Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I want to ask something: Does RAM timings with n-n-n-3n (i.e. 8-8-8-24, 9-9-9-27, etc) perform better or is optimized more than the other who doesn't (i.e. 7-8-7-21, 9-9-9-24, etc)?

share|improve this question

migrated from serverfault.com Jan 14 '11 at 8:47

This question came from our site for system and network administrators.

4 Answers 4

Both the long and short answers yield the same thing: no.

While you may believe by intuition that multiples of eachother would cause better system performance, this is not the case. Those timing numbers simply represent the timing delay (in nanoseconds) of various operations for the memory to perform. In every single case, you want these numbers to be as low as possible.

The lower the numbers are, the faster the RAM performs, and thus, the more operations and bandwidth can be performed per unit time. At the same frequency, 8-7-8-20 will outperform (albeit not much more) memory running at 8-8-8-24. If you compare these propagation delays to the time between CPU cycles, you will find that the RAM is always slower, and thus, the limiting factor.

For example, if you have a 2 GHz processor, the time between cycles is only 0.5 nanoseconds. As you can infer from this, obviously you want the fastest possible RAM. There is no advantage to matching the various timings as multiples of each other.

share|improve this answer

Essentially, the lower the memory timings, the better.

Of course, there are other variables that affect performance, like memory channel configuration (dual/triple channel). Also, if they're all not similarly matched specs-wise, I believe they will downgrade and match themselves to the lowest-spec module.

share|improve this answer

Theoretically, yes. However, in practice, you probably won't be able to notice a difference unless you know you are bottlenecked by the memory. Most of the time the storage device is the slowest part (hard drives especially), not the memory speed or latency.

In addition, higher bandwidth memory is usually better than lower latency memory, unless the bandwidth is limited by the memory controller (eg. i7 900 is limited to 1066 by default; motherboard manufacturers might allow higher frequencies with or without overclocking).

I wouldn't bother with anything other than DDR2 800MHz CL6 or DDR3 800MHz CL9, unless it runs at stock voltage (1.8V for DDR2, 1.5V for DDR3) and costs about the same.

share|improve this answer
    
This answer is wrong. Faster timings are always faster, and there is no sense in recommending 800 MHz RAM when there is much faster RAM with tighter timings available. –  cp2141 Feb 11 '11 at 18:36
    
@cp2141 Low quality flags are NOT for incorrect answers. They are to be used for answers that are too short, or thank you's. Moderators do NOT intervene in wrong answers, the community and voting takes care of that. –  Diago Feb 11 '11 at 18:40
    
@cp2141, sure they are faster, but can you actually notice the difference between 6 and 4ns timings? There are other areas that will grant a much larger improvement in performance. No point in getting anything than the standard 1.8V DDR2 800 or 1.5V 1333, unless everything else is maxed out and you KNOW that the memory is the bottleneck. –  Mircea Chirea Feb 11 '11 at 18:59
    
I'm not doubting whether or not it's worth the extra money for tighter timings, but you allude the readers to believe the n-n-n-3n RAM is faster then ANY other combination. Otherwise, why did you write "Theoretically, yes."...? –  cp2141 Mar 12 '11 at 14:48
    
@cp2141: "However, in practice, you probably won't be able to notice a difference...". RAM is the fastest component in a computer; 3% or so improvement in throughput/latency won't matter when it's already too fast. –  Mircea Chirea Mar 12 '11 at 16:50

Modern CPUs have enormous caches. For typical software, the cache hit rate is around 90% to 95% (and sometimes as high as 99%). Improvements in memory timing will have a very small effect overall.

Trust me, I know this from measurements I made for several different types of programs. I made a memory cache performance analysis tool a few years ago that uses the CPUs performance counter registers. Free tools exist if you want to prove it for yourself (cachegrind).

If you care about stability more than a couple of percent of performance, go for less aggressively timed memory. You will have less issues with it and it will probably last until you don't want it anymore.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.