The bandwidth and latency numbers we just looked at provided rather uniform results. We will now see if this holds true for a pair of programs that heavily rely on the memory bandwidth and latency for their performance.

If the above numbers don't convince you that the WinRAR benchmark relies heavily on memory frequency, then you will never believe it. Let's look at the overclocked results and the fact that there is a 13 second gap between the 6-6-5 and 7-6-5 timing sets at the same CPU frequency. Then compare that to the only 9 second difference between the 7-6-5 and 8-7-6 results where the CPU frequency has a 300MHz difference. Clearly the CPU frequency had very little impact on performance at a certain point, otherwise the 8-7-6 timing set would be well ahead. There is a large difference between the first two sets of numbers which have a much lower CPU frequency and the three overclocked results. This would be indicative of a memory bandwidth bottleneck between 3.70GHz and 4.00GHz in WinRAR.

SuperPi is another program whose results are heavily influenced by memory frequency and timings. Generally speaking, frequency is king providing better results at higher memory frequencies regardless of timings. Of course there are always trade-offs between frequency and timings and we almost see one of them in the 32M results above between the first and second red bars. The gap is a minuscule 3.3 seconds in favor of the higher memory frequency. This is a direct result of the lower tRD value for the 6-6-5 timing set. If this setup was on P35, then the 7-6-5 timing set would also have been able to run PL5 and the gap would be around 10-13 seconds but because this is the X38 chipset, our tRD can only be a single value lower than the CL of 7. One final result of note is the 4GHz result. It is the first time we have seen a sub-12 minute 32M in a review here at HWC so that is certainly nothing to discount and is still a decent accomplishment for an all air 65nm setup. The overwhelming power of DDR3 at high frequencies can't be denied in SPi 32M.