The Economical Way to a Pentium 4 System: Five Motherboards with the SiS648 Chipset

DDR333: Only With Two!

Why should SiS have it any differently than VIA? All current chipsets for fast DDR memory are hampered by the Achilles heel of fast clock rates and aggressive timing: three or even four DIM modules cannot be had at high processing speeds, as the resulting signal lengths create problems.

In practical terms, this means that DDR400 can only be achieved by using a single memory module. This, however, was mitigated by the fact that currently neither Intel nor SiS nor VIA supports DDR400 - performance beyond DDR333 cannot be achieved.

But even with DDR333, two DIMMs are as good as it gets. If the main memory is to be expanded to its maximum (up to 3 GB), three 1 GB modules are necessary. But the chipset can only manage this with 133 MHz (DDR266).

It Works, It Doesn't Work

The SiS brand alone is a basis for endless discussion. Numerous chipsets such as the 735/ 745 (Athlon) or 45/ 645DX/ 648 have regularly generated lots of talk about SiS in the past few months. But there are still many skeptics who question whether or not the current chipsets from SiS are really stable when running.

In many ways, SiS cannot even hope to be on a par with VIA and, above all, Intel. For example, its PCI performance is nowhere near that of the Intel chipsets. What's more, to a large degree, SiS has to vie with market competitors with its pricing. This means that it has to work as efficiently as possible in many areas. A manufacturer like Intel has more financial and hiring freedom when it comes to validating a new memory technology (like DDR333) to perfection before launching the product.

But for all that, SiS's work is not inferior. All five boards ran without a hitch in our lab using normal memory timings (DDR333 CL2.5).

Things get more difficult when it comes to fine-tuning: because motherboards with SiS chipsets have always sold primarily in the low-budget range, even manufacturers are cutting corners in order to remain competitive. As a result, aggressive memory timings are not thoroughly tested, or are only optimized to a modest degree. This makes it very possible that memory from manufacturer X will not work reliably with fast timings on the board from manufacturer Y.

But this is a question of attitude. Anyone who decides in favor of a low-cost product knows there is a risk of certain minuses. This decision, however, is not compatible with the desire for a perfect system. To sum up, all five candidates are perfectly capable of meeting requirements that are within reason. That is what they were built for.