iXBT Labs - Computer Hardware in Detail

Platform

Video

Multimedia

Mobile

Other

AMD Athlon 64 3400+ compared to other top processors from Intel and AMD

January 25, 2004



The recently announced desktop AMD Athlon 3400+ processor doesn't offer any surprising innovations, it's just an Athlon 64 3200+ with core clock increased by 200MHz. Therefore, to those who want more details about the Athlon 64 (and the entire AMD K8 series) architecture, we recommend our test and theoretical articles, while in this introduction we'll attract your attention to a number of some ambiguous (and funny) facts accompanying the 3400+ exactly.

First, Athlon 64 3400+ core clock is identical to that of Athlon 64 FX-51. Therefore this “extreme” desktop CPU from AMD surprisingly received a rival in person of more usual (and cheaper) desktop CPU, so the only advantage left to FX-51 is a dual-channel Registered DDR400 SDRAM controller. And we know that not all applications are critical about the memory.



Second, it's also important that we probably for the first time witness such a strange AMD processor rating increase. You see, the clock speed of Athlon 64 3400+ is 200MHz higher than Athlon 64 3200+ (while other features are identical)... And its rating is the same 200 points higher! We hadn't witnessed this for some very long time, as AMD always claimed that the performance growth (which this “####+” rating is called to indicate) should grow faster than clock speeds! E.g., Athlon XP 3000+ has 2100MHz or 2167MHz clock speed (there are two models), while Athlon XP 3200+ (+200 to the rating) has 2200MHz clock speed that is just 100MHz or even 33MHz higher! Funny, isn't it? Either AMD changed the ratings policy again, or... Anyway, there's no need to surmise as we have some more ponderable performance estimations – test results.

Testbeds and software

Testbed

  • Processors:
    • AMD Athlon 64 3200+ (2000 MHz), Socket 754
    • AMD Athlon 64 3400+ (2200 MHz), Socket 754
    • AMD Athlon 64 FX-51 (2200 MHz), Socket 940
    • Intel Pentium 4 Extreme Edition 3.2 GHz, Socket 478
    • Intel Pentium 4 3.2 GHz, Socket 478

  • Motherboards:
  • Memory:
    • 2x512 MB PC3200 DDR SDRAM DIMM TwinMOS
    • 2x512 MB PC3200 Registered DDR SDRAM DIMM Corsair (for Athlon FX)
  • Graphics card: Manli ATI Radeon 9800Pro 256 MB
  • HDD: Western Digital WD360 (SATA), 10000rpm

Software

    • Windows XP Professional SP1
    • DirectX 9.0b
    • Intel Chipset Installation Utility 5.0.2.1003
    • NVIDIA UDP 3.13
    • VIA Hyperion 4.51
    • VIA SATA Driver 2.10a
    • Silicon Image Driver 1.1.0.52
    • Terratec DMX 6fire Drivers 5.40
    • ATI Catalyst 3.9
    • Gray Matter Studios & Nerve Software Return To Castle Wolfenstein v1.1, demo checkpoint.dm_57
    • Croteam/GodGames Serious Sam: The Second Encounter 1.07, demo Grand Cathedral
    • Digital Extremes/Epic Games/Atari Unreal Tournament 2003 v2225, botmatch antalus
    • Cacheburst 32 0.91.07

Motherboard ABIT KV8-MAX3 ASUS P4C800 Deluxe Gigabyte K8NNXP
Links ABIT KV8-MAX3 ASUS P4C800 Deluxe Gigabyte K8NNXP
Chipset VIA K8T800 (K8T800 + VT8237) Intel i875 (RG82004MC + FW82801ÅB) NVIDIA nForce 3 Pro 150
CPUs supported Socket 754, AMD Athlon 64 Socket 478, Intel Pentium 4, Intel Celeron Socket 940, AMD Athlon FX
Memory sockets 3 DDR 4 DDR 4 DDR
Expansion slots AGP / 5 PCI AGP Pro / 5 PCI AGP / 5 PCI
I/O ports 1 FDD, 2 PS/2, 3 FireWire 1 FDD, 2 COM, 1 LPT, 2 PS/2, 2 IEEE1394 1 FDD, 1 LPT, 2 COM, 2 PS/2, 2 FireWire-800
USB 4 USB 2.0 + 2 2 USB 2.0 4 USB 2.0 + 2 2 USB 2.0 2 USB 2.0 + 2 2 USB 2.0
Int. IDE ATA133 + SATA RAID ATA100 + SATA ATA133
Ext. IDE / SATA Silicon Image SiI3114CT176 Promise PDC20378 Silicon Image SiI3114CT176 + ITE IT8212F
Audio AC'97, Avance Logic ALC658 AC'97 , Analog Devices AD1985 AC'97, Avance Logic ALC658
LAN 3COM Marvell 940-MV00 3COM Marvell 940-MV00 10Base-T/100Base-TX + Realtek RTL8110S-32
I/O Winbond W83627HF-AW Winbond W83627THF-A ITE IT8712F-A
BIOS 4 Mbit Award BIOS v 6.00.PG 4 Mbit AMI BIOS v. 2.51 4 Mbit Award BIOS v. 6.00PG
Form-factor, size ATX, 30.5x24.5 cm ATX, 30.5x24.5 cm ATX, 30.5x24.5 cm


