OC expectations with DDR3 1866 C9 2x8gb CMZ16GX3M2A1866C9

[carbide]

Hi there!

Just started playing around with overclocking my memory. I've set the voltage to 1.65 and been using off the shelf timings to go with some additional mhz jumps - 2000, 2133, 2400 all look fine and stable.

My question is, what are the rule of thumb boundaries? if I keep the timings very loose, can I just keep going? assuming I don't change the voltage, will I soon reach the ceiling by which the sticks cannot function at that high frequency?

Are temperatures a big concern with ram when going for higher frequencies? I have a chipset cooler blowing over them, but have no idea of when enough is considered enough.

Thoughts, appreciated...

 

[10tacle]

I have never seen any real world performance boost with overclocking memory beyond their rated speed (games and video encoding specifically). With that said, heat is not going to be a major issue with a decent air flow across them in the case (and they have a decent heat sink design like your Vengeance).

It's like overclocking anything else: go as high as you can go without crashes. Keep in mind that even if you get it "stable" it doesn't mean it won't corrupt registry files or a program running while overclocked. It's a gamble when overclocking anything, but too much of a good thing yields diminished returns and shortens the life of the hardware. Your hardware, your roll of dice over and over going higher and higher. Just hope you don't roll a craps and destroy your memory.

As an interesting perspective, overclocking memory on video cards shows FPS improvement in games. It's all about the hardware application being overclocked.

 

[Karadjgne]

I have 16Gb 1866MHz 10-11-10-30 1T. Default being the usual 9-9-9-27 @1600/1.5v. It'll sit just fine at 2400MHz 10-11-10-30 1T at 1.58v but goes no further, even bumping SA agent, ram to 1.65v or relaxing timings as far as 12-13-12-38.

Theres no hard set boundaries for any OC, whether it's gpu, cpu or ram, the only boundaries are set by the ram itself. It's silicon and therefore bound by the silicon lottery.

As far as I've personally been able to tell IRL, there's no difference between 1866MHz at 9-9-9-27 and 2400MHz at 10-11-10-30, other than the slightly faster setting of 2400 means the windows logo on boot doesn't quite fully show, it's cut short I get 3 colors up, with the 1866 at 9-9-9-27 it starts showing the 4th color before ending and at regular XMP 1866 10-11-10-30 I get all 4 colors. Grand total savings on boot is about 1 second. Every game runs the same fps, web pages take the same amount of time etc etc..

When Intel decided to move the memory controller to the cpu and off the board, for almost all apps/programs timings became pretty much a moot point as now ram had instant access to the cpu, and ram itself is usually sitting partially idle, just waiting for the cpu to get around to using what's in it. So burning out ram trying to push limits basically does nothing for 98% of programs on newer than lga775 architecture.

Best bet, set the 1866 as XMP and forget about it being there. It's less volatile at normal 1.5v than running hard at 1.65v for no IRL gains.

 

[carbide]

I have never seen any real world performance boost with overclocking memory beyond their rated speed (games and video encoding specifically). With that said, heat is not going to be a major issue with a decent air flow across them in the case (and they have a decent heat sink design like your Vengeance).

I neither, so far. I thought 2133 at 9-10-9-27-2T made programs feel more, 'snappy' to open, but it could well have been placebo.

I was under the impression Dram would be like Vram - 80-90c max operational temps, so pushing to a 'certain' level would be ill advised.

recently found this: http://www.ocdrift.com/review-corsair-dominator-platinum-ddr3-1866mhz-c9-8gb-memory-kit/

Which, although a better binned set of sticks, gives me something to go on, although I see if I were to be able to get C8 at 2133 it would perform close to C10 2400

It's like overclocking anything else: go as high as you can go without crashes. Keep in mind that even if you get it "stable" it doesn't mean it won't corrupt registry files or a program running while overclocked. It's a gamble when overclocking anything, but too much of a good thing yields diminished returns and shortens the life of the hardware. Your hardware, your roll of dice over and over going higher and higher. Just hope you don't roll a craps and destroy your memory.

Well, yes, for a system I need stability for, for gaming and for work, any instability is unacceptable. I just thought, provided it was aida64 stable, it could be considered as stable as the factory XMP settings that the kit ships with?

As an interesting perspective, overclocking memory on video cards shows FPS improvement in games. It's all about the hardware application being overclocked.

True, giving my memory a +150 bump and core +100 bump on my 1070, I have seen instantly tangible benefit in FPS

 

[carbide]

... It's less volatile at normal 1.5v than running hard at 1.65v for no IRL gains.

Interesting point I'd not read or considered - I assumes as the operational voltage was between 1.5v - 1.8v the level the voltage was at was inconsequential to performance, provided that there was enough for the speed/timings it was set to.

Are you saying there is a higher chance of failure/corruption with a higher voltage, or just that, with a higher voltage the settings are often tighter/faster?

