by Ilya Gavrichenkov
01/05/2004 | 10:05 PM
The beginning of the year 2004 was marked by the announcement of a new AMD Athlon 64 processor model. The new CPU is rated as 3400+ and differs from the predecessor with 3200+ performance rating by the higher working frequency, which has grown up to 2.2GHz. This way, the CPU family with AMD64 architecture positioned by AMD Company for the mass market now includes three processor models: 3000+, 3200+ and 3400+.
These three company’s offerings are priced very differently, that is why those of you who wish to acquire a Socket754 platform have something to choose from. In this article we are going to find out what the newcomer from AMD, Athlon 64 3400+, is capable of, if it will be able to become a worthy competitor to the top Intel processors, and what it actually looks against the background of the already existing AMD processors.
In fact, AMD Athlon 64 3400+ is absolutely identical to Athlon 64 3200+ with that only difference that it works at 200MHz higher core clock frequency, which is now equal to 2.2GHz. Besides that, there are no other differences. The newcomer, just like Athlon 64 3200+ and Athlon FX-51, features a 1MB L2 cache. This way the maximum working frequency of the Athlon 64 processor family has now reached that of Athlon 64 FX-51, so that now the only difference between Athlon 64 3400+ and Athlon 64 FX-51 is the memory controller: Athlon 64 FX-51 features a dual-channel memory controller requiring Registered DIMM modules, while Athlon 64 3400+ features a single-channel memory controller supporting regular memory modules. The table below will help you figure out all the distinguishing features of the today’s AMD processors based on Athlon 64 architecture:
Athlon 64 FX-51
Athlon 64 3400+
Athlon 64 3200+
Athlon 64 3000+
Typical heat dissipation (Cool’n’Quiet disabled)
Integrated memory controller
Supported memory types
Registered DDR400/ DDR333/ DDR266 SDRAM
DDR400/ DDR333/ DDR266 SDRAM
DDR400/ DDR333/ DDR266 SDRAM
DDR400/ DDR333/ DDR266 SDRAM
SIMD instructions support
I have to stress that Athlon 64 3400+ is based on the same C0 core stepping, which is now used in all contemporary processors on AMD64 architecture, including Opteron, Athlon 64 and Athlon 64 FX-51. you can clearly see it from the CPUZ utility:
Note that even the launching of the new Athlon 64 processor didn’t help AMD to overcome the 2.2GHz barrier. All top CPU models from the today’s AMD processor families starting with Athlon XP work at this particular frequency, but not any higher than that. However, AMD doesn’t need any further frequency increase yet. Intel also hasn’t raised the working frequencies of its processors lately, waiting for the introduction of the 90nm production process. This way, AMD has no causes for concern until February, I assume.
As far as the potential for further frequency increase, AMD might actually face some problems connected with the fact that the introduction of 90nm technology in their Dresden Fab30 can actually take place only in H2 2004. Nevertheless, Athlon 64 and Athlon 64 FX processors working at 2.4GHz still have a chance to appear a little bit earlier than that. At present AMD is finishing the work on the new Athlon 64 core stepping aka CG, which should allow increasing the core clock frequency a little more even with the current production technology. Besides, the new core stepping should also help solve some memory compatibility problems.
This way, AMD didn’t introduce anything brand new by launching their Athlon 64 3400+ CPU model. This allows us to pass directly to the benchmarking results skipping the CPU features discussion this time. If you would like to refresh your memory and learn more about the architecture and technologies implemented in Athlon 64 processors, please see our pervious articles on the matter in the CPU section of X-bit labs site.
The main goal of this test session was to find out the performance level of the new Athlon 64 3400+ processor against the background of its predecessors and closest competitors.
I would like to point out that in the meanwhile we still have to stick to testing the processors with AMD64 architecture in classical 32bit applications. The matter is that there are no mass operating systems in the market yet, which support 64bit AMD extensions. For example, the release of Microsoft Windows XP supporting x86-64 is expected to happen only in the second half of this year. That is why is you have an Athlon 64 or Athlon 64 FX processor, you should realize that your systems beast a few advantages, which will prove efficient and worthy a little later, but not now. However, we still cannot evaluate how great these advantages will be, as well as confirm if they will be there at all.
