Review: ASUS P4S800D-E Deluxe

by Ryszard Sommefeldt on 5 November 2004, 00:00

Tags: ASUSTeK (TPE:2357), Intel (NASDAQ:INTC), VIA Technologies (TPE:2388)

Quick Link: HEXUS.net/qawl

Add to My Vault: x

Memory Tests

The memory biased tests give us a basis from which to guage the performance in all other tests. Since the CPU is involved in the memory tests we run, we get a good picture on CPU performance too, performance we confirm in all other benchmarks. So what are we looking for with the ASUS and more specifically SiS655TX? It's up against Springdale-PE on a board running a PAT-like BIOS. We're nearly spot on with Canterwood performance in this case, so it's a straight fight. If the 655TX can keep pace with the fast Intel solution, SiS should be proud. If it can do better then they are onto something good. Let's start with Pifast.

It's my custom test to 10 million places. Fancy a go? Click here.



The lower the better. A fast FPU on the CPU and fast, low latency memory controller are what you need for good results here. The Athlon 64 system beats the P4s by virtue of the best FPU in the business, along with an on-die memory controller with nearly half the latency of the northbridge-based controllers that i865PE and SiS655TX sport. In the battle between 655TX and i865PE+PAT, the 655TX holds a slight edge. That points to either more available bandwidth, lower latency, or both. Let's see what's true.

Sciencemark can tell us both things. As well as simple bandwidth tests, it can also calculate the time taken for certain memory accesses. We make it fetch a piece of memory outside of the CPU's cache to test main memory accesses by the CPU, and therefore latency. Here's the bandwidth graph first.



~80MB/sec doesn't seem like much, but it's enough to help explain the Pifast result. More bandwidth means we go faster in bandwidth limited apps. What about access latency?



Well hello there. With Springdale-PE + PAT BIOS and Canterwood accepted as having some of the lowest latency on a bridge-located memory controller, it was a surprise to see 655TX best it. And not by a nanosecond or two either. It's nearly 5%, outside the realms of statistical anomaly. That'll be the HyperStreaming technology going to work, putting SiS's own fast paths into action.

Summary

I was expecting 655TX to be close, but not that close. Lower latency memory accesses, at least in Sciencemark's usage case, are something to be proud of. The tiny smidge of extra memory bandwidth won't hurt either. So look for 655TX and the ASUS implementation to keep pace or better the DFI's performance in everything else. On to some non-graphical system tests.