for 64-Bit Extended Systems and is available for free download on the Microsoft web-site. Of course, we couldn’t leave out this fact and took a closer look at the new operation system.

I would like to stress that this is the second time already we get to work with a Windows version supporting AMD64 technologies. In our article devoted to AMD Athlon 64 FX-51 we have already provided some benchmarks results obtained in this OS. However, at that time we had an earlier beta-version of Windows XP 64-Bit Edition for 64-Bit Extended Systems at our disposal. Although at that time already it worked flawlessly, without any evident bugs. The public demo version of this OS is marked with a newer date and works without any problems on Athlon 64 FX/Athlon 64 processors. However, this system is hardly suitable for every-day use. In fact, it can be useful only for software developers and testers who simply have to use a 64bit system. Regular users do not need to shift to Windows XP 64-Bit Edition for 64-Bit Extended Systems and sometimes simply cannot do it at all.

The first problem you face when installing Windows XP 64-Bit Edition for 64-Bit Extended Systems is the lack of necessary drivers. Very few hardware developers offer drivers for their devices, which are compatible with the Windows versions supporting AMD64. Regular 32bit drivers cannot be used for 64bit operation system. For this reason, for instance, the owners of Windows XP 64-Bit Edition for 64-Bit Extended Systems cannot use ATI based graphics cards in their systems: the Canadian company doesn’t supply the graphics card drivers for this operation system yet. Also the owners of some network cards, RAID controllers and audio solutions may face similar problems. That is why if you decide to install Windows XP 64-Bit Edition for 64-Bit Extended Systems in your PC, make sure that you’ve got drivers for all the hardware devices you have in your system.

The second problem is the lack of appropriate software. Although, one of the strengths of AMD64 architecture is its backward compatibility with 32bit software. Therefore, this is no fatal problem at all. Most 32bit tools and utilities work fine with Windows XP 64-Bit Edition for 64-Bit Extended Systems due to the so-called WOW64 (Windows-on-Windows) emulator, which transfers the CPU from the native 64bit mode into the compatibility mode, so that the application can manipulate the CPU registers and address the memory in 32bit format. During the tests we carried out for various 32bit programs we revealed just a few issues. The first one had to do with the failing installation of Windows Media Encoder 9 with all the resulting consequences, while all other issues were connected with different artifacts popping up in 3D games every now and then. Of course, it was the graphics card driver that was responsible for the latter problems.

At the same time, the variety of software developed specifically for 64bit mode of AMD64 processors will keep increasing. Some programs are already available today, although there are very few of them, I should say. In particular, we know about 64bit versions of TweakUI Power Toys utility, Sun Java 2 platform and beta version of Computer Associates eTrust Antivirus.

Windows XP 64-Bit Edition for 64-Bit Extended Systems operation system looks very much like Windows 2003 and not like Windows XP. It looks as if they used exactly the Windows 2003 code when they were working on the Windows XP version with AMD64 support. The OS includes a set of 64bit DirectX 9.0b libraries and a set of standard programs ported for AMD64 architecture. It is a pretty curious fact that Windows XP 64-Bit Edition for 64-Bit Extended Systems includes not only a 64bit version of Internet Explorer 6.0, but also a classical 32bit version of this program. Also the 64bit Windows XP package will include a number of critical updates absent in the standard Windows XP SP1 package for 32bit systems.

I would also like to say a few words about the test system, which we used for Windows XP 64-Bit Edition for 64-Bit Extended Systems testing. The thing is that the standard platform configuration didn’t suit our goals this time, as we failed to find corresponding drivers for a few components of our system. That is why we had to replace an ATI based graphics card with an NVIDIA based one. As a result, Athlon 64 FX-53 was tested on the following system:

In order to finish our story about the preliminary version of Windows XP 64-Bit Edition for 64-Bit Extended Systems in a logical manner, we decided to test AMD Athlon 64 FX-53 processor in this operation system during the processing of 32bit and 64bit applications.