Our test systems were configured as follows:
The performance of the new Athlon 64 3400+ in the popular “semi-synthetic” 3DMark 2001 SE 3D test appears very close to that of Athlon 64 FX-51 processor. Since these both CPUs differ only by the integrated memory controller we can conclude that the average performance of the single-channel memory controller in Athlon 64 3400+ is comparable with that of the dual-channel controller used in Athlon 64 FX-51 CPU. How is that possible if the bandwidth of the latter is twice as high? In fact, everything is very simple: Athlon 64 FX-51 uses registered memory modules with higher latency than that of the regular unbuffered memory modules used in Athlon 64 systems, even despite the similar timings we set for both test systems.
CPU Score in 3DMark03 is not quite a CPU benchmark. This result is very dependent on the memory bandwidth, too. This is one of the reasons, why Athlon 64 FX-51 is tangibly faster than Athlon 64 3400+, despite the same core architecture and working frequency.
All in all, Athlon 64 3400+ is just a little bit behind Athlon 64 FX-51 in the general result.
Aquamark3 test based on a real gaming engine is one of the few gaming applications supporting multi-threading. Therefore, Pentium 4 and Pentium 4 EX processors supporting Hyper-Threading technology perform somewhat better here.
The increase in the core clock frequency of Athlon 64 up to 2.2GHz brought to us by the new Athlon 64 3400+ processor helped this CPU family to outpace significantly the Pentium 4 processor family in Quake3 test, which has always been won by Intel’s solutions (here I am not taking into account the Extreme Edition CPU).
In Unreal Tournament 2003 we witness a very funny situation: less expensive Athlon 64 3400+ outperforms Athlon 64 FX-51, even despite the single-channel memory. However, we have always stressed extreme sensitivity of this game to the memory subsystem latency.
In this pretty popular helicopter flight simulator Athlon 64 3400+, Athlon 64 FX-51 and Pentium 4 XE 3.2GHz processors ensure about the same performance level. By the way, their price is almost twice as different.
The tests in three pretty new gaming applications, namely Gun Metal, Tomb Raider: The Angel of Darkness and demo X2: The Threat, show about the same results. The CPUs with AMD64 architecture show their strengths here and outperform the competitors, even including Pentium 4 Extreme Edition.
As a result, I could say the following. The performance of Athlon 64 3400+ in gaming applications is very close to that of Athlon 64 FX-51, and sometimes is even higher than that. Both these processors work at the same core clock frequency and the differences in their memory controllers, which seem radical at first glance, do not let Athlon 64 FX-51 processor win an indisputable victory with a significant advantage. As a result, Athlon 64 3400+ processor can easily compete with Intel Pentium 4 Extreme Edition 3.2GHz, which is the today’s fastest CPU offered by Intel, leaving it notably behind in most cases.
As usual, here we are going to share the results obtained in Winstone test packages. Now we are using newer versions of these benchmarks released in the end of 2003. They are Business Winstone 2004 and Multimedia Content Creation Winstone 2004.
Business Winstone 2004 is a benchmark, which shows the average performance of the tested platforms during work in regular office applications. The test imitates the everyday work of an average user in a bunch of widely spread programs and displays the result basing on the time it took the programs to complete the tasks. The list of applications used in this benchmark for proper performance analysis is pretty long and includes Microsoft Access 2002 SP-2, Microsoft Excel 2002 SP-2, Microsoft FrontPage 2002 SP-2, Microsoft Outlook 2002 SP-2, Microsoft PowerPoint 2002 SP-2, Microsoft Project 2002, Microsoft Word 2002 SP-2, WinZip 8.1 SR-1 and Norton AntiVirus Professional Edition 2003.