Low-level testing

CPU core performance

Well, we guess it's high time to run our CPU RightMark to benchmark the processors and also try out the latest version of RightMark. Though there are many options in this test, we run them all and analyzed all the results to select the most indicative: SSE/SSE2 instruction sets, 1 and 2 threads (considering the virtual multiprocessors of Pentium 4). We must say this selection is optimal even for the Athlon 64 solutions as they performed best using these instruction sets exactly (we also tested MMX/SSE2, MMX/FPU, SSE/FPU). Essentially this is good indicating the good level of SSE/SSE2 support in AMD64 architecture.

In this diagram you can see one of two intermediate results: physical model processing performance. As you remember, these results are used as a source for CPU RM rendering module. Anyway, it's interesting from the very beginning: in a purely mathematical task the leadership belongs to all AMD processors though the lag is not that great. But there's another thing: the results of both Pentium 4 CPUs worsen with two threads (i.e. Hyper-Threading)!.. But this turned out to be the problem of Performance and Stability test benchmarking modules, which we'll fix (remember that CPU RM 2003 is still beta.)


And the rendering module behaves as expected: two threads provide significant boost to Pentium 4. It doesn't double actually, but who promised that two virtual CPUs were to work like two physical? So, here the situation is predictable: Intel loses to AMD without Hyper-Threading and wins when HT is enabled.

 

The bottom mark (essentially the final) still gives the palm of supremacy to Intel though Athlons succeed to catch up a bit. In general, we witness the struggle of architectures: within one of them different CPUs perform almost the same depending on clock speed only. The large L3 cache doesn't help Pentium 4 EE as well as the dual-channel controller doesn't help Athlon 64 FX-51. Moreover, Athlon 64 3400+ operating at the same clock speed as FX-51 does, is even slightly ahead with non-registered/ECC memory (the latency?..)

Memory subsystem performance

We'll traditionally estimate it using another low-level benchmark - CacheBurst32. As usual, we are interested in memory read, write and latency (the lower result is naturally the better).

 

Well, nothing surprising here: every processors performs accordingly to its bus. Pentium 4 with its Quad Pumped Bus (equivalent to 800 MHz) is hors concours. What's more interesting - Athlon 64 3400+ won over Athlon 64 3200+ just a 1%. By the way, the clock speeds differ by 10% already. Does it mean that Athlon 64 with its single-channel DDR400 is almost totally about the memory subsystem, if the 3000MB/s read throughput seems to be the limit at the given RAM clock?..

 

Write results are more complex. Here the large cache did a good job for Pentium 4 EE, while Athlon 64 FX-51 nicely relied upon its dual-channel controller (if you remember, it's the main difference between FX-51 and 3400+; you can see its role in this test yourself). And again the comparison of Athlon 64 3200+ and 3400+ aggrieves as the increased clock speed provides almost no performance boost. It really seems the single-channel controller is stalling.

 

The high (relatively to AMD64) Pentium 4 latency is not a secret. And the highest latency among AMD processors is indicated by FX-51. Well, it's no surprise, the more reliable and stable registered/ECC is not a free gift, its latency compared to the similar unregistered is always slightly higher.

Testing in actual applications

Graphics performance

 

Since version 4.26 3ds max became “addicted” to Intel and the version 5.0 introduced even more optimizations for Hyper-Threading. However, remember that in our case an external render is used so 3ds max rendering engine is not related to the test results at all. And we don’t know for sure about the optimizations for Pentium 4 Brazil Rendering System. Yes, it supports SMP and SSE, but that’s all. However, Pentium 4 wins anyway – not that significant, but enough. But that’s not the most interesting – note that Athlon 64 3400+ is again slightly ahead of Athlon 64 FX-51. Well, you just have to get used to it, as this trend will develop in further tests.