At first I would like to say a few words about the 64bit performance. Unfortunately, our opportunities were pretty limited because there were hardly any applications compiled for native 64bit mode. The only available benchmark for Win64 is a 64bit version of SiSoft Sandra 2004 SP1. This test package includes short synthetic benchmarks for the CPU and system memory. Due to these tests we get at least some idea of the system performance in Windows XP 64-Bit Edition for 64-Bit Extended Systems. At the same time I would like to say that SiSoft Sandra 2004 SP1 also exists for Win32. So, we get an excellent opportunity to compare the results of this test during 64bit code processing in 64-bit Windows XP with those during 32bit code processing in 64-bit and 32-bit Windows XP versions.

SiSoft Sandra 2004 SP1

Win64
64-bit exe

Win64
32-bit exe

Win32
32-bit exe

CPU Arithmetic Benchmark

Dhrystone ALU

10213

10441

10387

Whetstone FPU

4277

3797

3790

Whetstone SSE2

4547

4960

4854

CPU Multimedia Benchmark

Integer MMX/SSE

17729

22913

22888

Floating Point SSE2

25451

24631

24611

Memory Bandwidth Benchmark

Int Buffered

6051

5633

5668

Float Buffered

6006

5613

5629

The results are pretty ambiguous. In some tests the shift to 64bit mode leads to a notable performance gain, while in some cases the performance drops significantly. However, first of all I would like to say that 32bit benchmarks from the Sandra package were run equally fast in 32bit and 64bit operation system. This way, I can conclude that WOW64 emulation doesn’t cause any delays of the 32bit code processing.

Returning to the results in 64bit SiSoft Sandra 2004 in Win64 I would like to point out that the shift to 64bit mode allows AMD64 processors to speed up the work with the system memory quite tangibly. However, this is almost the only case (besides Whetstone FPU benchmark) when 64bit technology has a positive effect on the performance, according to SiSoft Sandra 2004. In all other cases the performance in 64bit mode is either the same or lower. Unfortunately, we can’t state with all certainty why the 64bit performance is considerably lower than 32bit performance in Multimedia Integer MMX/SSE and Whetstone SSE2 benchmarks. The most probable explanation here is the poor optimization of the test algorithms for AMD64 technology. All exiting compilers generating the code for Windows XP 64-Bit Edition for 64-Bit Extended Systems are only available as alpha- or beta-versions, that is why no one can guarantee their efficient work today.

At the same time, if the code of the 64bit SiSoft Sandra version is not very well optimized for AMD64 architecture (for instance, it uses nearly none of the additional general purpose registers), we can also face a noticeable performance drop but for a different reason. In 64-bit mode Athlon 64 processors perform certain instructions (namely integer multiplication and division) somewhat slower than in 32-bit mode. However, this explanation suits only for Dhrystone ALU benchmark, because Athlon 64 performs all SSE2 instructions equally fast in 32-bit and 64-bit modes.

Unfortunately, besides SiSoft Sandra 2004, there are no other applications at our disposal today supporting 64-bit mode and offering some benchmarking opportunities. That is why let’s pass over to a more indepth investigation of the 32bit applications performance in Windows XP 64-Bit Edition for 64-Bit Extended Systems.

Since we are now talking only about a preliminary version of this OS, it doesn’t make much sense to carry out very detailed test session with the whole bunch of applications involved. For our preview we will simply run PCMark 2004 test, which measures the performance of the most widely spread algorithms.

PCMark 2004

Win64

Win32

File Compression, MB/s

3.321481

3.327296

File Encryption, MB/s

37.08963

37.18693

File Decompression, MB/s

29.30771

29.32725

Image Processing, Mpixels/s

14.73462

14.75512

Virus Scanning, MB/s

2695.335

2755.923

Grammar Check, KB/s

3.490216

3.697021

File Decryption, MB/s

74.32847

74.46515

Audio Conversion, KB/s

3166.141

3177.966

Web Page Rendering, pages/s

7.107715

6.114258

DivX Video Compression, fps

71.95704

71.63315

Physics Calculation and 3D, fps

197.4499

218.8646

Graphics Memory - 64 lines, fps

445.9108

2432.193

As we see, typical 32bit procedures loading the CPU and the memory show almost the same performance level in 64bit Windows version as they do in 32bit Windows. In other words, this is another proof to the fact that the use of WOW64 emulator doesn’t cause any performance drop in 32bit applications running under 64bit operation system. However, according to the obtained results there are two exceptions to this rule.