Multimedia Content Creation Winstone 2004 is very similar to Business Winstone 2004, however, it uses absolutely different applications. All of them are intended for creation and processing of images, audio and video streams. The complete list of applications includes a number of popular and widely spread professional tools, such as Adobe Photoshop 7.0.1, Adobe Premiere 6.50, Macromedia Director MX 9.0, Macromedia Dreamweaver MX 6.1, Microsoft Windows Media Encoder 9 Version 9.00.00.2980, NewTek LightWave 3D 7.5b and Steinberg WaveLab 4.0f.
In both benchmarks Athlon 64 processors win the race, having left the regular Pentium 4 as well as its Extreme Edition counterpart far behind. Just like in the gaming benchmarks, Athlon 64 3400+ and Athlon 64 FX-51 again perform close to one another, balancing the scale to one or another side, depending on the way the applications included into the test package react to the two key parameters of the memory subsystem: latency and bandwidth.
Besides these two benchmarks we also decided to use a new Futuremark PCMark04. This test also measures the system performance during typical tasks processing. However, unlike Winstone tests, Futuremark developers included not only a few popular applications, but also a few popular algorithms apart from the applications where they are actually used. Here is a list of applications and tasks which are used within the PCMark04 package to measure the performance of our test system: ZIP archiving, spelling checking with the help of Link Grammar Parsing Library, web-sites rendering in Internet Explorer 6.0, image conversion into JPEG format, mp3 files decoding with the help of Ogg Vorbis library, video decoding with Windows Media encoder 9 and DivX 5.0.5, 2D graphics primitives processing, work in 3D via Microsoft DirectX 9 with Havok Physics engine 2.1 physical modeling system, anti-virus checking with F-Secure Anti-Virus, info encoding and decoding with Blowfish Algorithm.
If we disregard the results of the memory subsystem tests where Athlon 64 FX-51 with the integrated dual-channel memory controller demonstrates the highest performance possible, the AMD processors do not look that attractive in PCMark04 any more. The reasons for this unpleasant failure lie in the test itself. The thing is that the first half of PCMark04 test runs a few tasks simultaneously thus involving Hyper-Threading technology supported by the Intel Pentium 4 processors. And the second half of the test contains mostly algorithms for streaming data processing, which have been always faster on CPUs with Pentium 4 architecture.
Besides WinRAR archiving utility, which we usually use to measure the processors performance during data compression, we are going to involve a new archiving tool, which is getting more and more popular today. It is 7zip (http://www.7zip.org). Its distinguishing feature is extremely high (maybe even the highest) level of data compression today, which is achieved due to its own algorithm. Besides, it boasts an integrated benchmark, which allows evaluating the data compression and decompression speed for dictionaries of various sizes. For 7zip tests we used a 32MB dictionary.
Athlon 64 3400+ appeared the fastest during archiving due to the low latency of its memory subsystem and large L1 and L2 cache-memory. As for the fast data extraction from archives, the ALU processors performance is the major determinative here. That is why no wonder that all CPUs with Athlon architecture and 2.2GHz working frequency ran equally fast.
In most applications for audio and video data encoding Pentium 4 Extreme Edition 3.2GHz and Pentium 4 3.2GHz processors are the fastest. The only exception is the Canopus Procoder 1.5 benchmark, which has always been known for favoring AMD CPUs. As for the performance of Athlon 64 3400+ and Athlon 64 FX-51, the difference exists only in two tests out of four, where the CPU with a dual-channel controller works a little bit faster.
To test the CPUs in scientific applications we used a new beta-version of the ScienceMark 2.0 test package, which measures how fast the tested platforms can solve real math1ematic modeling tasks.
One of the most significant advantages of the Athlon architecture is a powerful FPU unit, which is very actively used in applications like that. That is why it is not at all surprising that even Athlon XP 3200+ demonstrated very high results in ScienceMark 2.0. Athlon 64 3400+ works at the same clock frequency but features larger L2 cache and an integrated memory controller, which actually do not improve its performance in ScienceMark 2.0 that much. However, even despite this fact we cannot accuse this CPU of low performance compared with the competitors.
Moreover, we have also tested the performance of our today’s racers in a popular symbol calculations package aka Mathematica 5.0 from Wolfram Research. To do this we used a benchmark, which is described in detail here.
