Corsair RAM, errors in Memtest86


Jun 12, 2012

I resently build a new gaming computer,
Motherboard: MSI Z77A-GD55
CPU: Intel Core i7 3770K Ivy Bridge
PSU: Corsair Enthusiast Series Modular TX85
RAM: Corsair 16GB DDR3 1600MHz Vengeance

after a week i started getting BSOD twice a day, so i ran a memtest86 on the RAM, and found out that one of the ram stick were faulty, so i returned them and got, 4 new stick, did a overnight test on the new sticks, 4 passes 0 errors, now after one more week, i got BSOD, tested the RAM again, and the RAM stick that was installed in Slot 1 has the same error as the first RAM stick.

is it my motherboard or is the RAM not compatiable with my motherboard, cpu or something else, BTW nothing is overclocked only XMP enabled, and all drivers is up to date.
Did you have all the RAM inserted when you ran the tests or did you test them 1 stick at a time?

If you didn't, try testing 1 stick at a time and try a stick that showed errors in another slot other than slot 1.


Jun 12, 2012
Hey and thanks for the Welcome :)

what am i looking for in the programs, is it just to check if the RAM is running at 1,5v and have the right timmings?

i tested them separately all in Slot 1, just tried testing the faulty RAM in Slot 2, still giver me errors.

think i am just going to return the motherboard, to be on the save side.
Note some of those applications might not be able to read the voltage for your MOBO -- that's up to your MOBO. I just ran across a case a few days ago where the DRAM Voltage was way too high. Both I and MOBO MFG agreed shut it down and RMA the MOBO to avoid damage.

RAM fails but at a very low rate <1%, 3-4 sticks -- you've got a voltage issue.

IMO contact MSI and ask them, they might have an App to suggest that can properly read the voltage. Also, look and see if there's any RAM Temperatures > 60C is of concern.

/edit - Just because you or the BIOS sets the DRAM = 1.50v doesn't mean the controller is damaged and has it pegged to 2.5v.
Two things I wanted to point out:

1) Only Crucial and Kingston RAM fail at less than 1% per package (variable number included). Retailers process replacement requests for all other RAM brands at a rate higher than 1%. Corsair quite a lot more than 1%. Some model lines of Corsair RAM (like XMS) are more like 7%.

2) The people with no problems don't come in here. You would have to measure the RAM failure rates as a percentage of all new computer problems instead. If that was done, then RAM would be well north of 10% likely. Maybe north of 20% likely.

It may really not be a bad stick instead only bad voltage or it may not be the RAM at all, but quoting some 1% statistic is misleading at best.
As a whole maybe, like MOBO's 2%~3% but less than 1/2 are actual failures. Most MOBO's are damaged either by the End User or the End User's PSU.

RAM - it depends on the kit, Corsair Dominator's <1% ... kits like Corsair XMS no doubt much greater. Vengeance are 1% according to Corsair.

RE (2) above if RAM had a 10%~20% failure rate OMG the problems would be non-stop here or in any forum.

9.9+/10 RAM 'problems' are End User induced, I've lost count how many 'RAM = FAIL' went to 'RAM = PASS' once I helped folks set their RAM properly in the BIOS.

/edit - based upon the OP's comments (if true) I have little doubt the MOBO is killing the sticks no matter what kit, kits, stick or stick are used. Going from Pass -> Fail in short order ... come on. Try the 'bad' stick or stick's in another PC or fiends PC and run Memtest.
The problems here are non stop.

I am not saying that RAM kits are 10% likely to fail, just that if there is some kind of failure on day 1 that its probably at least that high that the failure is RAM related.
If anyone knows this section -- it's me x10. Most Threads are: Buying A or B, Wrong Frequency, assuming a RAM issue but not one, OC RAM, and a small percentage are a real Memtest issue. Most issues (Memtest) can be solved with either a BIOS update when appropriate (mostly 4GB/stick on AMD) and/or a slight increase of either the DRAM or 'CPU IMC' voltage (QPI/VTT/VCCIO/VCCSA/CPUVTT/CPU-NB/etc).

