ScienceMark 2.0, Pifast
Our look at the ST61G4's performance begins with ScienceMark 2.0's memory analysis. Generally, one can kind of predict probable benchmark results on similar chipset architectures if overall bandwidth and latency is known. So we can say that the Canterwood chipset is generally faster than the Springdale by dint of its extra bandwidth and lower DRAM latencies afforded by PAT-enabled BIOSes. Let's see how the ATI RS300 fares in this regard.
Not too well, I'm afraid. We usually expect to see around 4GB/s bandwidth for a clutch of i865G/PE / i875P boards. VIA and SiS have managed to equal that bandwidth figure with their latest dual-channel S478-based chipsets. ATI, it appears, has not. The benchmark was run numerous times with slight changes to the performance section of the BIOS. Any setting had negligible effect on overall bandwidth, according to ScienceMark 2.0. That automatically puts the ST61G4 at a serious disadvantage when compared to the i865G, its closest rival.
The latency analysis does nothing to help the ST61G4's cause. Low bandwidth and high latency will negatively impact upon most of our benchmarks that follow. Sure, this XPC is running the test CPU, and by inference the memory, slower than the SB62G2, but that in no way explains why it's so far behind in these synthetic tests. It's not looking good at the outset. A case of a memory controller that's not quite as efficient as it should be ?.
The situation is as expected in Pifast. It's simply a question of the DRAM not being able to get information quickly enough to the CPU. It can't transfer as much of it (bandwidth) and it takes longer to do so (latency). The inevitable result is a slower Pifast time than we had hoped for. Pifast absolutely thrives on both of those factors, so the near-5s deficit to the SB62G2 is not too difficult to explain away. The saving grace for the ST61G4 will have to be the performance of the onboard graphics. We'll come to that in a moment.