And again Athlon architecture shows its real best. Intel processors are significantly behind even Athlon XP, not to mention the newer Athlon 64 and Athlon 64 FX. Here I have to say that Athlon 64 FX-51 and Athlon 64 3400+ again perform very close to one another.
We decided to test the performance of our processors also in ABBYY FineReader 7.0. This is a popular tool for optical text recognition, which requires powerful computational resources for proper functioning. To test the CPUs we used a text 48 pages long, which we scanned from four books of different type and quality. This way the text to be recognized included columns, complex tables, formulas, pictures, different fonts and different print quality. The diagram below shows the time it took each test platform to recognize the selected text fragment.
As we see, Athlon 64 FX-51 and the new Athlon 64 3400+ processors cope fastest of all with text recognition tasks.
To test the CPUs performance during final rendering we used a new sixth version of 3ds max. Besides the rendering speed for a single frame, we also measured the animation rendering speed with simultaneous recording of the results into an avi-file.
As usual, Pentium 4 processors are faster during single frame rendering (mostly due to Hyper-Threading technology), however, during the movie rendering the leadership goes to CPUs with AMD64 architecture.
In Cinebench 2003 test showing the performance in CINEMA 4D package, the situation seems pretty typical. Single frame final rendering is done faster by Pentium 4 processors with Hyper-Threading technology. However, the work in OpenGL is completed faster on AMD CPUs. Note that in this case the performance difference between Athlon 64 FX-51 with the dual-channel memory controller and Athlon 64 3400+ with a single-channel memory controller is more than evident.
Adobe Photoshop CS 8.0 is a very popular graphics application used mostly for 2D graphics editing that is why we paid special attention to the performance of our today’s testing participants in this application. We used PSBench7 benchmark with a 100MB image to be processed.
Just like in the previous version of Adobe Photoshop, Pentium 4 processors again appeared faster here. As for the performance of the new Athlon 64 3400+, it is not any different from that of the Athlon 64 FX-51: 6% lower than the performance of Intel Pentium 4 3.2GHz.
Now let me show you more detailed results illustrating how fast various Photoshop CS 8.0 filters are performed on different systems. The table below shows time in seconds:
Pentium 4 3.2
Pentium 4 XE 3.2
Athlon XP 3200+
Athlon 64 3200+
Athlon 64 3400+
Athlon 64 FX-51
Gaussian Blur 1
Gaussian Blur 3.7
Gaussian Blur 85
Reduce Size 60%
To test the platforms performance in the new AutoCAD 2004 we used a modified Cadalyst C2001 test.
In the new AutoCAD 2004 version AMD processors continue running very fast. Pentium 4 CPUs can successfully compete with the new Athlon 64 only in wireframe mode. As for the performance differences between Athlon 64 FX-51 and Athlon 64 3400+, they take turns in getting ahead depending on the type of actions they perform.
Well, we do not expect anything special from the overclocking experiments with Athlon 64 3400+. As I have already mentioned, this processor is based on the same C0 core as Athlon 64 3200+, with that only difference that this core now works at higher clock rate. Our previous experiments with Athlon 64 3200+ overclocking revealed the ability of this core to work at the maximum frequency of 2.3-2.4GHz without any extreme cooling solutions involved. Therefore, we expect the new Athlon 64 3400+ to demonstrate the same potential. The maximum may grow up only if the new CG core stepping comes out, however, there are no CPUs in the market based on the new core stepping yet.
I have to point out that the new Athlon 64 3400+ doesn’t allow increasing the clock frequency multiplier above the nominal 11x, just like the predecessor, Athlon 64 3200+. This is AMD’s policy: only more expensive Athlon 64 FX processors allow you to modify the clock frequency multiplier the way you like. That is why we will overclock Athlon 64 3400+ just like its predecessor, by speeding up the FSB frequency.
We didn’t use any special cooling solutions for our overclocking tests. All we had was a regular cooler from a boxed processor shipment. We increased the Vcore by 10% from the nominal 1.5V up to 1.65V. The maximum FSB frequency when the CPU remained stable notched 215MHz. This way, our Athlon 64 3400+ got overclocked from the nominal 2.2GHz to 2.37GHz. This result fully corresponds to what we saw by other Athlon 64 processors: this is evidently the maximum frequency for the current core stepping.