 

During the previous similar Athlon 64 FX-51 testing we stated it was ideal for Photoshop. However, that time Intel still won over the purely desktop Athlon 64 3200+. Today it doesn't have even this advantage: purely desktop Athlon 64 3400+ easily outperforms Pentium 4 3.2 GHz and even wins 3 seconds over Pentium 4 eXtreme Edition. Actually we witness the strengthened and now unambiguous leadership of one rival platform over another in this application. Also note that Athlon 64 FX-51 is considerably ahead of 3400+. This testifies that Adobe Photoshop is very sensitive to memory operations that are more important to it than latency. However, the core architecture is still more determinative otherwise the winner would be Pentium 4 eXtreme Edition with its fastest memory access.

Audio encoding

Our audio encoding method is wide and diverse enough, but due to this there’s no sense in publishing the whole bunch of test results. Therefore we decided to pick out two tests: LAME MP3 encoding and Vorbis Tools OGG Vorbis encoding.

 

 

These tests emphasize each other well forming the similar, but still opposite pictures. Intel leads in LAME, while AMD in Oggenc. Such a calm competition without any significant lags. Note the indifference of both tests to anything except the clock speed: Pentium 4 eXtreme Edition is even with its usual sibling as well as the other same-clocked couple - Athlon 64 FX-51 and Athlon 64 3400+. We guess we’ll not be including non-specialized audio encoding tests into reviews until there are processors of other (or changed) architectures as the situation is clear enabling rather precise predictions with only processor clock speed values and a calculator.

Video encoding

From this method we selected the majority of tests – two “ripping” and “film-making”. We must say that a Windows Media Video 9 encoding test to spite the more popular XviD might seem not very logical, but it’s related to some peculiarities of the latter. Actually there’s a new version available, but we didn’t have time to adjust it to our test kit. Besides, XviD deserves a separate article anyway. As for the “film-making” (MPEG2) part, we naturally selected Mainconcept and Canopus.


 

It’s totally identical to LAME + Oggenc — in one test the leader is AMD, in another the leader is Intel. It’s funny that AMD64 was favoured by a product from Microsoft often blamed for strong bias to Intel. It’s more intensified by the fact that Windows Media Video 9 supports Hyper-Threading (and naturally SMP), but in this case this doesn’t help Intel’s processors. The differences between “extreme” and usual desktop CPUs are minimal, so the expediency is questioned more and more often...

 

 

You might blame us for selecting tests to create the initially desired picture. We assure you it's not true. We were actually surprised with such clear results ourselves – just take two similar applications and you can see that each is dominated by a different architecture. Well, maybe we just stumbled upon the truth? Note the repeating situation when Athlon 64 3400+ outperforms its “extreme” FX-51 sibling by several seconds. Of course, if we separately consider the results of a single test, we might think it’s just inaccuracy. But this actually repeats very often in some very different tests!

Archiving

From this article we start using the latest version of 7-zip (3.13). This explains the difference comparing to the previous tests. We also stopped using the special (and rather complex) options for the sake of more common <-mx —mmt> for Intel and <-mx> for AMD (‘-mmt’ enables multi-threading, and the application author advises not to use it with non-Hyper-Threading processors.)

 


 

In WinRAR Athlon 64 3400+ won already 5 seconds over FX-51 instead of one or two. However, we already used to this. 7-zip shows more interesting results: we decided to check its Hyper-Threading support and tested it on and off on Pentium 4 systems. The results are just startling: it’s exactly Hyper-Threading that enables both Pentium 4 to win. Without HT the performance drop is disastrous. We also checked how Intel processors are affected by HT forced off - 17% of performance in both cases.

Games

Gaming tests are traditionally combined into a single section in our reviews with minimal comments. Here’s our position again: almost nobody purchases top processors for gaming. First, it’s economically inexpedient; second, it’s just stupid. If you have surplus money and want to improve PC performance in games, you should purchase a more powerful graphics card. Of course, it will require a powerful CPU – powerful, but not the top. And the difference between the top and its slower neighbours can be very considerable. And, besides, those who purchase Pentium 4 eXtreme Edition or Athlon 64 FX-51/3400+ for gaming do this not because these are fast, but just because it’s cool :).


