how do I OC ram?

[omghaxcode]

So I have that samsung ram everyone talks about (low profile 13cas). I have no idea how to OC it though. Can anyone point me to some advice?

 

[pantsaregood]

If the system in your signature is the one in question, just raise your FSB. RAM speed is directly related to FSB.

 

[omghaxcode]

no I'm sorry, it is i7 3770k, z77x-ud3h, samsung MV-3V4G3D/US

 

[2is]

In a nutshell, you don't... More trouble than it's worth. The gains are only seen in benchmarks, even then they are marginal. It puts a lot more stress on the IMC as well as the modules themselves (particularily if you need to overvolt them) and you're adding an extra variable when you are stress testing/troubleshooting instability.

 

[BonzaiDuck]

no I'm sorry, it is i7 3770k, z77x-ud3h, samsung MV-3V4G3D/US

The Samsungs are "rated" DDR3-1600, 1.35V at the spec timings. This means as much that they simply chose to "test" a portion of product off the assembly line at that voltage so they could sell them at that speed.

However, if you run a search under "memory and storage" for posts by "CmdrDredd," he shows how he put the Samsung sticks through their paces.

I think -- from what he posted -- you can boost the voltage to only 1.475V with speed set to maybe DDR3-2133 and the same latency settings.

Main thing -- if you can get the system to post at the higher speed, you should be able to tweak the latencies and voltage to make it stable.

You may want to try them first at DDR3-1866, same voltage, possibly CAS = 9 or 10. I still suggest that you look for those recent threads here -- get as much information as you can.

In a nutshell, you don't... More trouble than it's worth. The gains are only seen in benchmarks, even then they are marginal. It puts a lot more stress on the IMC as well as the modules themselves (particularily if you need to overvolt them) and you're adding an extra variable when you are stress testing/troubleshooting instability.

Supposedly the processor IMC is designed to handle the RAM, even for speeds beyond 1866 or 2133. The biggest risk is the voltage, which -- in these Samsungs -- is particularly low. The VCCIO termination voltage must be within 0.5V of the RAM, and probably "safe" up to 1.2V. So there's extra "wiggle room" with these Samsung modules.

There are other indications that the higher speeds can reap a 30% advantage for certain tasks like rendering, but not for other normal computer usage. My personal observation is that you can notice a "snappier" response with the higher speeds, even though the benchmark tests are "synthetic."

 

[pantsaregood]

Sandy Bridge IMC is only rated for four DIMMs at DDR3-1333. SB-E is rated for eight at DDR3-1600. Ivy Bridge is rated for four DIMMs at DDR3-1600.

Anything more than that and Intel isn't responsible for any instability that comes with higher speeds/more DIMMs. I have a 2500K at 4.4 GHz pushing 2x4GB DDR3-2133 at 10-11-10-26-1T, though. Completely stable at 1.572v - it may be stable at lower voltage, but the RAM is rated for up to 1.6v, and the Sandy Bridge memory controller is rated for up to 1.575v.

Running RAM at higher speeds isn't going to damage your CPU or RAM, but it could lead to instability. The performance gains are minimal, as well. They can give decent improvements to synthetic benchmarks (AIDA64 memory bench went from ~18,000 MB/s to 25,000 MB/s when comparing DDR3-1600 to DDR3-2133 for me), but the difference in practice is minimal.

 

[BonzaiDuck]

Sandy Bridge IMC is only rated for four DIMMs at DDR3-1333. SB-E is rated for eight at DDR3-1600. Ivy Bridge is rated for four DIMMs at DDR3-1600.

Anything more than that and Intel isn't responsible for any instability that comes with higher speeds/more DIMMs. I have a 2500K at 4.4 GHz pushing 2x4GB DDR3-2133 at 10-11-10-26-1T, though. Completely stable at 1.572v - it may be stable at lower voltage, but the RAM is rated for up to 1.6v, and the Sandy Bridge memory controller is rated for up to 1.575v.

Running RAM at higher speeds isn't going to damage your CPU or RAM, but it could lead to instability. The performance gains are minimal, as well. They can give decent improvements to synthetic benchmarks (AIDA64 memory bench went from ~18,000 MB/s to 25,000 MB/s when comparing DDR3-1600 to DDR3-2133 for me), but the difference in practice is minimal.

"Could lead to instability." I would think that one would eliminate this problem from the beginning -- by doing adequate stress-testing and adjustment. Anyway, I'm not going to be able to use my warranty with Intel because I lapped the i7-2600K. I lap all my processors.

 

[ShintaiDK]

Unless you really must OC for the plain sake of OC. Then stay at 1600Mhz like it is as default.

Else raise the voltage for the memory slightly. And change the memory multiplier.

 

[BonzaiDuck]

Unless you really must OC for the plain sake of OC. Then stay at 1600Mhz like it is as default.