So, there is no doubt that AMD will have to undertake a certain core redesign, if they decide to release a CPU with 2.4GHz actual core frequency, before they shift to a new finer production technology.
Well, AMD introduced a new Socket754 processor from Athlon 64 family. The newcomer rated as 3400+ supports 2.2GHz core clock, which is the same as that of the Athlon 64 FX-51 announced in September 2003 and positioned as a solution for hardware enthusiasts. It will allow a considerably low-cost Athlon 64 3400+ demonstrate very attractive price-to-performance ratio, as it will offer comparable performance at a pretty low price of $417 at launch. The single-channel memory controller integrated into Athlon 64 3400+ doesn’t slow down its performance too much compared with that of the Athlon 64 FX-51 boasting a dual-channel memory controller. Another advantage of the new processor is its ability to work with unbuffered memory modules featuring lower latencies than the registered ones used in Athlon 64 FX-51 based platforms. As a result, the performance difference between Athlon 64 FX-51 and the new Athlon 64 3400+ is hardly noticeable at all.
The lag of Athlon 64 3400+ behind Athlon 64 FX-51, %
Advantage of Athlon 64 3400+ over Pentium 4 3.2, %
Advantage of Athlon 64 3400+ over Pentium 4 XE 3.2
Business Winstone 2004
Multimedia Content Creation Winstone 2004
3DMark03, Default, CPU score
3DMark2001 SE, Default
Aquamark3, Default, CPU
Quake3 (four), High Quality, 1024x768x32
X2 - The Threat, 1024x768x32
Unreal Tournament 2003 (dm-antalus), 1024x768x32
Gun Metal (Benchmark2), 640x480x32
Tomb Raider: The Angel of Darkness (paris3), 640x480x32
Comanche 4, 1024x768x32
ScienceMark 2.0, Molecular Dynamics Benchmark
ScienceMark 2.0, Primordia
ScienceMark 2.0, Cipher
MP3 Encoding, LAME 3.94b
Data Compression, WinRAR 3.2, Best
Data Compression, 7zip 3.11, 32MB, Compressing
Data Compression, 7zip 3.11, 32MB, Decompressing
MPEG-2 Encoding, Canopus ProCoder 1.5, Mastering
MPEG-4 Encoding, FlasK 0.78.39/DiVX 5.11
Windows Media Encoder 9, MP2 to WME
Adobe Photoshop CS 8.0
3ds max 6.0, Final Rendering, Underwater
3ds max 6.0, Final Rendering, Stadium animation
CINEMA 4D, Raytracing
CINEMA 4D, Shading
CINEMA 4D, Lighting SW
CINEMA 4D, Lighting HW
AutoCAD 2004, Total Index
AutoCAD 2004, 3D Wireframe Index
AutoCAD 2004, 3D Gouraud Index
AutoCAD 2004, Non-graphic Index
AutoCAD 2004, 2D Graphics Index
So, it is now possible to build the fastest desktop PC on a Socket754 platform. Thanks to the higher working frequency of the new Athlon 64 3400+, this CPU can successfully compete not only with the top Pentium 4 processor models, but also with the Pentium 4 Extreme Edition.
However, the situation in the market will change dramatically in the beginning of February 2004. This is when Intel is expected to announce its new Prescott core and Pentium 4 Extreme Edition 3.4GHz, which should definitely raise the performance maximum for Intel solutions.
As for Athlon 64 FX-51, it can hardly remain an attractive buy right now. The CPU itself, as well as the mainboards designed for it and the memory it supports cost considerably more, however, there are hardly any evident advantages of this more expensive platform over the fresh Athlon 64 3400+. Therefore, we can expect a new Athlon 64 FX-53 CPU with higher 2.4GHz core clock to be released soon, as it will have to “show its place” to Athlon 64 3400+ and to become a worthy competitor to the upcoming Extreme Edition 3.4GHz from Intel.