by Anton Shilov
08/26/2002 | 04:05 AM
From the introduction of the SYSmark 2002 earlier this year, numerous reviewers have complained about poor scores on systems based on AMD CPUs compared to those powered by Intel`s processors. A lot of questions addressed to BAPCo concerned adding more tests that favoured Intel and increasing their contribution in the calculation of the final score.
Well, it now seems that the scandal begins to become even more serious as AMD issued a presentation that explained their discontent regarding the latest benchmarking software from BAPCo (available at VHJ). We may say that AMD`s dissatisfaction is quite understandable, though, we do not think we can now make any comments on the benchmark itself without an explanation from BAPCo itself. Keeping in mind that AMD joined BAPCo in the end of July (see this news-story), we expect they will somehow manage to find-out a solution for the problem.<%BANNER[article]%>
We want to attract your additional attention to the fact that AMD promised to introduce a new performance rating (see this news-story) that is intended for the new x86-64 processors next year. When the brand new architecture appears on the market, the developers certainly need support from the major developers of any types of software to drive the innovations to the real life. AMD will also need supporters among benchmarking software makers in order to be able to ground their new performance index.
It is clear that the industry will only win if BAPCo and the Sunnyvale based CPU-manufacturer work out an objective and comprehensive way of measuring x86-64 processor’s speed. Of course, BAPCo is not enough, but when it comes to certain situations, when big PC-suppliers choose CPUs for their product line, for instance, the results of their benchmarks are examined as the gospel truth. Presently, AMD may have to explain their partners why the company uses an older version of the benchmark instead of the new one.
Well, it seems that at this point, there are no universal benchmarks that could help customers to understand true performance a CPU can provide. Some benchmarks are rumoured to be biased, while the other may offer unstable or unexpected results. The only thing we can currently offer the observers are reviews made by our own test-laboratory, where the speed of different processors are measured using numerous benchmarks and analysed afterwards. As for universal performance ratings, we still have to wait a bit. AMD`s present approach is sensible enough, however, its main disadvantage is that performance is compared against comparatively old CPUs which does not support new technologies and trends.
To sum up, we have come to the same conclusion as before: Intel, AMD and possibly Transmeta accompanied by VIA have to collaborate in order to drive a universal way of measuring CPU performance. Of course, any rating becomes obsolete after one or two years, however, at the moment there is a total mess with megahertz and ratings accompanied by architecture-efficiency of totally incomparable CPUs.