RAM deteriorating under Over-clock settings?

[BonzaiDuck]

I was very pleased when I built my Sandy Bridge system in mid-2011 -- with the G.SKILL DDR3-1600 "GBRL" Ripjaws kit. I started with 2x4 8GB.

For the price, I upgraded to 4x4GB of the same "GBRL." Discovered through one of our colleagues here on the forums that you could fill all four slots and STILL OC the RAM to 1866 with looser timings -- same voltage at the 1.5V spec.

I've been experiencing random restarts and occasional BSODs every 5 to 7 days with the machine running 24/7. Trying to troubleshoot this, I didn't panic at first because it happens when the machine is only being used for Media Center at the EIST speed -- with no other apps open. The occasional CHKDSK-at-reboot hasn't shown any disk corruption, but I know how this can play out -- eventually.

So I finally reset everything back to stock settings. I am now more and more inclined to think that the RAM was somehow failing at idle under the 1866 settings. I need to wait a few more days to be absolutely sure, but I'd had these random restarts occur more frequently over the last couple weeks while fiddling with settings that would only improve system stability otherwise. Now that the RAM is running under its spec XMP settings, it looks OK -- for now.

I'm inclined to purchase a set of 2x8GB G.SKILL sticks that run at spec 1866 and use the same timings and 1.5V spec as the 1600's.

Of course, I can test the old sticks in pairs and then as one-at-a-time.

I'm just wondering if it's possible that the sticks are going bad only for the overclock setting, or how exactly they go south. An HCI-Memtest on the whole 16GB set of four will take close to five days. I can probably use the Windows version of HCI to run the tests alongside "regular" usage. But -- it's still a B****.

 

[ShintaiDK]

You overclcoked the memory and started the accelerated degrade. You can try relax the timings and/or increase the voltage. But its just how things work.

 

[BonzaiDuck]

You overclcoked the memory and started the accelerated degrade. You can try relax the timings and/or increase the voltage. But its just how things work.

Thanks for your response, ShintaiDK. I won't argue with you about "what these modules were capable of" or any of that. I think I spoke to a G.SKILL tech-rep when I put the second kit into the mobo maybe 18 months ago. He didn't say "don't overclock them" but seemed to think they might be capable of that -- including a CMD=1 setting.

The voltage on the RAM had always been close to the 1.50V spec. So I think you're saying that the speed setting might have caused them to deteriorate? [Or -- for that matter -- the choice of command-rate.] I think the cost to me for the two kits was $100 to $120 total -- over 30 months.

I'll probably find out before middle of next week if setting them back makes a difference. I could run them through an HCI-MEmtest, but the 1000% "thorough" testing will take too long. Probably -- attempt using the Windows feature of HCI.

It would almost seem easier to just buy a 2x8GB kit and set the old ones aside. And I still can't be sure if the problem is related to PSU or even mobo. Still -- beginning to look more and more like RAM . . .

 

[ShintaiDK]

A simple fix might simply be to set the command to 2T.

 

[BonzaiDuck]

A simple fix might simply be to set the command to 2T.

Well -- that might be correct. After fiddling with this and that -- very minor tweaks to VCCIO and some other BIOS items -- I decided to reset everything to stock.

Done that for about a day or so, but decided to put the RAM at stock under the profile of the CPU's over-clock and other settings. So far, so good. But I still probably won't know for sure until middle next week.

All I can say at this point -- I'd left the CPU voltage settings alone while testing the OC profile. I maybe moved VCCIO up four notches to 1.14V, and changed some VRM and other power-related settings EVER SO SLIGHTLY. So for a few days of that, it was crashing twice per day -- not once in a week of 24/7 operation between bootups.

Now -- with the CPU overclock and the RAM set to stock (CMD=2, also) -- it all seems rock-solid. But I have to wait and see.

Meanwhile, I'm going to re-examine my mobo fan hookups and how I've allocated the 12V plugs to the PSU rails -- at least take a look at it.

Funny how something like this makes one begin weighing a long-intended parts order. Wish I had this much restraint when I was young and fulla beans.

 

[bryanl]

I'm just wondering if it's possible that the sticks are going bad only for the overclock setting, or how exactly they go south.

They are not. They were bad all along.

