X58 memory comparisons

Working on a new tool; recent discussions brought up the fact that, for the main part, i7 setups are pretty immune to memory speed changes (given the same memory); but, there are so many DIMMs spec'd at different speeds, that it's pretty hard to compare actual latency (which determines actual throughput), apples to apples... I wanted to post the beginning of this for comment - if anyone thinks of a field I'm missing, give a holler! What I'm doing is calculating latency nanoseconds from nominal frequency and latencies, then dividing by nanoseconds/cycle at a standardized ("Adj'd" labels...))speed of 1066, to get basic latency at Intel's only officially supported speed - whether you intend to 'abide' by Intel's speed limit or not, this will give you a way to compare latencies for all the different products available... Another feature is under "LV"; it's a 'latency value' calc, normalized to a max of 10 - trying to show the 'bang for the buck' relationship: the lower the latency, and/or the lower the price, the higher the LV...

3 answers Last reply
More about memory comparisons
  1. Interesting, if I understand what you're doing. You're showing the rated freq/latencies, then converting them to 1333 so they can be directly compared?

    Can you post the formula you use to get this? I'm interested in the same thing for DDR2 RAM. If I understand it correct, you can take the mfg's rated timings at say, 1066, and create your own spd figures to cover 800, 533, etc... using the formula which takes into account nanoseconds.

    Seems like this would solve forever the arguments about whether xxxxMHz @ Y is faster than yyyyMHz @ X.

    Is this how actual SPDs are created? They start say from 800 and go up to 1066 if the RAM will stand it, then go down to the 533, etc..?
  2. See, it's confusing for me because when I look at the SPD/EPP for my modules, I see this:

    1066/533 - 5-5-5-15 48 2T
    800/400 - 5-5-5-15 24 __
    533/267 - 4-4-4-10 16 __

    Why are my 1066 numbers the same as the 800s? Seems like either the 1066 should be higher, or the 800s lower.

    And I actually run my RAM at 800 and use the 533 numbers of 4-4-4-10 16
  3. The RAM in the table is 'normalized' to 1066 - Intel's only 'supported' speed for the 1366 processors (the 1156s support 1333). The calculation is ((1 / NominalClock)[which is 'cycle time' at rated speed] * RatedLatency) [which gives you 'latency time', rather than cycle count...]) / (1 / StandardClock) [which is 'cycle time' at the standardized clock] rounded up to the nearest integer = latency count at standardized speed. An example: if we take the OCZ3P1600LV6GK toward the bottom, which is rated 1600 7-7-7-24, for the seven latency, we get ((1 / 1600) * 7) / (1 / 1066) = 4.66375, so at 1066, it should run a five latency; likewise, for the twenty-four latency, it's ((1 / 1600) * 24) / (1 / 1066) = 15.99, so there, a sixteen latency should work at 1066...

    As for your SPD, I'm boggled, too... My G.Skill F2-8500CL5D is the same, but runs well at 4-4-4-12 800... Go figure! Safety margin, perhaps? Maybe insurance against board that don't 'take to' low latencies well?
Ask a new question

Read More

Gigabyte Latency Memory Motherboards Product