 

[Karadjgne]

Older ddr3 was made to be compatible with the lga775 boards so was initially 1.65v but that could be raised as high as 1.8v because the memory controller could handle that voltage. Newer ddr3 made the swap to lga1156/1155 but in turn the voltages were lowered for the MC to a standard 1.5v. But the MC could still use 1.65v as that was also available in the exact same models. Taking 1.5v ram upto 1.8v is ill-advised for any lga11xx series cpu, it does nothing but heat up the cpu considerably since that's where the memory controller is now located.

And as with anything to do with electronics, raising voltages beyond stock will always have a greater chance of failure or corruption. The tighter the timings, the harder/faster the ram works on the i/o. To compensate for that, often you'll need to bump voltages. But bumping should be the minimal amount needed to maintain stability. No different than OC of a cpu. You don't bump vcore to 1.4v just for a 3.4 to 4.0 OC. Very stable, huge heat.

 

[carbide]

I thank you for explaining the operational voltage variation of DDR3 - something I'd still yet to find out anywhere other than 'it's just rated that high'

I originally purchased this kit because it was rated at 1.5v - after reading the intel haswell datasheet, page 103 table 49, it states VDDQ - Processor I/O supply voltage for DDR3/DDR3L as 1.5v +- 5% so I always assumed 1.575Vmax.

However, at the time haswell (devils canyon in my case) was the bees knees, everyone on here, reviewers etc. were using performance DDR3, which, 95% (unless you could the modules like mine) simply enabling the XMP profile would set the voltage at 1.65v.

Had a bit of an epiphany last night, thinking about this thread, trying to get the ram stable, I decided to drop the CPU overclocking method I've used for the last few years for something simpler. I dropped my offset voltage (+0.040v which would sometimes boost up to 1.3v for AVX) in exchange for a fixed voltage - went from 1.3v to 1.25v under load. Dropped VCCIN from 1.8v to 1.75v which still allows a droop down to 1.72v (intel minimum operational voltage is 1.65v) set Vtt/Vring/cache from 1.18v to Auto.

All seems stable, idle 5c higher, load temps 10-15c cooler.

For the ram, had a scary moment where I though I'd corrupted windows sailing too close to the sun (2400 10-11-11-25-1T) but I've backed off to 2400 10-12-12-31-2T which seems to be CPU/FPU Aida64 /realbench stable so far.

I suppose all that really remains is to try to drop the Dram voltage back as far as i can

 

[Karadjgne]

Setting xmp on my ram puts it at 1866MHz 1.5v, doesn't raise the voltage to 1.65v. So that's a little odd.
My wife's pc has a 1.65v kit, and default is 1600/1.65v xmp is 1866/1.65v. That's fine for Ivy-Bridge cpus. For Haswell and newer I'd be trying for 1.5v or close, just to give the MC a break.

 

[carbide]

well, 1.6v seems to be totally stable, which I wasn't expecting straight off the bat!

I understand setting timings too tight can damage windows if they're unstable, same with dropping the voltage? or just cause a crash like too little CPU voltage settings?

 

[Karadjgne]

Any or all of the above. Data goes through the ram, one little bleep and the game crashes, or OS gets corrupted, causing a bsod or even failure to boot. Ram is kinda touchy about things, if it's not stable you'll know soon enough.

 

[carbide]

Karadjgne - have you used Realbench v2.56 before? Only ask as it seems to be the best (whether it was really intended or not) stability test I've found so far. windows memory check tool didn't detect any errors, but this program, particularly the 'multitasking' part of the benchmark can BSOD setting even you thought were stable for some time...

back to 1.65v Dram, any lower, no dice

 

[Karadjgne]

Yes, I've used (do use) Asus RealBench. There's a lot of misconception regarding many torture tests, and ppl get it in their heads they are doing it right, because it's a stress test. Occt, Aida 64 and many others are all valid stress testers, as far as stability goes, but ppl seem to use it as temp tests as well. They aren't. Only Prime95 small fft should be used for temps. The others use linpack, which uses your ram to varying degrees, so you don't get a clean 100% cpu load. On p95 small fft does that. On the other side of the coin is RealBench, which uses everything, not just cpu and ram, but the cpu in different scenarios, blender of various lengths, gpu, ram, basically the entire pc. It's very good for stability as it'll hit all the major and most minor instruction sets that you don't normally see, until you do, and bsod.

Prime95 v26.6 and prior, small fft. Gaming temps for any cpu. More important for Haswell and prior.
Prime95 v26.7 and later, small fft. Includes AVX instructions, so adjust AVX offset for Intel newer than Haswell, by - 2 (200MHz) under OC, to prevent cpu overheats. Games really don't use AVX instructions but some applications can.
RealBench. Best overall stability.
Aida 64, Occt, IBT and the rest. Use with caution.