4x4GB Samsung 30nm Overclock

[cytoSiN]

I recently grabbed 2 kits of Samsung's low-profile, low-voltage ram. There's been a lot written about how well these kits overclock on their own, but I wanted to see how far I could push them (stably) on my sig rig. Results below. But first, a special thanks to Termie for posting the recent Newegg deal and encouraging me to play around with these things.

So, I played with LOTS of settings to find a sweet spot, but I won't bore you with all the gorey details. Instead, I'll show you the most stable 2133 OC as well as the 1866 OC that I've settled on for every day use (about the same performance as the 2133 OC). Keep in mind that these sticks are rated at 1600/11-11-11-28-2T/1.35v. Also keep in mind that I'm not willing to even try anything more than 1.5v on my rig.

2133/9-10-10-25-2T/1.5v

(I hit PrtSc before everything scaled back to idle, as you can see in the RealTemp window. See the pic below for idle temps/voltages)

I liked this OC, but it was giving me wonky results in Prime95...it would run for an hour stably, and then I'd get a rounding error on one core while the other 3 workers churned away. And it wasn't the same core every time. I tried loosening it up to 9-10-10-28, but same strange results. In my experience, a rounding error will occur soon after starting P95, or it won't happen at all, so this was a new experience for me. I probably could have pushed the DRAM voltage or VCCIO very slightly and gotten away with this OC, but as I said above, not gonna happen on this rig. Anyway, I gave up on the 2133 OC and decided to scale down to 1866, while tightening everything else, and ended up here:

1866/9-9-9-24-1T/1.5v

As you can see, performance is nearly identical to the 2133 OC, but without the wonky P95 rounding errors, so I think I'm settling here for an everyday OC.

Conclusions
This RAM is amazing, even when combining kits. I was stable with one kit at 2133/9-10-10-25-1T/1.5v, and I really didn't have to sacrifice much to get two kits stably OC'ed. I had some issues booting to Win7 at some of my higher tests, but it never failed to POST, with one kit or two, even at some stupid settings like 2133/8-8-8-24/1T for BOTH kits (no Win7 there, unfortunately). I'll be buying more as soon as it goes on sale again, if not sooner (it's only $7 cheaper when it's one sale)! Highly recommended for overclocking, even when combining kits.

Some more pics (taken with Samsung GSIII, so forgive the graininess):

Pre-surgery:

Old RAM:

Old RAM out:

New RAM

New RAM in (it barely sticks out of the slots, and I didn't have to remove either fan on my 212+ to get all 4 sticks in!):

 

[AlexAL]

awesomeness

 

[Termie]

I recently grabbed 2 kits of Samsung's low-profile, low-voltage ram. There's been a lot written about how well these kits overclock on their own, but I wanted to see how far I could push them (stably) on my sig rig. Results below. But first, a special thanks to Termie for posting the recent Newegg deal and encouraging me to play around with these things.

My pleasure! Especially when someone else is willing to test equipment for me!

Also keep in mind that I'm not willing to even try anything more than 1.5v on my rig.

I'm with you on that, man - makes your testing even more useful for me!

2133/9-10-10-25-2T/1.5v

(I hit PrtSc before everything scaled back to idle, as you can see in the RealTemp window. See the pic below for idle temps/voltages)

I liked this OC, but it was giving me wonky results in Prime95...it would run for an hour stably, and then I'd get a rounding error on one core while the other 3 workers churned away. And it wasn't the same core every time. I tried loosening it up to 9-10-10-28, but same strange results. In my experience, a rounding error will occur soon after starting P95, or it won't happen at all, so this was a new experience for me. I probably could have pushed the DRAM voltage or VCCIO very slightly and gotten away with this OC, but as I said above, not gonna happen on this rig. Anyway, I gave up on the 2133 OC and decided to scale down to 1866, while tightening everything else, and ended up here:

1866/9-9-9-24-1T/1.5v

As you can see, performance is nearly identical to the 2133 OC, but without the wonky P95 rounding errors, so I think I'm settling here for an everyday OC.

