Safe ddr3 voltage?

[SickBeast]

Hi guys, I need help overclocking my memory. I have it at 2400mhz and it's not quite stable at 1.65v. Unfortunately the next voltage step in my bios is 1.75v. Would it be safe to run my RAM at 1.75v for daily usage? My CPU is a haswell 4690k. Thanks!

 

[VirtualLarry]

What are you doing on your PC that is so memory bandwidth intensive, that you are willing to overclock and overvolt your RAM to the point of failure?

 

[SickBeast]

I game a lot and I find that my computer is a lot more responsive with the memory overclocked.

 

[BonzaiDuck]

And the RAM make, model and spec speed, voltage and timings?

You should really see what happens if you underclock the RAM so that you can drop the voltage, and then change the command-rate to 1 at stock timings. Or there might be a balance of speed, voltage and tighter timings for which the command-rate setting is still in the cards.

 

[SickBeast]

What does the command rate do? The memory is pny 1866 cas 10 1.5v. It won't do cas 9 at stock voltage. It is almost stable at 2400mhz at cas 11, 1.65v.

 

[thetuna]

I think the general consensus on 'safe' dram voltages is within .5v of your uncore/vtt voltage.
And uncore really shouldn't go above 1.35v.
On the topic of uncore, have you tried raising that to get some more stability?

Personally, I would never run my dram above 1.65v for long term.

At the moment, due to a combination of not-too-great ram and the base clock I'm running at, my ram is at 1150MHz 9/9/9/24 @ 1.4v. (The next ratio up is 1533MHz, and this ram won't go that high.)
I notice literally no difference between this and when it was running at 1400+ 9/9/9/24.

 

[BonzaiDuck]

What does the command rate do? The memory is pny 1866 cas 10 1.5v. It won't do cas 9 at stock voltage. It is almost stable at 2400mhz at cas 11, 1.65v.

If you want higher than 1866, then try 2133.

You know! . . . It's not going to make a lot of difference, and if you wanted to run RAM at that speed, it would've been more cost-effective to buy RAM rated higher than 1866. There isn't that much price variation to make memory overclocking much worth it anymore.

But the command-rate -- the number of clock-cycles need to execute a memory operation or command -- can give a little speed-boost without raising voltage or changing the timings. It COULD require a slight increase in the VTT/VCCIO or "System Agent" voltage for the integrated memory controller, but for some makes (G.SKILL) -- not much. At their spec speed, voltage and timings, I have a 4x4 set of Corsair XMS3 with my current VCCIO setting at only 1.075V, and I think I can drop that another 0.030V. The stock value is somewhere close to 1.005 to 1.009V. For whatever overclock on the CPU and the number of modules, there is a sweet-spot for the IMC voltage neither too high nor too low, but the lower it will run, the better.

I had used benchmarks before to assess the impact of CMD=1 for both DDR2 and DDR3 RAM. With my G.SKILLs, my ASUS Z68 and Z77 motherboards automatically override the XMP settings and configure under "Auto" to a 1T command-rate. The result certainly shows up in the benchies. Although I could be experiencing a "Placebo effect" to some extent, I nevertheless thought that I could "feel" the difference.

Back in the day of DDR and DDR2 when I had to RMA a pair of Ballistix, a Crucial tech-support rep told me "you're not supposed to run them at 1T. It's outta spec! You'll invalidate your warranty . . yada-yada-yada." But the exchange occurred at a time when Crucial was having a barrage of RMA returns because (perhaps) they'd spec'd the VDIMM voltage warranty limit too high.

I'd had a good e-mail exchange going with G.SKILL tech-support on several occasions with DDR2 and DDR3. Their reply on the command-rate question was something like this: "Oh, yeah! You should be able to do that. You shouldn't need to bump up the VCCIO very much! [if at all]." The implication being that you needn't change the VDIMM voltage, and their Ripjaws or other kits were perfectly capable for it.

 

[SickBeast]

Thanks for your input guys. I'm not going to push it past 1.65v. I will let you know how I do.

 

[Soulkeeper]

Just a little input here:

"According to JEDEC,[3]:111 1.575 volts should be considered the absolute maximum when memory stability is the foremost consideration, such as in servers or other mission-critical devices. In addition, JEDEC states that memory modules must withstand up to 1.80 volts[a] before incurring permanent damage, although they are not required to function correctly at that level"

It's been awhile since this jedec spec, and the process tech has only strank more since.
So if you're running a recently purchased set of ddr3 i'd really caution against increasess above where you're at now.

It really amazes me that they sell these factory overclocked kits at 1.65v using chips that are essentially the same as used on the ddr3l.
I know it's apples and oranges, but 30/20nm class chips (on the newest ddr3/ddr4) with less cooling than a cpu die and much higher voltages it's insane.

 

[SickBeast]

Wow maybe I should just stick to 1.5v.

 

[Soulkeeper]

I'd run my kit under 1.35v if my bios would let me, the voltage don't really help much.
fiddling with the timings helps a lot more

 

[SickBeast]

Thanks Soulkeeper. I have it back at stock speed. It's not worth killing my CPU or my RAM for the small performance boost.

Hey guys is it normal for my CPU temps to go up after upgrading to dual channel memory? I had a single stick before. My temps have gone up almost 10C! Do you think I might have damaged my CPU? I did test the memory at 1.75v for about 10 minutes earlier today.

 

[thetuna]

Honestly, you don't need to worry about killing anything as long as you keep your voltages and temps within the realm of sensibility.

This thread might be helpful for you: http://www.overclock.net/t/1411077/haswell-overclocking-guide-with-statistics

As far as adding more ram, yeah that could have an effect if your CPU was memory bandwidth starved before.
It also typically takes a disproportionately higher amount of power the more memory ranks you fill up.
If you go from barely stable with 2 sticks to 4, that could put you over the edge of stability.

That said, 10 C does seem like a big increase.
Maybe the weather just got hotter?

 

[Dasa2]

im running 1600c11 @ 2133 10-10-10 1t
i can tighten it to 9-10-10 with a .1v increase but its only borderline stable at that so i dont bother other than for benching

if your running games that see a improvement from overclocked cpu\ram then i think its worthwhile

1.65v ram is a fairly common 24/7 v for a overclocked system so i wouldnt worry to much about that but higher probably isnt a good idea for 24/7 use

1.5v is the better choice for stock clocked work systems

2133c9 is faster than 2400c11 so see how you go

 

[SickBeast]

Well it turns out that my system actually was stable with the ram at 2400mhz, my CPU was overheating. It is a very hot day here so perhaps that is why. I don't have very powerful cooling so I just backed my CPU down to 4.2ghz at 1.1v. Now it barely touches 70C under full load.