When I see Good -> in //// Bad -> out with XMP setting I guess you 'could try' increasing the appropriate CPU IMC voltage to prolong Good -----> Bad.

Q - What do you tell the guy with 5-6 bad MOBO's in a row??!!

Q - Why do you think the high-end MOBO's have actual voltage contacts?
I hang out non-here in the "Home Built" section. I am just cross training atm.

I am not a RAM master, but I do see RAM problems all the time and they are a large percentage of the total new build problems in the Systems sub-forum.

Over there, there is a whole lot less of "if you configure it this way it might work" and more "it should work out of the box or you should get something else instead".

For my part, I haven't seen anyone ever have 5-6 bad motherboards in a row, or more than 2 for that matter, but I would probably tell them it wasn't a motherboard problem.

I also probably wouldn't tell anyone to get a $250 motherboard in the first place so the nit picky details of why it has or doesn't have some niche feature would fly right by me. The people that come to my section are average Joes and they really don't need $250 motherboards.

For that matter they don't need anything else to be happy other than RAM that works out of the box (and other stuff that works out of the box too). When I design new computers for people, it is with "working out of the box" in mind for that reason.
Some kits don't do DDR3-1333 very well at all, try running out of the box a DDR3-2400/2600 kit at 'Defaults' -- it's an effort for those to post until properly set. Further, some CPU's i.e. SB-E won't run properly (Turbo 200MHz+ lost) on DDR3-1333 (Default).

Ditto here when I see multiple sticks fail it's not a RAM problem. Sadly, I've seen folks with the 5-6 bad MOBO's -- blaming it on the MOBOs...for sure.

I won't argue the benefits of cost, I'll just say some of our servers cost more than folks cars and even my i7-3930K with 32GB isn't nearly enough for anything more than 'small' pieces of SQL. Folks who want an i7-3960X, 4-WAY GTX 680's, water blocks don't want me to tell them get DDR3-1333 there's little benefits...they want every last morsel.

Out of the box on what? If you want Defaults then buy a Dell/HP/GW/etc.
Well, I guess the average Joe that I always deal with doesn't care to query multiple million long databases and they don't expect millions of results so different worlds I guess.

The people that come to me with the idea of getting 2600 RAM usually have no idea what they are getting above 1333/1600 (not much) and don't care to spend the extra $ if it isn't cost effective.

They also tend not to be overly interested in 3960x too.

Although, I did help a guy one time build a $7000 gaming setup and a different guy do some crazy 2x Xeon core on a SR-2 Blackhawk motherboard with 2x of the highest end Quadro cards on the market before.

Those are the only two people I have met for whom 2000+ RAM is even a question in my entire IT career.

Everyone else seems to care about bang for their buck.

Out of the box on what - The sorts of things in my signature I guess. Gaming PCs priced at $600 - $1500 mostly.

Average Joe who comes in wanting to build a gaming computer doesn't want to spend average 100 hours tuning settings, they want to start gaming the same day they receive the parts. No configuration, no RMA, just gaming.
I build at least a couple of 'those' folks rig's per month locally and by referral, but I refuse to build another SR-2 (SR-X now) rig.

Well you've met another...I am one of the suppliers to RealtyTrac, and several financial and broker, etc companies. REO/IDX/GIS/CRM/TAX databases with an obscene amount of data and images. My wife's dinky databases, textual data only, in total is 3GB+.

Folks wanting 'those' builds rarely ask here, though I've run across and referred David Trubby more than one build - example I ran across this guy -> and told the guy all the problems ahead of time. I made David's background as a joke when he gave me his 2 cents on my SR-2 build -

None of my builds are anything more than 10~12 hours (SR-2 exception), most all are fully blocked, and they're limited to a few cases and core components.