[an error occurred while processing this directive]

AMD Athlon 64 X2 4800+/5000+ For Socket AM2: Continuing To Benchmark The New Platform

In this article we'll review another two AMD Socket AM2 processors: Athlon 64 X2 5000+ and Athlon 64 X2 4800+. These are rather interesting because the former has performance rating higher by 200 points due to its 2.6GHz clock rate versus 2.4GHz clock rate of the latter. But the latter has 2MB L2 cache, which is twice as capacious comparing to that of 5000+! Athlon 64 X2 5000+ has only 1MB L2 (512KB per core), while the X2 4800+ Socket AM2 offers us 2MB cache, classic for AMD's top dual-core. So, the main question is: is it justified to increase performance rating by 200 points only because the clock rate is higher? Especially, if that is accompanied by halving L2 cache?

Hardware and software

Testbed configuration

CPU
Mainboard
Memory
Athlon 64 FX-62 MSI K9N SLI Platinum (BIOS 9.03) Corsair CM2X1024-6400 (5-5-5-12)
Athlon 64 X2 5000+ ASUS M2N32-SLI Deluxe (BIOS 0603) Corsair CM2X1024-6400 (5-5-5-12)
Athlon 64 X2 4800+ ASUS M2N32-SLI Deluxe (BIOS 0603) Corsair CM2X1024-6400 (5-5-5-12)
Core 2 Duo E6400 Intel DG965SS (eng. sample) Corsair CM2X1024-6400 (5-5-5-12)
Core 2 Duo E6600 Intel D975XBX (BIOS 1181) Corsair CM2X1024-6400 (5-5-5-12)
Core 2 Duo E6700 Intel D975XBX (BIOS 1181) Corsair CM2X1024-6400 (5-5-5-12)
CPU Athlon 64 FX-62 Athlon 64 X2 5000+ Athlon 64 X2 4800+ Core 2 Duo E6400**** Core 2 Duo E6600 Core 2 Duo E6700
Process 90nm 90nm 90nm 65nm 65nm 65nm
Core clock, GHz 2.8 2.6 2.4 2.13 2.4 2.66
# of cores 2 2 2 2 2 2
L2 cache*, KB 2x1024 2x512 2x1024 4096 4096 4096
Bus clock**, MHz 400 DDR2 400 DDR2 400 DDR2 266 QP 266 QP 266 QP
Multiplier 14 13 12 8 9 10
Socket AM2 AM2 AM2 LGA775 LGA775 LGA775
TDP***, W 125 89 89 55-75 55-75 55-75
AMD64/EM64T + + + + + +
Hyper-Threading
Virtualization Technology + + + + + +

* - "2 x ..." means per core
** - for AMD processors this is memory controller bus clock rate
*** - measured differently for Intel and AMD processors; impossible to compare directly.
**** engineering sample; might differ from serial product

Software

  1. Windows XP Professional x64 Edition SP1
  2. 3ds max 7.0
  3. Maya 6.5
  4. Lightwave 8.5 x64 Edition
  5. WinRAR 3.51
  6. 7-Zip 4.32 x64 Edition
  7. LAME 3.98
  8. Monkey Audio 4.01
  9. OGG Encoder 2.8 (Lancer)
  10. Windows Media Encoder 9 x64 Edition
  11. MATLAB 7.1
  12. Pro/ENGINEER Wildfire 2.0
  13. SolidWorks 2005
  14. Microsoft Visual C++ Professional 6.0
  15. CPU RightMark 2005 Lite x64 Edition
  16. F.E.A.R. 1.3
  17. Half-Life 2
  18. Unreal Tournament 2004 build 3339
  19. Quake 4 Point Release 1.1
  20. FineReader Professional 8.0
  21. Adobe Photoshop CS2 (9.0)
  22. Canopus ProCoder 2.01.30
  23. DivX 6.1.1
  24. Windows Media Video 9 VCM
  25. x264 v.438
  26. XviD 1.1.0 Release
  27. Apache 2.0.55 for Windows

Drivers

  1. NVIDIA ForceWare 81.98
  2. NVIDIA nForce SMBus Driver 4.50
  3. Intel INF Update

Benchmarking

Essential foreword to charts

Our test procedure features two peculiarities of data representation: (1) all data types are reduced to one - integer relative score (performance of a given processor relative to that of Pentium D 805, given its performance is 100 points), and (2) detailed results are published in a Microsoft Excel table, while the article contains only summary charts by benchmark classes.

3D Modelling & Rendering

Athlon 64 X2 4800+, 5000+ and FX-62 formed standard though different "jaggies". It would be naturally to assume that 5000+ outperformed 4800+ less because of higher clock rate and smaller cache. So, a fully-fledged cache of same-clocked FX-62 would make that difference more noticeable.

CAD & CAE

Strangely enough, resource-critical CAD applications preferred clock rate over cache size. We can also see that the system performance bottleneck here is seemingly the graphics card. Because the high 11% dash of A64 X2 5000+ (comparing to 4800+) reduces to 7% (A64 FX-62 comparing to 5000+, respectively.) But the clock rate difference is the same 200MHz. Also significant that usually the fastest Intel Core 2 Duo E6700 couldn't outperform Athlon 64 FX-62 much, thus indirectly proving our assumption.

Compiling

In this benchmark clock rate advantage provided no benefit to A64 X2 5000+. However, compilers' sensitivity to L2 cache is not a secret (there's even an assumption that it's the 4MB L2 that makes Intel Core 2 Duo win at single-threaded compilation.)

RightMark

RightMark, like many other "calculators", considers the clock rate almost exclusively, given the architecture is identical. So, AMD processors form nearly identical jaggies.

Photo Processing

This is similar in case of Adobe Photoshop CS2...

Web Server

Such an amusing parity of Athlon 64 X2 4800+ and Athlon 64 X2 5000+: clock rate advantage was negated by cache size. Actually, if you take a look at the results table, you will see that 5000+ did better with short packets, while 4800+ performed better with long ones.

File Packing

Archivers use much memory to store their large dictionaries, so the victory of slower Athlon 64 X2 4800+ equipped with twice as much L2 cache was natural.

Audio Encoding

Nearly identical jaggies. The clock rate is the key here.

Video Encoding

This is actually similar to 3D modelling and rendering

Nearly identical jaggies between AMD processors. We are already getting used to those.

3D Games

Such an obvious love for cache makes us wonder if gamers are to actually pay attention to AMD performance ratings in the first place. Otherwise they might end up purchasing more expensive and formally faster processors - only to find out that they are being outperformed by lower-rated cheaper ones.

Total Score



Efficiency Per GHz



Conclusions

Considering the rather minor difference in performance (and presumably greater difference in price), we can say that Athlon 64 X2 4800+ looks somewhat more attractive than its senior sibling. This mostly has to concern home users (especially gamers!). But, in general, this benchmarking didn't produce any striking surprises. Everything was as expected.

Memory modules for testbeds were kindly provided by
Corsair Memory Russia

Stanislav Garmatyuk (nawhi@ixbt.com)
November 30, 2006
[an error occurred while processing this directive]