Performance

First of all, let’s check the memory subsystem performance of the nForce2 and KT880 in synthetic benchmarks. This will allow us to draw conclusions later, when we run a few real applications.

So, according to SiSoft Sandra 2004 results, as well as to the data obtained in ScienceMark 2.0, the memory controller of VIA KT880 is faster than the memory controller of NVIDIA nForce2 Ultra 400. Moreover, the VIA KT880 provides lower latency for CPU addressing the memory, as well as higher bandwidth. Is this enough to congratulate VIA on the victory? Not yet! The matter is that the performance of the processor-memory bus, which we are trying to measure with the help of these synthetic benchmarks is not limited by the memory subsystem performance in this case. Here the limitations are imposed by the processor bus, which provides twice as low bandwidth as the dual-channel memory subsystem. That is why the results obtained in these synthetic tests may not reflect the actual memory subsystem performance of the tested platforms.

Moreover, as we have already said a few times, NVIDIA’s major trump is not their dual-channel support. The strength of NVIDIA chipsets hides in the DASP unit (Dynamic Adaptive Speculative Pre-Processor). Its second version is implemented in NVIDIA nForce2 Ultra 400. Theoretically, this unit is just an additional buffer (or cache, if you want), built into the chipset North Bridge. Then this buffer is loaded with the data, which should be required by the CPU later, as the logical part of the DASP unit assumes. So, if the predictions and data sampling from the memory have been successful, the latencies during work with real applications can be significantly reduced due to more efficient use of the available memory bandwidth. It is exactly due to DASP that NVIDIA’s chipset gets some chances to outperform VIA KT880 in real applications, even though its results in the synthetic benchmarks are far from being impressive.

Well, let’s see.

Well, this is exactly what we have been expecting. Due to the DASP unit, NVIDIA nForce2 Ultra 400 chipset didn’t let the new VIA KT880 take the lead in real applications. In 19 benchmarks out of 25, the leading position belongs to the dual-channel solution from NVIDIA. In order to evaluate how much slower VIA KT880 appeared than its rival, take a look at the graph below:

So, the average performance of VIA KT880 is about 0-3% below the leader. However, there are a few more interesting things. Take WinRAR, where NVIDIA chipset with the DASP unit outperforms VIA KT880 by the good 6%. At the same time, in Aquamark3 VIA KT880 is on the contrary, a little bit faster than nForce2 Ultra 400. However, let me repeat once again: the NVIDIA chipset demonstrates evidently higher performance than the newcomer from VIA.