Else raise the voltage for the memory slightly. And change the memory multiplier.

When Intel over-clocking was done with locked CPU multipliers through changing the FSB, CPU and RAM voltages, this was done all the time. It became a balancing act to get the RAM to run at the highest speed with the most efficient ratio -- ideally 1:1. You couldn't avoid it.

Now, it's a different situation. You can OC the processor and leave the RAM alone, or attack the RAM as a separate project.

For costs in time and effort to achieve slivers of performance improvement, it would seem almost unpractical. Want to do the 1000% coverage tests on 16GB RAM with HCI Memtest? 400% took me twenty hours before the test failed. It becomes a temptation to set the voltages higher than they need be, just to get the desired confirmation of stability without days of waiting to see the result fall short of "stable."

But I'm not an oddball on this forum, for "OC for the plain sake of OC." I do, however, believe in setting reasonable goals to build a stable overall system. Wild extremes are for contests and deep pockets.

Even so, you're going to test the RAM whether you OC it or not, and there are different approaches to choosing RAM. You could, if you wanted, buy more expensive RAM rated at a higher speed and voltage, and underclock it to a speed still higher than "1600." That might save some time in testing for higher speeds. Or, you can count on a little common sense, results of others etc. and a little luck to overclock the 1600 RAM.

The high-performance RAM generally relies on higher voltages and looser timings, anyway. When you find a kit that has especially tight timing specs, that advantage can be exploited -- for the price, that is.

 

[Termie]

This thread might be interesting:

http://forums.anandtech.com/showthread.php?t=2256637

BTW, this question really should have been posted in the "Memory and Storage" forum.

 

[BonzaiDuck]

This thread might be interesting:

http://forums.anandtech.com/showthread.php?t=2256637

BTW, this question really should have been posted in the "Memory and Storage" forum.

There are people there who seem a tad naive about over-clocking . . . or so I had that impression per some thread or interaction there.

I'd have to confirm with some additional investigations per the difference between SB and IB on this issue, but it seemed that IB can use memory (overclocked or spec'd) at speeds above 2133, while SB can't. There may be some others here who know with greater clarity . . . .

 

[Termie]

There are people there who seem a tad naive about over-clocking . . . or so I had that impression per some thread or interaction there.

I'd have to confirm with some additional investigations per the difference between SB and IB on this issue, but it seemed that IB can use memory (overclocked or spec'd) at speeds above 2133, while SB can't. There may be some others here who know with greater clarity . . . .

That may or may not be true, but what the OP will see is both the link to the Xbitlabs article, as well as cmdrdredd's very helpful analysis of his own experiment with Samsung 30nm overclocking. He runs at DDR3-2133 10-10-10-28 CR1 1.475v, but showed that raising the voltage and running at 2400 would probably be the sweet spot for OC'ing, if one were into that kind of thing.

 

[BonzaiDuck]

That may or may not be true, but what the OP will see is both the link to the Xbitlabs article, as well as cmdrdredd's very helpful analysis of his own experiment with Samsung 30nm overclocking. He runs at DDR3-2133 10-10-10-28 CR1 1.475v, but showed that raising the voltage and running at 2400 would probably be the sweet spot for OC'ing, if one were into that kind of thing.

No disagreement there. I was riveted by Dredd's expose' on those Samsungs. But we're talking about the sweet spot for the Ivy -- which I'm not using at the moment.

I came back to this topic -- having built my i7-2600K system almost exactly a year ago. Whether then or now, I couldn't (or can't) shake the feeling that running at a 1T command-rate -- IF STABLE -- "seems" faster or just "is" faster. Another guru here was running the same G.SKILLs at 1866 -- two kits worth -- 1T command-rate. I got sidetracked after buying a second kit for $50 back in December or January. I finally socketed the second kit about four or five days ago.

I had actually considered spending $100 on a 2x8GB kit, but apparently the four-stick configuration is not the disadvantage it used to be.

It seems the upper bound for worthwhile memory speed on the Sandy is about 1866. It's the Ivy Bridge for which the higher speeds pay off, making the Samsungs very promising.

 

[Diogenes2]

What exactly is the payoff?

 

[BonzaiDuck]

What exactly is the payoff?

It was an article somewhere . . . TechPowerup, Anandtech -- can't remember. Although I came across it during the last week.

Certain tasks, such as rendering, see an improvement of 30%.

Many other mainstream computer uses wouldn't see it.

For me, with this "fine-tuning" exercise (inconvenience, waste of time, etc.) -- I figure a 7% improvement just for CR=1. The synthetic latency benchmark shows as much as a 10ns improvement for the extra 266 Megahertz.

For me, it's important to "stay connected" and have a reliable machine. But I'm quite willing to over-clock that "reliable machine." My experience tells me that it's best to get all the tweaking done at the same time, and before the new computer is first used for any practical purpose. I don't much like making hardware changes after that point. Ideally, tweaking the RAM or adding the second kit should've been done last year.

