L O S T C I R C U I T S

Benchmark Selection

Knowing this, it is a matter of selecting appropriate benchmarks to elucidate the differences between platforms. Likewise, the focus can be on other benchmarks to emphasize the absence of differences. The entire situation comes down to a statement like trains are running faster downhill than at night.

Most office applications are handled by the CPU cache and, therefore, there can and will not be much difference between benchmarks using either SDRAM or DDR as long as the other system parameters are the same. Another deciding criterion is the ratio between I/O latencies of the mass storage components and the computational performance contribution. This means, that there are benchmarks that measure performance and there are benchmarks that measure the impact of bottlenecks on system performance. In other words, any benchmark tacking the slowest system component will primarily measure bottlenecks rather than performance.

The most commonly used benchmarks for office applications are Winstone2000 and CSA Research OfficeBench 1.2. Winstone2000 is running mostly from the cache but is handicapped by intermediate loading and saving of data to the HDD with frequent swapfile accesses. Therefore, for a cross-platform comparison, it may not be the best suitable tool. OfficeBench 1.2 is probably the most precise implementation of a typical, human, user pattern into a benchmark in that it mostly scrolls around in Excel spread sheets and Word documents without doing too much computation. Despite the fact that it was developed specifically to show off the advantages of the Rambus architecture and, thus, may be a good tool for looking at certain aspects of memory transfers, my take is that, for a high end machine, this benchmark is highly irrelevant (no pun intended and no offense either).

Similarly dependent on HDD performance is ContentCreation Winstone2001. Please keep in mind, that all these benchmarks are valuable for evaluation of system performance but not for a cross platform comparison since the possible differences in performance are masked by bottlenecks imposed by I/O latencies.

ContentCreation Winstone2001 is dominated too much by HDD latencies to show much difference between DDR (purple) and SDRAM (turquoise) systems at the same speed.

All contents of this site are protected by international copyright laws. Reproduction of the contents even in parts is not allowed except after written permission by the author and referral to this site.
Copyright 2000 - 2001 LostCircuits