Conclusion: It Only Works With Optimized Software
Our comparison doesn't reveal anything terribly new: the Intel Xeon 2200 architecture is nothing more than a "normal" Pentium 4/2200 based on the Northwood core. The only difference is in the CPU platform - Intel provides the Xeon with Socket 603 and the Pentium 4/2200 with Socket 478. What's new is the dual capability as well as the Hyperthreading function, which, technically, is also integrated in the Pentium 4. It's a similar picture with the AMD Athlon MP 2000+, whose architecture is identical to the XP version - they are both based on the Palomino core. To enable dual operation, AMD only changed the coding on the L-Bridge at the top of the processor.
This comparison between the two workstation platforms from AMD and Intel clearly shows that dual operation does not necessarily mean increased speed for all applications. Rather, you need software that has been specially adapted to multiprocessor operation, so that the load is equally distributed between the two CPUs. With its Athlon MP 2000+, AMD has added a high-performance processor to its portfolio. And what's more - it certainly holds its own against the Intel Xeon 2200.
The AMD 760MPX platform is interesting in that two processors with varying clock rates can be used. This makes it possible for parallel operation of an Athlon MP 1200 along with an Athlon MP 2000+. However, with this variation of multiprocessing, the performance increase tends to be limited. Both manufacturers propagate the use of their processors specifically in workstations. Our selected benchmarks, chosen specifically for dual-processing, show that using two CPUs makes sense only with 3D rendering and MPEG encoding (MPEG-4, Divx). However, there's not much more to say about this topic, considering the state of technology at the moment.
A few final thoughts on Intel's Hyperthreading technology, which virtually doubles the number of processors: when using typical applications that are optimized for dual processing, Hyperthreading brings no advantages with it. Rather, the overhead on data slows down the application. Only software that is specially adapted for Hyperthreading enables an increase in performance. In addition, when Hyperthreading is activated, the memory performance decreases drastically, which is partially reflected by the memory benchmark.