 

[BonzaiDuck]

They are not. They were bad all along.

Not true. With the OC speed, latency and voltage settings, I had tested the memory (as I said) for between 4 and 5 days with the thorough "1000%" HCI test.

Of course, the memory test software applies some stress and some voltage, so the system was running at either the spec CPU speed or the over-clocked speed during the test run. IT was not running at EIST idle. This latter state had always characterized the instability.

[Dunno . . . maybe Bryan doesn't like G.SKILL kits . . . ]

 

[john3850]

I noticed that 8GB sticks of my gskill is rated 1.5v but runs perfect at 1.35-1.38v which leads me to believe that much of there memory is ddr3L and set to 1.5v to avoid mb problems.

With the SB and IB if the vcore at idle is set too low it leads to a crash when the multiplier goes up to fast.

 

[Blain]

I'm just wondering if it's possible that the sticks are going bad only for the overclock setting, or how exactly they go south.

They are not. They were bad all along.

Not true. With the OC speed, latency and voltage settings, I had tested the memory (as I said) for between 4 and 5 days with the thorough "1000%" HCI test.

Of course, the memory test software applies some stress and some voltage, so the system was running at either the spec CPU speed or the over-clocked speed during the test run. IT was not running at EIST idle. This latter state had always characterized the instability.

[Dunno . . . maybe Bryan doesn't like G.SKILL kits . . . ]

Why would you surmise that based on the response he posted?

 

[DigDog]

hmm ...

maybe, it's the mobo, that is going bad, and not the ram?

 

[ali1988]

4-5 days is ridiculous. If the memory is bad enough to cause intermittent BSODs then it should pick up in couple hours on regular memtest86+. Or just run the HCI memtest overnight. Just because it takes 4-5 days to complete doesn't mean you need to let it complete to be reasonably certain your memory is fine. It should start off course and become progressively fine. After 12hrs of testing it should pick up a fault that is large enough to cause BSODs in windows during light use.

 

[BonzaiDuck]

4-5 days is ridiculous. If the memory is bad enough to cause intermittent BSODs then it should pick up in couple hours on regular memtest86+. Or just run the HCI memtest overnight. Just because it takes 4-5 days to complete doesn't mean you need to let it complete to be reasonably certain your memory is fine. It should start off course and become progressively fine. After 12hrs of testing it should pick up a fault that is large enough to cause BSODs in windows during light use.

[I'm overwhelmed with the attention, now . . . ] This was HCI's advice, generally accepted (I thought) for validating RAM. "Thorough" testing is 10 passes or 1,000%. It wouldn't take so long, but I'd told people all "goo-gah" about getting 32GB or 64GB into their motherboards that it takes proportionately longer to apply this level of testing the more RAM you have.

I noticed that 8 sticks of my gskill is rated 1.5v but runs perfect at 1.35-1.38v which leads me to believe that much of there memory is ddr3L and set to 1.5v to avoid mb problems.

Well, you're probably right. When those 1.35V Samsungs were released, I thought I read somewhere that G.SKILL was using the same "black parts" for some of their models. But it wasn't "official." I usually just set them at their spec, or even a little above it just for testing or using stock settings. My own "memory" is a bit hazy, but I thought for sure I'd investigated the MAXIMUM WARRANTY voltage for these G.SKILLs ("-GBRL") -- or -- I'd asked their tech-support. I think it was 1.60V. These were never run higher than 1.52V or 2-hundredths above the stock spec -- even at 1866/10-10-10-28-1T. Truth be told, their faster models at 1866, 2133 etc. have the former spec at 1.60 or 1.65, and I could only speculate about any "wiggle-room" under their warranty -- for those, there may not be any.

Fact is, I recoiled from buying the faster RAM with that 1.60V spec, because it would also require running VCCIO to at least 1.15V with a temptation to run it higher. Now, G.SKILL has produced 2x8GB DDR3-1866 models that run at the tighter 9-9-9-24 timings and only require the 1.5V setting.

With the SB and IB if the vcore at idle is set too low it leads to a crash when the multiplier goes up to[o] fast.

This was the first thing that occurred to me, and it's an observation one finds on various forums in addition to Anandtech.