The first exception. Web Page Rendering test shows a much higher result working in 64-bit Windows version. This happens because the test measures web-pages rendering speed with the help of the Internet Explorer version installed in the system, and of course, in Windows XP 64-Bit Edition for 64-Bit Extended Systems this is a 64-bit version too. This way, we can clearly see that 64-bit Internet Explorer is tangibly faster than 32bit one, and Athlon 64 processors really do “speed up” web-surfing.

The second exception. It has to do with the last two benchmarks, which indicate a performance drop during their work in 64-bit operation system. Moreover, this performance drop is not so crucial in the 3D benchmark, while in 2D benchmark (the very last one) it is simply dramatic. In this case I would blame the graphics driver for such a failure of the 64-bit system. It looks as if the situation with the drivers for Windows XP 64-Bit Edition for 64-Bit Extended Systems is aggravated not only by the absence of these drivers. As we have just seen, the existing drivers are far from ideal as well. The best example here is Detonator from NVIDIA. However, we should be fair: ATI has no drivers at all, which would support processors with AMD64 technology in 64-bit operation systems.

To prove everything I have just said, here are the results of 32-bit games tests in Windows XP 64-Bit Edition for 64-Bit Extended Systems:

 

Win64

Win32

3DMark 2001 SE

11588

18075

3DMark03

2820

5273

Quake3 (four), High Quality, 1024x768x32

227.0

464.3

Unreal Tournament 2003 (dm-antalus), 1024x768x32

63.38

96.44

The numbers speak for themselves, I suppose. 32bit games run under Windows XP 64-Bit Edition for 64-Bit Extended Systems work much slower than in the regular Windows XP Professional. Moreover, there are some problems with the actual work of the gaming applications. We discovered very bad artifacts in a number of our test 3D applications, such as Aquamark3 or X2 – The Threat. All in all, NVIDIA still has a lot to do about the improvement of its 64-bit graphics drivers.

Summing up everything mentioned above, the verdict about Windows XP 64-Bit Edition for 64-Bit Extended Systems will not be praising. So far this system may be of interest only to the developers. Regular users will not be happy with it because of the availability matters as well as the drivers quality. However, it is very nice that the process of developing applications supporting AMD64 architecture has finally taken off. The official launch of Windows XP 64-Bit Edition for 64-Bit Extended Systems is scheduled for H2 2004, so hopefully the hardware guys will be able to eliminate the problems with their drivers by then, and the software developers will successfully port their programs for 64-bit operation system and 64-bit working mode.

Conclusion

As we see, the situation in the high-performance processors market has become very favorable to AMD lately. The launch of 64-bit processors allowed the company to make a successful breakthrough, so that now they can at least compete on equal terms with Intel. New high-performance processor, Athlon 64 FX-53, which we have reviewed today, is another proof to the point.

In most benchmarks Athlon 64 FX-53 working at 2.4GHz frequency outperforms all alternatives from Intel including even Pentium 4 extreme Edition 3.4GHz featuring 2MB of L3 cache memory. Of course, we have to stress that during audio and video data encoding as well as final rendering NetBurst architecture makes Intel solutions more efficient than Athlon 64. Moreover, Hyper-Threading technology, which is not available by AMD solutions pushes different Pentium 4 CPUs to the lead in a number of practical applications. However, the gaming performance of Athlon 64 FX-53 is beyond any competition, which is its best trump in front of dedicated gamers.

Nevertheless, even though Intel has already introduced its new processor generation on the Prescott core, AMD will have to stand one more blow from Intel in the near future: when Intel announces a new D0 Prescott core revision and shift to a new LGA775 processor socket, the clock frequencies of their CPUs will grow up immensely. Moreover, systems based on these new processors will have a few marketing advantages at their disposal, such as DDR II SDRAM and PCI Express support. Will AMD manage to retain its positions in this case? We will see, but so far AMD seems to be just fine :)