RAM fails Memtest86+ when tested together

September 20, 2011 at 17:35:41
Specs: Windows 7 x64, Core i7-920/12GB DDR3-1600
I recently noticed that Windows 7 was only recognizing 8 of my 12GB installed RAM. I have also been getting BSODs that point to memory problems. I have 6 x 2GB of DDR3-1600 installed in all 6 slots on my motherboard (GA-EX58-UD5) with an Intel Core i7-920 CPU. I tested the RAM with Memtest86+ and it failed with errors. I continued to test each stick separately and they all passed. I tried every combination of sticks in slots that my system would allow and some combinations would fail and some would pass with no apparent pattern at all. There isn't a single slot that always passes or fails. The best I could come up with is it is more likely to pass with combinations that use less channels. The memory controller supports up to triple channel configurations. Triple channel configurations nearly always fail. Dual channel often fails. And single channel only occasionally failed. So it seems that there is a problem with either the motherboard or the CPU's memory controller, but I don't know which. They are both under their 3 year warranty until November of this year. So I can have them replaced, but I don't know which needs to be. I have never had a CPU stop working ever, but then again they didn't have memory controllers in them until recently and continue to become exponentially more complex. Sadly, I don't have an LGA 1366 CPU or motherboard to test them with separately. Any suggestions with what to do?

See More: RAM fails Memtest86+ when tested together

Report •


#1
September 20, 2011 at 17:59:31
You can't test dual data rate memory sticks one at at time. If they fail in pairs then you need to try to figure out which set or single one fails. Same here. They simply must pass memtest or microsoft's memory diagnostics. It may be the motherboard but you can check that your bios settings are set correctly and that the ram is an approved ram. Otherwise you may have to de-rate the settings to get to pass or they may never pass.

May have to go to some tech center to double check ram or board. Or send the ram back.

1/3 of highway deaths are caused by drunks. The rest are by people who can't drive any better than a drunk.


Report •

#2
September 20, 2011 at 18:52:47
What's wrong with testing the sticks by themselves? They just operate in single channel mode. You get lower memory bandwidth, but the OS will boot and the system will run that way. I don't see why testing the individual sticks that way shouldn't work either.

This RAM has been working in my system for 2 years, so it's not the BIOS settings or RAM type or timing that causes the problem. I did upgrade my BIOS to the latest beta version and reset all relevant settings to default (I had to change back settings like RAID, etc. to preserve my HDDs), but this did not change anything. I also lowered the data rate to the default 1066 MHz and conservative timings, although the RAM is rated at 1600MHz in XMP, but continued to get errors in Memtest86+.


Report •

#3
September 20, 2011 at 21:14:09
If it ain't broke don't fix it. If the BIOS worked well and there was no overriding reason to update then WHY would you try a Beta BIOS? About the only reason I would resommend that is if you were upgrading to the latest CPU and only the newest BIOS would support it.
Try rolling back to the last BIOS before the newest one and see if that helps. The Beta version may not be completely stable with all RAM installed and then inform the MB mfg about your failure on the Beta test. You were a guinea pig, you were a Beta tester, now report the error and roll back.

You have to be a little bit crazy to keep you from going insane.


Report •

Related Solutions

#4
September 20, 2011 at 22:14:33
Sorry, I wasn't exactly clear, but I didn't upgrade the BIOS and then start receiving the memory errors. I upgraded the BIOS in an effort to see if it would help at all with the errors. The problem was not affected much at all by BIOS change, but if anything it helped a little. Sometimes when I booted with memory populated in certain slots on the old BIOS, it would only report some of the memory as being there. That's how I first noticed the problem, because the OS could see that there was 12GB in the slots, but only reported 8GB as available. The same thing in Memtest86+, it would show the 6 slots populated at the bottom under SPD information, but it would only have 8GB cached. I could boot multiple times with the same memory configuration and have 8GB, 10GB or all 12GB available on different boots. However, since I upgraded to the beta BIOS, I have not had this problem of the RAM not all being available. But I receive memory errors with both BIOSes.

Report •

#5
September 21, 2011 at 14:50:22
Let me make this more simple.

Memtest has to pass.

You wasted your time with a bios update.

1/3 of highway deaths are caused by drunks. The rest are by people who can't drive any better than a drunk.


Report •

#6
September 21, 2011 at 18:12:22
Thank you for simplifying my complex problem with your profound statements. I'm much better off having been told the obvious.