Conclusions
This RAM is amazing, even when combining kits. I was stable with one kit at 2133/9-10-10-25-1T/1.5v, and I really didn't have to sacrifice much to get two kits stably OC'ed. I had some issues booting to Win7 at some of my higher tests, but it never failed to POST, with one kit or two, even at some stupid settings like 2133/8-8-8-24/1T for BOTH kits (no Win7 there, unfortunately). I'll be buying more as soon as it goes on sale again, if not sooner (it's only $7 cheaper when it's one sale)! Highly recommended for overclocking, even when combining kits.

Some more pics (taken with Samsung GSIII, so forgive the graininess):

snip

New RAM in (it barely sticks out of the slots, and I didn't have to remove either fan on my 212+ to get all 4 sticks in!):

So glad this worked out for you. So just a couple of questions:

(1) Have you tried lower than 1.5v at 1866? I'm betting these sticks are good for it, not that it matters much.
(2) Did you try to increase the memory controller voltage (I think the answer is no), or alternatively did you leave it on auto, which might do it automatically (can you confirm)?
(3) So in your test of 2 sticks on their own, you found a stable 9-10-10-25-1T, where as with two sticks, you had a moderately stable 9-10-10-25-2T? Is that correct? Given that they'd run with 2 at those settings, I really am curious whether upping the memory controller voltage would have gotten you to the same place with 4 sticks.
(4) Did you try setting the sticks at cas 10, 2133? I would have thought that the cas 9 setting is probably what's getting in the way of stable P95 runs. I assume you tried and spared us the results though.

And don't you love that low profile? That's why I used it in an ITX build - there was no way it wouldn't work - if you've got ram slots, these will fit!!! Next time they're on sale, I'm doing exactly what you did, and retiring my 2x4GB Corsair sticks to my HTPC. I was getting out of memory errors in Batman:AC yesterday with 4GB...so much for 4GB being sufficient for modern games.

 

[cytoSiN]

So glad this worked out for you. So just a couple of questions:

(1) Have you tried lower than 1.5v at 1866? I'm betting these sticks are good for it, not that it matters much.
(2) Did you have to increase the memory controller voltage (or alternatively did you leave it on auto, which might do it automatically)?
(3) So in your test of 2 sticks on their own, you found a stable 9-10-10-25-1T, where as with two sticks, you had a moderately stable 9-10-10-25-2T? Is that correct? Given that they'd run with 2 at those settings, I really am curious whether upping the memory controller voltage would have gotten you to the same place with 4 sticks.
(4) Did you try setting the sticks at cas 10, 2133? I would have thought that the cas 9 setting is probably what's getting in the way of stable P95 runs. I assume you tried and spared us the results though.

And don't you love that low profile? That's why I used it in an ITX build - there was no way it wouldn't work - if you've got ram slots, these will fit!!! Next time they're on sale, I'm doing exactly what you did, and retiring my 2x4GB Corsair sticks to my HTPC. I was getting out of memory errors in Batman:AC yesterday with 4GB...so much for 4GB being sufficient for modern games.

(1) I'm going to see if I can run the 1866 OC at 1.4v or lower later this week. I'll follow up here.
(2) I set DRAM voltage at 1.5v. VCCIO is on auto (just over 1.0v -- 1.0067v or something like that).
(3) Yep, that's correct. I suspect that a slight increase to VCCIO would have kept me stable at 9-10-10-25-2T with 4 sticks, but as the AIDA results show, the 1866 OC (9-9-9-24-1T) is nearly identical in terms of performance, so I figured no reason to mess with voltages to force the 2133 OC. If I was able to tighten the 2133 OC to get a decent performance boost, I would definitely try that though. Maybe next weekend
(4) I did try at 10-10-10-28, and several other random permutations, but the 1866 OC performed as well or better than all of the stable 2133 OCs, so I didn't bother posting those screenshots.

