Which RAM mathematically works out to be the fastest?

zacw

Honorable
Oct 25, 2013
113
0
10,690
Okay, I know there is a age old debate determining CAS latency vs raw frequency. I also understand that they share an interconnected relationship of stability making one high naturally will make the other one lower for stability.

Anyway I am not looking to create a debate, I'm after somebody who can mathematically calculate the best speed. I have read a few guides but I have very limited mathematical capabilities.

The below listed speeds are from the G.Skill website detailing the Trident X and the different versions. I myself have 2x 8 GB 1.65v 2400 Mhz kit but I am looking at changing it depending on which offers the best performance. I have tinkered enough with my CPU lately that I do not wish to change these values for any further performance however I have tested my kit at all of those speeds and I am stable.

RAM Settings

1600 1.5 7 8 8 24 2n
1866 1.6 8 9 9 24 2n
2133 1.6 9 11 11 31 2n
2400 1.65 10 12 12 31 2n
 
Solution
Look into this (memory-scaling-on-haswell). In this test/review they used same ram modules and OC them from 1333 cas 7 to 3000 cas 12 and there are tons of comparison and tables to look into.

It looks like fastest option is around 2400MHz with cas [strike]10[/strike] 9 or 2666 cas 10, but there are not big differences in general.

EDIT: It's not mathematical, but real world test. Mathematical/theoretical calculations are often far from real performance.
there is a formula to calculate latency
2000 x (CL / Speed ) = Latency in nanoseconds (ns)
so 1600Mhz has a latency of CL7 - 8.75ns
1866Mhz at CL8-8.75ns
2133Mhz at CL9-8.43ns
2400Mhz at CL10-8.33ns
 


Thanks for the quick reply, however does that formula take into account the variables that come after the initial CL ? like for CL10 then you have 12 12 31 2n
 
Look into this (memory-scaling-on-haswell). In this test/review they used same ram modules and OC them from 1333 cas 7 to 3000 cas 12 and there are tons of comparison and tables to look into.

It looks like fastest option is around 2400MHz with cas [strike]10[/strike] 9 or 2666 cas 10, but there are not big differences in general.

EDIT: It's not mathematical, but real world test. Mathematical/theoretical calculations are often far from real performance.
 
Solution


Not much about timings its more about latency
 
Thanks everybody for the quick replies, I took a little time to reply as I have been OC all day .. and falling really far behind at uni because of it 😛. I think i have found a great OC for my kit I will post the specs shortly once I can confirm stability
 
for reference to everyone, 1866Mhz with a CL of 9, works out and does appear to be the best, but it depends on the CPU, as different types prefer different speeds best. For instance;

Intel 2nd/3rd generation i3/i5/i7's prefer 1600Mhz CL 8/9

almost all AMD CPU's will benefit to the fastest ram capable of being supported by their FSB, usually 2400/2666MHz

Intel 4th generation CPU's best with 1866Mhz CL 9.
 
It all really depends on what you do with your system....if simply gaming, where DRAM is but a conduit for data, then yes 1866/9 is around the sweet spot, however, if you really use your rig, things like heavy multi tasking, imaging, video, VMs, CAD anything with large data sets, etc then faster DRAM will really shine. Running BMs simply show the slight increases while doing a singular thing, when multi-tasking, even just having multiple taabs open on a browser the DRAM is constantly updating itseld and being used, that's where the higher bandwidth comes to play with the higher freq DRAM
 


Yeah but for AMD it's like a paradox. The AMD chips benefit the most from fast RAM, but often the weak IMC cannot run it~

Side note. I always wanted to test some G.Skill Flares DDR3-2000CL7. Did Powerchip go out of business or something? They had the best overclocking ICs like in the ECOs that could do DDR3-1600CL6 with ease.

 
Hey guys, sorry once again for a late reply, I really do appreciate the help everybody has given me, running a few tests and here and there I have found that by tweaking the ram at best I have documented has seen a rough 3-5% increase. But obviously depending on the benchmark or test the results vary, and its extremely hard to keep results consistent as its measuring such a tiny difference and the smallest variable can throw off the results. So with that being said again at a later date Ill post my own findings once i'm not being blown up by the infinite ammount of uni work I have to do.

So my final overclocking specs are as follows:

CPU 4670k: 4.7 Ghz @ 1.35vcore with 45 cache multiplier @ 1.3v, 1.9v CPU input.

Ram G.Skill Trident X 2400Mhz: 1.65v (Stock) 2400 Mhz speed.

CAS: 9, TRCD 11, TRP 11, RAS 30 and CR 1T

I have run out of time for how much tighter I can get those timings but I am going to keep on pushing once uni concludes for this semester.

On a side note, has anybody here achieved a 4.8 GHZ OC with a 4670k I have read so many forums on the internet and read countless reviews, and other posts but for the life of me I cannot figure this out.

I can get 4.8 GHZ @ 1.43Vcore with 40 Cache multiplier @ 1.15v , Input voltage of 1.9-2. With these settings I can boot and post, I ran Aida 64 for 2 hours, I ran Intel burn test on maximum for 5 cycles, I ran Intel Extreme Tuning Utility for 20 minutes, I ran Passmark on 5 different tests, I also ran a benchmark of heaven for one full cycle and it held. Now the problem is when I run Cinebench it will run 2 instances in a row without fail but the third always blue screens my computer. Now I have found Cinebench to be the most accurate for testing stability, as If i crash in cinebench once out of 5 runs, that indicates that If i was to play rome total war 2 (my benchmark for stability game) it will crash within 2 hours. I found this out as I could pass all the above benchmarks with 4.7 Ghz at 1.31v but for it to pass cinebench 5 tests in a row it needed 1.36 and again it crashes with anything up to 1.35 the extra V was for my cache stability.

Now I have adjusted my ram frequency and after numerous tests I can confirm that the XMP profile or higher speed ram makes no difference to my stability. I have even tried raising my voltage from 1.43 where I can run rome for 2 hours stable to 1.48v and it still crashes.

Heat is not the problem I have a custom watercooling loop that keeps everything nice and cool and I monitor with openhardware and real temp and everything stays below 75 degrees.

So I am not sure what I might be able to tweak to hit 4.8 Ghz incase I have missed something? I was contemplating changing the CPU strap to aim for like 4.77Ghz but I haven't found much on the internet explaining how to do this as it gets complicated with its close interaction with the black strap which I don't really want to adjust to ruin my other parts. And I am sorry that, this post kinda hijacks my original post.

Once again thanks everybody for your help, let me know what you think of my timings and if you think I could get them lower which I should adjust, I found that CAS 8 was not achievable at that speed.