System Benchmarks

With the system benchmarks, we have added a couple more real-world programs to the testing fold to get a better idea how memory interacts with some day to day usage programs.


As always though, we start with Super Pi and the 1M runs I can discount. Anytime you run 1M you can have this much variation on a run to run basis, but 32M is a different story. With three runs averaged and the 6-6-5 numbers actually pulling ahead of the 7-6-5 runs rather significantly, as far as 32M is concerned, it starts quite an interesting discussion. Personally, this reviewer uses SPi 32M to gauge memory performance before Everest or any other bandwidth test. 32M is so sensitive to memory performance, both bandwidth and latency, that it tends to be a better judge of character. These results might indicate that the 7-6-5 clocks we are getting on this board really are not in line with the other two timing sets, which was mentioned in the overclocking section. Hopefully NVIDIA can do something about this in future BIOS upgrades but it doesn't look good for CL 7 thus far for high clocking stability.


The WinRAR benchmark was replaced for 7Zip benchmark recently due to some inconsistent results and just like 32M SPi, we see a little bit of crack in the armor that bandwidth is always better. At the frequencies we are running each timing set, 6-6-5 again appears to be slightly superior to 7-6-5. At the other end, 8-7-6 just continues to dominate like it has in all benchmarks up to this point.


It seems to be the theme for the system based benchmarks that 6-6-5 at 820MHz is faster than 7-6-5 at 913MHz. So despite latency being almost identical and the bandwidth programs giving the advantage to 7-6-5 at these frequencies, all of the programs actually testing performance are leaning towards 6-6-5. Again, it isn't so much that 6-6-5 is walking about from the 7-6-5 numbers in the DivX and Lame encoding tests but it is keeping up and that alone is significant. The Lame results showing an almost dead even heat amongst the three overclocked timing sets may be an indication of a bottleneck elsewhere with that program, possibly in the hard drive reading and writing. We may have to find a better, multi-threaded, MP3 conversion program for benchmarking in the future.