I'm betting that there are VRM -- "power" and current settings that would resolve it. But the multiplier is at EIST level when these crashes happen -- the computer sitting there feeding nothing but my HDTV at an idle voltage of ~1.008V.

My problem: I sometimes fail to keep thorough notes as I did when I was OC'ing the system and running stress-tests. And again -- my own "memory" without notes is hazy -- but I think this problem began not long after I'd set the RAM at 1866 with the corresponding timings. It was "regularly" infrequent, as I said -- once per week or so. Actually, I went through the entire month of August without any instance of the problem. I found the instances of occurrence in the Event Logs shown by Event ID 41, and keep a spreadsheet of the time between failures. There is no change in pattern with the adjustments I made prior to resetting these RAMs to stock settings a couple days ago. Only some minor changes I made last week resulted in it happening twice over a range of hours.

Now that I've set the RAM back to their stock settings, it looks pretty good. But I'll still need to wait and watch for another four days or so to see if it's disappeared or the frequency of occurrence has diminished over what it had been.

For the motherboard, I'm very skeptical that it's lost even an iota of its original operability. Where have I twisted up the voltages that would strain the board? How many minutes or hours has the board been run at the OC settings and load VCORE? With the features they built into these boards, the extra heatsinks and solid-state components -- I don't see it that way. I think -- barring the RAM actually going "bad" -- my biggest worry is any sort of data loss or OS corruption.

Time will tell -- of course . . . Point being -- when we "tune" our systems, we expect instability to occur in a couple days' time. This sort of thing is seldom anticipated.

 

[Gillbot]

4-5 days is ridiculous. If the memory is bad enough to cause intermittent BSODs then it should pick up in couple hours on regular memtest86+. Or just run the HCI memtest overnight. Just because it takes 4-5 days to complete doesn't mean you need to let it complete to be reasonably certain your memory is fine. It should start off course and become progressively fine. After 12hrs of testing it should pick up a fault that is large enough to cause BSODs in windows during light use.

Not really, i've had sets pass memtest for a short run but leave them go for 24 hours and errors build.

 

[BonzaiDuck]

Not really, i've had sets pass memtest for a short run but leave them go for 24 hours and errors build.

Long Time, Bro! Like the old historical radio series -- "You were there" when I was tuning this sucker. You and John3850 and some others -- we created a pretty lively thread.

Point here has been noted before: People have different standards. I've seen some gamers run their overclock stress settings for 8 hours and call it a day. And also fact of the matter: We don't NEED to leave these OC settings operable interminably -- it's really not an inconvenience to store them as BIOS profiles, re-boot occasionally and set them back to stock between game sessions.

. . . Unless your gaming is a nervous habit and you multi-task a lot. . . .

 

[BonzaiDuck]

Well . . . the plot thickens, but it should take another 5 to 7 days to confirm it. It is now extremely likely ("almost" certain) that my problem was the RAM over-clocking (in general). It may prove (in specific) that it was only the choice of the command-rate or 1T instead of 2.

Looking back, these errors occurred so infrequently that I didn't associate the RAM settings with them. If I had, this would have been resolved months ago.

So that leaves me with "enthusiast" versus "practical" choices. Let's face it: on these Z68 boards, you don't reap much in gains for running RAM beyond 1866. OR -- the daily performance difference between 1600 (stock/default) and 1866 is not really noticeable. The benchmarks with AIDA 64 however are "neatsy-keen!" And of course -- the command-rate.

As an "enthusiast" with a balanced-budget and clear credit, I suppose it would be nothing to buy these:

http://www.newegg.com/Product/Produc...82E16820231625

They run stock at 1866 with timings 9-9-9-24 and voltage 1.5V. One customer review cites ability to drop the CAS latency from 9 to 8. Can't be sure if that requires upping the vDIMM or leaving it alone. OR-- MAY-BE -- I could set the command rate to T1 and "no problem."

Maybe. Five or ten years ago, for a machine that's now 2.5 years old, I might have placed the order by now without hesitation. $170 plus tax -- probably free shipping. I also want to buy a Solaire Anywhere table-top grille this year. Either way, the old grille is great for steaks and pork-chops; the old RAM seems (so far) to be running great on stock settings.