On the Ivy Bridge "question," I want to see what is done for reducing the temperatures. And I'm not going to flash my BIOS on this system to accommodate IB until I've reached that decision-point, which is not currently a pressing need.

So what are your thoughts on this, Diogenes?

 

[Termie]

What exactly is the payoff?

http://www.xbitlabs.com/articles/memory/display/ivy-bridge-ddr3.html

 

[2is]

I only did a quick glance so perhaps i missed it but I don't see a 30% gain in anything that isn't a static benchmark.

 

[Termie]

I only did a quick glance so perhaps i missed it but I don't see a 30% gain in anything that isn't a static benchmark.

WinRAR, 33% improvement going from DDR3-1333 to DDR3-2666. Also, 3dMark11 physics increases 14%, which is fairly substantial, easily equal to the difference between the 3570 and 3770.

 

[Diogenes2]

..

So what are your thoughts on this, Diogenes?

My thoughts are that the expense, time and effort of running memory higher than 1600 ( or 1333 for that matter ) on SB/IB is not worth the results.

That includes the time spent running benchmarks and discussing it on e-forums.. :biggrin:


Now, if I had a dedicated WinRARmachine, or got paid to run SuperPi, I might feel differently about that ..

 

[2is]

WinRAR, 33% improvement going from DDR3-1333 to DDR3-2666. Also, 3dMark11 physics increases 14%, which is fairly substantial, easily equal to the difference between the 3570 and 3770.

I saw that, but I was comparing it to 1600 since thats what the OP has, which is just under 20% in winrar which looks to be the best-case real world test.

 

[BonzaiDuck]

My thoughts are that the expense, time and effort of running memory higher than 1600 ( or 1333 for that matter ) on SB/IB is not worth the results.

That includes the time spent running benchmarks and discussing it on e-forums.. :biggrin:


Now, if I had a dedicated Winrar machine, or got paid to run SuperPi, I might feel differently about that ..

You may be right, there. The OP wanted the real dope, and he probably got more than enough. For me, personally, I've spent a lot of time chatting it up about this, but I think I'm going to save some money . . . . by no longer even considering purchase of another memory kit for this box . . . . OF course, I still have some stress-tests to run, now that I got "from there" to "here."

 

[Termie]

Well, the irony of all this is that the Samsung 30nm memory that the OP has actually costs less than practically any other memory available on the market today, and yet can easily run at DDR-2133.

I just bought some for a mini-ITX non-OC build that will run at DDR3-1333. Why? Because it was cheap. If you haven't read up on why this RAM is basically the best choice in every situation, you really should.

 

[BonzaiDuck]

Well, the irony of all this is that the Samsung 30nm memory that the OP has actually costs less than practically any other memory available on the market today, and yet can easily run at DDR-2133.

I just bought some for a mini-ITX non-OC build that will run at DDR3-1333. Why? Because it was cheap. If you haven't read up on why this RAM is basically the best choice in every situation, you really should.

No doubt about it. I've seen some reviews, and CmdrDredd's OC and benchie results. All kinds of advantage: low voltage and doesn't need heatspreaders; low profile; variety of speeds/latencies; not expensive.

Somebody was "wishing" that they had an 8GB or 2x8GB (kit) version, but I don't think they do. Even so, with the new CPU design/IMC etc., running 4 instead of 2 doesn't pose nearly the disadvantage it used to.

Like I said -- I picked a 2x4GB kit of G.SKILL GBRL 1600's back in June, '11. Bought a second of same around December/January, and only got arround to installing them last week. I can't justify buying the Samsungs since I already have these and they'll scale to a setting that at least can be used by a Sandy Bridge.

 

[Diogenes2]

I just bought these for my X79 .. ( 16GB )

http://www.newegg.com/Product/Produc...82E16820226184

I'll see what 1.5v will get me, and how it affects performance running a couple of applications I use a lot.. Handbrake and Folding@Home...

 

[BonzaiDuck]

I just bought these for my X79 .. ( 16GB )

http://www.newegg.com/Product/Produc...82E16820226184

I'll see what 1.5v will get me, and how it affects performance running a couple of applications I use a lot.. Handbrake and Folding@Home...

Nice. I don't say that for the "heatspreaders," but the price is right, the timings are right, and the voltage is right. Lemme see what those "customers" say . . . Again, for me, the lower voltage [assuming same or better scalability] would be a plus. But I don't think speeds above 1866 will be of great benefit to my own rig.

Those should be great if you just bought them. You'd have to compare results at different settings to see if they're as good as the Samsungs. The customer experiences didn't leave me with full confidence on that point, but don't rule it out, either. Anyway, you're using the Sandy "E" . . . I wouldn't know how socket-2011 and quad-channel changes the ball-game much . . .