 

 

Here you can see another “new drivers effect”. It’s clear that 64-bit (though still operating in 32-bit mode) AMD is the unambiguous leader. And we think this is due to graphics card drivers (probably optimized again) looking to Return to Castle Wolfenstein. Note that in the low-res “Fast” mode that is traditionally used to estimate processors’ and chipsets’ performance Pentium 4 eXtreme Edition wins. But it starts to lose to Athlon 64 FX-51 and 3400+ when there's more load on the graphics card at higher resolutions! We can only suppose that the higher graphics card load means more work done by its drivers, which are executed faster on Athlon 64. Well, maybe it’s time to cite a standard disclaimer for testers that do not have access to specific application source codes: We are not sure 100%, but this just seems logical enough”.

Summary diagrams for all tests

This time we decided to separate summary diagrams according to test groups due to their significant difference. As usual, comments to these are shortest, as everything is obvious anyway. If you remember, 100% stands for the worst result. Besides, if the resulting value is inverse of performance (e.g. time), worst to best results ratio is calculated - (K/X)/(K/Y) is correctly cancelled to Y/X. So the more is the better in summaries.

 

Intel wins, but the leadership is not astounding.

 

Some more complex situation here: each platform wins where it’s the strongest. For Intel it’s linear access performance, while for AMD it’s latency. You can’t just directly compare these things, as each is critical for the corresponding software types (both are critical simultaneously very seldom).

 

It’s almost parity, but the best of AMD wins more points over the best of Pentium 4 in Adobe Photoshop than the latter wins over Athlon 64 in 3ds max. AMD scores more.

 

Here Intel scores more due to considerable advantage in LAME.

 

Intel scores more again, but we insist on the draw: both Pentium 4 processors won clearly only in Mainconcept MPEG Encoder.

 

In WinRAR rivals are even, but in 7-zip Intel is the leader. Still, as always, we can’t say it “devastates” AMD.

 

AMD obviously wins in all the tests with no exception.

Conclusions

Let’s state the obvious first: Athlon 64 3400+ succeeds in winning the “desktop most powerful” from Pentium 4 3.2 GHz. You can check this according to the summary diagrams yourself. However remember that the leadership is very fragile. By the way, the previous leadership of Pentium 4 3.2 GHz over Athlon 64 3200+ wasn't firm as well, so it’s just the history repeating.

What’s more interesting – the arrival of Athlon 64 3400+ almost negates the effect of Athlon 64 FX-51. Just look: 3400+ is cheaper and almost doesn’t lose to FX-51 in performance. Besides, you can install it into a usual desktop board and doesn’t require the expensive Registered ECC DDR400. Actually, if you exclude possible very specific and exotic applications, Athlon 64 3400+ means death for Athlon 64 FX-51. And as AMD claims there can be only one FX, it seems we should expect FX-53 in the near future. It’s not that urgent, but it's just a real evidence of the fact that 2200 MHz is not a problem for K8.

And if you finally analyze the current situation on the background of the past few years, you will see that despite crowds of users adherent to both companies are breaking lances in their virtual battles, neither Intel and AMD couldn’t demonstrate any serious and long-term technological and/or engineering dominance since the release of Athlon in 1999.

When AMD releases Athlon, Intel announces Coppermine. When AMD unveils Thunderbird and then Palomino, Intel promotes Pentium 4 (here’s actually the lag until Northwood, but it’s not long-term). Intel presents Northwood and later Northwood with Hyper-Threading, AMD starts to lag behind, but only until Opteron and Athlon 64 announcements, which enable AMD to catch up and even slightly outperform Intel in the desktop field. So, if Prescott doesn’t show fantastic performance (does anyone still believe in fairy-tales?), then we shouldn’t see anything astounding in the x86 CPU segment, maybe just another formal swap.

So, if both companies have their businesses in good condition (the opposite seems just impossible), the ultra popular soap opera called “Intel vs. AMD” will live long on our displays. “…The year 2004 was marked by the strong competition between the two primary rivals in the x86 segment and brought a number of insignificant victories to both, but no radical changes were witnessed… This phrase is already in my concluding review.

Stanislav Garmatyuk (nawhi@ixbt.com)
Dmitry Mayorov (destrax@ixbt.com)

Write a comment below. No registration needed!


Article navigation:



blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook


25

Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.