And frankly, ShintaiDK may be right: These 4x4GB GBRL's might be totally rock-solid at 1866 10-10-10-28 CMD=2. And again -- only time would tell . . .

Maybe setting CMD=1 would only mean bumping up the vDIMM to 1.55V and the VCCIO to 1.13 or 1.14. And again -- only time would tell.

[You figure 'time would tell' when I ran the HCI-Memtest 1,000% "thorough" test, though, wouldn't you? Live and learn, folks!]

 

[BonzaiDuck]

Resolved -- with observational caveats . . .

This somewhat regular but week(s) apart problem of unanticipated resets and a few BSODs is apparently due to the RAM settings.

It doesn't appear that the RAM is damaged in any way; the voltages were all well-within the spec-to-warranty-limit range and close to the spec.

The elimination of the problem events might continue even if I run up the speed of the RAM again to 1866, but leave the command-rate at "auto" or CMD=T2.

I'm trying to remember who it was who shared his own G.SKILL "GBRL" settings with me; IDontCare? Gilbot? VirtualLarry? Maybe it was Larry, but I was grateful for the insight, because these modules indeed do run at the higher speed with the looser 10-10-10 timings.

But who would imagine, that you could stress the modules for four days and get past the 1000% "thorough" memory test -- then see the instability pop up after a week's 24/7 operation or longer? To see it happen with roughly the same continuing frequency?

If you can't detect the potential instability in 4 days of testing, it throws a different kind of uncertainty into attempts to wring extra performance out of your sage and modest investment.

There might be another slight factor in Larry's(?) comfort with this versus mine: He had the Maximus V mobo. But some features on my Z68-v-pro are almost as prodigious.

Running the G.SKILLs at 1866 instead of 1600, or at CMD=1 as opposed to 2 -- not essential to my happiness or inner fuzzy feel-good. Something to play with, though . . .

 

[BonzaiDuck]

Just one more thought -- a correction.

Among longstanding colleagues and associates on this forum, it was IDontCare who tipped me off about the G.SKILL "flexibilities." Looking back through my PMs, I see he might have implied that even looser timings of 11-11-11 . . . for the GBRL's could be in order, although he was fine with 10-10-10 on his MIVE mobo.

Again, though, this entire episode shows how you can validate hardware and settings over a 5-day continuous period, then believe you've found the gold at the end of the rainbow (IDC's metaphor). Then, further down the road, seemingly disjoint for time-elapsed, a mysterious and infrequent "event" reoccurs.

Now, I may hunt down my correspondence with G.SKILL techs, so I can confirm their own sense of confidence about their product.

 

[BonzaiDuck]

http://forums.anandtech.com/showthread.php?p=36044890#post36044890

 

[VirtualLarry]

I'm trying to remember who it was who shared his own G.SKILL "GBRL" settings with me; IDontCare? Gilbot? VirtualLarry? Maybe it was Larry, but I was grateful for the insight, because these modules indeed do run at the higher speed with the looser 10-10-10 timings.

But who would imagine, that you could stress the modules for four days and get past the 1000% "thorough" memory test -- then see the instability pop up after a week's 24/7 operation or longer? To see it happen with roughly the same continuing frequency?

If you can't detect the potential instability in 4 days of testing, it throws a different kind of uncertainty into attempts to wring extra performance out of your sage and modest investment.

There might be another slight factor in Larry's(?) comfort with this versus mine: He had the Maximus V mobo. But some features on my Z68-v-pro are almost as prodigious.

Must have been Gillbot. I never owned a Maximus V mobo.

 

[BonzaiDuck]

Must have been Gillbot. I never owned a Maximus V mobo.

It was IDontCare. And the point of it is: It was solid advice. Even so, I thought my board was overloaded with RAM sticks for the overclock. it wasn't. It was loaded to the max, but it was able to allow overclocking those DDR3-1600's. So Like I said elsewhere, I have a spare set of "quad" RAM sticks now for my next project.

The trouble was the *#%&!mn VGA driver!! And the troubles I'd had with media center when I update it, making me reluctant to update it!

But -- Z68 -- he had a Maximus IV Extreme board.

I'm rock-solid now -- after I uninstalled and then "clean-installed" the latest NVidia download. Seven days, 24/7 and counting . . . throwing everything I can at it.

Live and learn!