Calculating the factors a, b and cAs mentioned before, time is sadly enough not on my side. Because of this, the amount of hours investigating the exact memory performance scaling was limited to ‘just a couple’. In order to find the numbers fitting variables a, b and c we used the Lavalys Everest Memory Bandwidth application, this to maximize the effect of changing the memory parameters.
Note that by using more benchmarks to determine the effect of the memory parameters, you will have a better and more realistic chart.
After analysis of the scaling results on the AMD DDR3 platform, we found that the effect of the different parameters can be written as three simple linear equations:
f(mhz) = 2780,4x + 5344,9
f(tcl) = -1126,3x + 11967
f(trcd) = -123,2x + 11015
In these equations of the form y = ax + b, the variable a describes the steepness of the slope. The higher the absolute value of this figure, the more steeply the line or, in this case, the more important the memory parameter plays a role in terms of performance. Note that ‘minus’ does not mean that the slope is less steep, it just says that the slope is angled in a different direction. For more information: Wikipedia!
To make our life a bit easier, we can use the relative effect of each memory parameter to add to our formula. So,
MHz => a = 2780,4
tCL => b = 1126,3
tRCD => c = 123,2
tRCD => c = 1
tCL => b = 1126,3 / 123,2 ~ 9,1
MHz => a = 2780,4 / 123,2 ~ 22,6
For the Intel X58 platform, we can do the same calculations and come to the conclusion that:
a = 7,5
b = 3
c = 1
Note that the variables a, b and c for the AMD DDR3 platform and Intel X58 platform cannot be compared to each other! They are relative within the platform. This means that they can only be talked about in terms of relative weight, not absolute.
Collecting the data
Now that we’ve determined how to calculate the rating for the different memory kits, we can go to the next step: determining what kind of data we will use for calculating the ratings. As mentioned before, the principle we are using states that we determine the overclocking capabilities of a memory kit by how easy it is to reach a better performing configuration (frequency and timings).
The query used to extract the useful data from our database will have to be written accordingly, which means that we have to remove all biased data. In order to do this, we have removed the overclocking data from the following applications:
All results that were blocked or soft-deleted
All results where the memory was clocked over 1350MHz (DDR3-2700)
All results where the memory was clocked lower than 600MHz (DDR3-1200)
All results with tCL lower or equal to 5
All results with tRCD lower or equal to 4
All results from applications Sisoft Sandra, CPU-Z, Reference Clock and Memory Clock
Mainly, these limitations were set in place to remove as much inaccurate data from the results. We removed the data from the Memory Clock application because it’s likely that people enable settings such as B2B-Cas Delay in order to achieve the maximum frequency.
Follow us on Twitter
Follow us on Facebook
We found out ourselves that the real world differences between fast/slow memory is only a couple of % (1-3%), however, everest makes the differences much bigger. There's also the question whether a better everest bandwidth score automatically means a better real world performance : it often does, but not always.
My biggest gripe is with the charts though, and is not "your fault" in any way : there are plenty of types of OCZ blades for example, some good, some bad, with different chips on them and different specs. Still : they're all thrown into one category "OCZ blade". Contrary to that : Corsair GTX is always elpida, so naturally sticks out (but should devided into several types itself). What this leads to is over generalization : "If I buy Corsair GTX I'll be ok", whereas a blade series @ eg 2ghz cas 7 is still better than a 1866 cas 8 corsair.
All in all, forgive me PJ, I find the graphs presented this way much too general to provide any real use.