Feature Request for Memory & Cache Benchmark Tool

[dontbugme]

Hi

 

I'm in the process of benchmarking my RAM and not ashamed to admit that I'm a n00b at it. 

After every change of the timing settings in the bios, I Reload Windows again, and run the AIDA Mem/Cache tool to compare the scores. As we know, even with the same timings the scores can change 10 minutes later, showing different results. This makes it very confusing to determine which actual timings are the best to settle on.

What would be MOST EXCELLENT would be for the Mem tool to recognize the same memory being used (perhaps just have the user save it with a name? that can be loaded up manually?) and then after running a benchmark it'll compare all the stats and TELL YOU which timings resulted in the best. "Idiot proof" RAM overclocking so to speak. 

Then whether you only compare 2 separate timings, or 20, you'll always still have the same, calculated and compared results, with the absolute best settings to use, and no guess-work based on visually comparing the 'floating' results that you get just by manually running tests and eyeballing to guess the best timings to use. Since the results can change 5 minutes later for the exact same timings (as mentioned above), perhaps have the tool run 3+ benchmarks in a row and average the numbers, per timings?

i.e. I run the test with 15, 17, 17, 28: it runs 3 tests in a row, averages the respective numbers, saves them to the memory profile. I reboot, change timings to 14, 16, 16, 24, load that profile up and run the test, it runs them 3 times in a row, averages, compares, you repeat the process 4 more times, all with different timings, and after each test the tool says "hey, based on our math, the (example) 15, 17, 17, 28 timings were the fastest to use."

MY GOD  would this save me a lot of headaches! 

[Fiery]

It sounds like a sophisticated and also quite complicated idea One immediate issue about it: how to decide which is the best setting?  Because sometimes a higher memory clock combined with worse timings could yield to a higher bandwidth but also higher (worse) latency.  One would have to gear the overclocking towards maximizing bandwidth _or_ minimizing latency.  The 2 together may or may not work out.

A loop that would cycle through 10 or even more tests of each benchmark may work out well, but then one would have to digest the results in a CSV or XML output, and that also would not be ideal for your imagined kind of usage I suppose.