Athlon 900MHz Platform Comparison

Pages: 1 2 3

System Level Performance

Application Benchmark Performance

Component level benchmarks are interesting from an academic viewpoint, but they obviously have serious shortcomings with regard to estimating ‘real world’ performance of a system. This is because they are intended to stress a particular component, and so do not act like any application would. As can be seen by the previous results, the design of the benchmark has a great deal of impact on what the results are. It is therefore important to use several different component level benchmarks, or understand exactly what the benchmark is testing so the results can be interpreted properly. Unfortunately, most benchmark manufacturers don’t want to reveal this information because it makes it easier for hardware manufacturers to build drivers or features that take advantage of it to make their product look better in the benchmark.

For this reason, application benchmarks are the most useful for determining how a ‘real world’ system will perform. However, there are inherent problems there as well, such as identifying exactly what is causing any performance differences. When the hardware changes can be kept to a minimum (such as swapping out only the processor or only the memory), it is obvious. However, when there are other changes (chipset, motherboard, cache size and speed, etc.) then it becomes more of a guess – unless one also has the component level results to compare it to.

For this comparison, I used ZDBOp’s Winstone 99 (both Business and High-end) and Content Creation 2000. At the time of this comparison, I did not have access to BAPCo’s Sysmark 2000, however it will be used in future comparisons. I also included the overall results from Passmark’s test, since it generates an aggregate score for all component level tests. It might be interesting to compare this with application benchmarks to see how well it correlates to ‘real world’ performance.

Benchmark

AZ11/SDRAM

AZ11/VCSDRAM

AK72/SDRAM

AK72/VCSDRAM

Content Creation 2000

27.7

29.3

28

29.1

Winstone99 Business

40.4

39.9

38.2

37.2

Winstone99 HighEnd

47.9

47.6

45.8

45.7

…AVS Express 3.4

6.76

6.84

6.39

6.53

…Front Page 98

4.47

4.39

4.16

4.17

…MicrostationSE

5.65

5.75

5.54

5.47

…Photoshop 4.0

5.27

5.28

5.18

5.15

…Premiere 4.2

4.75

4.74

4.65

4.62

…Sound Forge 4.0

3.52

3.46

3.48

3.42

…Visual C++ 5.0

4.39

4.29

3.93

3.94

     

Passmark Overall

87

88.3

94.8

86.2

With the exception of Content Creation 2000, the results of the application benchmarks indicate that the smaller, but faster, cache of the Tbird Athlon performs a bit faster. It is important to recognize that the ZDBOp benchmark results can vary by as much as 10%, so even using the average there is still a margin of error of somewhere less than 5%. With the exception of one test, all of these scores fall within that 5% range, and should therefore be considered roughly equivalent. The one test that seems greatly affected by the cache speed is Visual C++ 5.0, and even that is only about 11% faster.

Conclusion

While this probably doesn’t come as a big surprise to anyone, the only conclusion that can be drawn here is that the performance difference between the classic Athlon and the Tbird is negligible. This means that if you have a Slot A system now, and plan to upgrade, the only reason to replace the motherboard is if you simply cannot find a classic Athlon for it in the speed you want. The one possible exception to this is if your primary job involves frequent and/or large compilations, you might enjoy that 10% boost.


Pages: « Prev  1 2 3  

Be the first to discuss this article!