DDR3 1600@7-7-7-19 vs. 1866@9-9-9-24

[Edrick]

I have been running at 1600 since I put this system together. I always thought 1600@7-7-7 would be better than 1866@9-9-9. I decided to start trying out new settings and this is what I found to my surprise. But hopefully this will settle some of the questions as to which is better, speed or CAS.

Same RAM - CPU - MB - OC Settings - Voltage - Etc.
Only thing I changed was RAM speed and CAS settings.

 

[MrK6]

Great benchmarks! However, I've found that higher bandwidth doesn't always equate to better performance. For example, my sticks at 1568MHz 6-7-6-15 are faster in Super Pi compared to 1960MHz 8-8-7-21. Test out some applications to see what the actual performance difference is :thumbsup:.

 

[Spikesoldier]

pretty cool, good too see that sb can take advantage of faster memory frequencies.

do other benchies look better too?

 

[Edrick]

SuperPI had a very slight advantage to the 1866mhz setting. But it was so close that I would hardly call it an advantage (half second). My Win7 score went from 7.8 to 7.9 after going to 1866. That was confirmation enough for me to keep my settings at 1866.

I can get my Ram to 1866 @ 8-8-8-24, but I have to crank up the voltage to 1.66v. At this speed I do see more of an advantage, but I do not want to run at that voltage 24/7.

 

[Spikesoldier]

i think youre ok at 1.66 vdimm.

1866 @ cas 8 would be blazing.

 

[Dadofamunky]

Doesn't look like much of a statistical difference - 1-2% either way. Any visible improvement? Seems to me the CAS9 balances out the faster access time vs the CAS7 and slower access time.

 

[Blain]

I decided to start trying out new settings and this is what I found to my surprise.
But hopefully this will settle some of the questions as to which is better, speed or CAS.

Sorry, but since your testing was nowhere close to being exhaustive, the questions remain.

 

[Rubycon]

Six of one vs. half dozen of another...

RAM is only going to work at its limit and a lower speed with lower CAS vs. higher speed with higher CAS work out very close. The latter is easier on chips with IMC as uncore is 2X (1.5X for Gulftown/Westmere chips).

The most important thing is to make sure it's stable!

 

[MrK6]

The latter is easier on chips with IMC as uncore is 2X (1.5X for Gulftown/Westmere chips).

Now I'm curious . What exactly is "easier on chips" by running higher speed and higher CAS? Less voltage needed?

 

[Blain]

Now I'm curious . What exactly is "easier on chips" by running higher speed and higher CAS? Less voltage needed?

I think you're misunderstanding the statement.

 

[MrK6]

I think you're misunderstanding the statement.

Then how so?

 

[Blain]

I won't presume to speak for Dawn, so I'll let her explain her statement.

 

[Rubycon]

Now I'm curious . What exactly is "easier on chips" by running higher speed and higher CAS? Less voltage needed?

My mistake! I meant former instead of latter.

Increasing memory speed pushes the uncore frequency higher. Most chips will not be stable at the required 3,732 MHz that DDR3-1866 needs. With DDR3-2000 the problem is worse - most chips are going to need a serious boost in QPI voltage to be stable there. Personally I won't go past 1.4V on a continuous basis. At least with Westmere chips the DDR:uncore ratio is 1.5:1.