I do love the low profile. It's almost laughable how stupid it makes the Ripjaws X look. Don't get me wrong, I actually love the way my blue Ripjaws X looked (especially in my previous Asus mobo, which had blue trim), but they're just so unnecessary, and there was NO way I was getting 4 sticks of that stuff under a 212+ with push/pull. I'm definitely buying more of these Samsung sticks to have handy when I upgrade the RAM in my HTPC, not to mention for any other ITX builds. And as for whether or not I needed to upgrade from 8GB to 16GB, probably not...but I wanted to play with this OC, and you can never have too much for modern gaming, as you've seen

 

[TaranScorp]

Best Ram I ever bought. 16GB
Running 1866 8-9-9-21-1T for many month's now.
1.470V

 

[cytoSiN]

Best Ram I ever bought. 16GB
Running 1866 8-9-9-21-1T for many month's now.
1.470V

Nice. I had issues in Prime95 at CAS8, but I may try again. What other hardware you using, and what CPU overclock, if any?

 

[MontyAC]

Ram looks good. I'll get some when they go on sale down the road.

 

[TaranScorp]

I'm using a Fatal1ty Z68 Professional Gen3 board with an i7-3770K chip overclocked to 4.4.
Hyper Threading on and using the IGP.

 

[Zap]

I love that low profile look! It totally makes RAM with huge heatspreaders unnecessary.

I only have one kit. Should have bought another. Oh well, next time it goes on sale...

 

[chimaxi83]

Nice man! I have one kit of the same RAM, and am running 2133 9-10-10-26-1T at 1.6V. I've had it since January, zero issues. Great kit!

 

[RussianSensation]

Samsung 30 nm Green PC3-12800 Low Profile 1.35 V DDR3
Review @ TechPowerup

This RAM is THE 'budget' ram to buy

Keep in mind higher memory is more beneficial than tight timings on the Sandy Bridge/IVB platform. Thus, DDR3-2400 11-11-11-28-1T is better than DDR3-2000 8-8-8-8 (This was not the case during Core i7 Nehalem/Lynnfield i7 920/860/i5 750 era where timings mattered more).

Xbitlabs has the scoop:
DDR3-1333 9-9-9-27-1N;
DDR3-1600 9-9-9-27-1N;
DDR3-1867 9-9-9-27-1N;
DDR3-2133 11-11-11-33-1N;
DDR3-2400 11-11-11-33-1N;
DDR3-2667 11-13-13-35-1N.

"DDR3-1867 with 9-9-9-27 timings turns out to have a better effective latency than DDR3-1600 with aggressive timings of 7-7-7-21"

What makes this ram so amazing is that it's able to scale to DDR3-2133-2400. Since the memory speed benefits SB/IVB platform more than latency, this makes it a remarkable value for $47.

 

[Coup27]

It would be nice if AT could do a feature on this RAM. It's flying in the face of current RAM with their massive heatspreaders, full height and up to 1.65 volts!

 

[davidthemaster3]

I have 4 of these in my rig too. I'm running them 2000Mhz @ 10-10-10-28 1T @1.45V ... but I think I can put CAS 9 in there, but I'll have to recheck cause I flashed my BIOs recently. (Asrock extreme 4 z77).

EDIT: Found a thread where cmdrdredd posted his timings (he's using 2x 4GB sticks though)

Benchmark at my current setting of DDR3-2133 9-10-10-24 CR1 1.475v. Think I'm gonna leave it here 24/7 as it seems to be a good compromise between speed and timings.

 

[cytoSiN]

I have 4 of these in my rig too. I'm running them 2000Mhz @ 10-10-10-28 1T @1.45V ... but I think I can put CAS 9 in there, but I'll have to recheck cause I flashed my BIOs recently. (Asrock extreme 4 z77).

Yeah I'm pretty sure they'd behave at 2000, but my motherboard only likes multiples of ~266. I got them stable at 1866-8-9-9-21-1T last night as well. Results were microscopically (technically nanoscopically, I guess, since we're dealing with latency in nanoseconds, but you get the idea) better than the 1866-9-9-9-24-1T results, but then I went back and got the cas9 settings and lowered the voltage to 1.42, so I'll probably stop there. Same AIDA64 results as above, but with slightly lower voltage.