Actually it turns out I didn't waste my time with a BIOS update since, as I stated, after the update I no longer have the problem with the system not recognizing all the RAM that is installed, which makes actually testing all the RAM that's installed a lot easier.

Since I already know that Memtest86+ has to pass, my problem is how to get it to pass, i.e. is it my motherboard or is it my CPU that needs to be RMAed? If you have no useful advice to contribute to finding an answer to this question, you need not respond, unless somehow making those sorts of posts boosts your self-confidence, in which case I can oblige that need I guess.


Report •

#7
September 21, 2011 at 18:33:05
Did you do any of the things I have suggested?

1/3 of highway deaths are caused by drunks. The rest are by people who can't drive any better than a drunk.


Report •

#8
September 21, 2011 at 20:06:00
Have you looked into memory support list on the motherboard manufacturer's website? If the memory is not on the support list, it might work, if it is on the list ir has been tested by the mfg to work. If it works in small groups but not all filled, this might be the case.

Have you tried increasing the memory voltage by 0.1V or 0.2V in your BIOS to see if that under manual memory settings to see if your memory stabilizes at the higher voltage. It is possible that with all of the memory slots filled you are getting a small voltage drop and the memory is getting unstable, increasing the voltage by a small amount that is still within the operating range of the memory may be just enough to stabilize it when you fill all of the slots.

You have to be a little bit crazy to keep you from going insane.


Report •

#9
September 21, 2011 at 21:53:07
I have done most of the things you suggested, including reseting BIOS settings and lowering frequency/timings of RAM. These changed nothing. I am not looking to take it to a shop and spend any money on it, which is the only other thing you suggested.

Using methods, some of which you said were worthless, I have convinced myself that the RAM itself is not the problem, since each stick will pass 10+ passes of Memtest86+ when tested alone. It seems to be something to do with the memory controller as it is multiple channel configurations that are more likely to fail.

The RAM (F3-12800CL8D-4GBRM) is not on the short list of approved RAM, but a similar model is (F3-12800CL9D-4GBNQ). However, I know the problem is not incompatibility because all 12GB have been operating flawlessly at full XMP frequency/timing/voltages in the system for the last 2 years. It only recently became a problem, which means something just started to fail.


Report •

#10
September 22, 2011 at 05:11:40
If this is a built system and it sounds like it, you can manually set up your memory and try increasing the voltage to them by a tenth of a volt and still be within the operating voltage of the memory. This is done by those who overclock to stabilize the memory, especially when running the memory at higher speeds (which you are not doing), but if there is some degradation in the memory controller or even the power supply voltage, this MAY be enough to fix this. Adjusting the memory voltage is not rocket science, and you can write down all memory settings when you set to manual, before you make any changes so you can always return it to the way it was. You may need to set the general memory to manual and then lower down set the memory voltage to manual, and then make the change(s) needed. If you have a friend that has done overclocking, then he may be more comfortable making these changes than you are. Most 'repair shops' would not do this for you but try to sell you new memory or a new motherboard, or both. They would not be capable and/or give you all kinds of reasons they would not and you should not touch these settings, but if you use reasonable caution and try only a small incremental increase, you will be well within any safety margin, and should actually be within the specified operating voltage of the memory.

The number of the memory you list that is 'approved' is one number higher in the middle, which MAY mean a later model than yours. This may or may not be significant, but since you have been using this memory for a while, may not be significant, except to keep this in mind.

You have to be a little bit crazy to keep you from going insane.


Report •

#11
September 22, 2011 at 12:54:29
You can't test one ram but since you refuse my help then please continue your efforts.

Some part of the memory system tests make it fail. Ram, ram controller, ram slots, motherboard, motherboard flex, power supply, cpu, associated ram chips.

1/3 of highway deaths are caused by drunks. The rest are by people who can't drive any better than a drunk.


Report •

#12
September 23, 2011 at 12:02:06
@Fingers Yes, I built the system myself and have built all my own systems (and many others) over the past 12 years. I've done my fair share of overclocking, although I usually don't have the need to push my system to those levels with the stock performances these days. I tested before by raising the DRAM voltage manually to 1.64V, as opposed to letting the XMP profile raise it to the SPD setting of 1.6V, but still received errors. There are also options to change some other DRAM voltages, so I will look into those and test with raising them as well.

And more than likely, the number in the middle of the approved RAM model number is representative of the RAM's CAS latency value. CL9 is actually slower than CL8, which is the timing of my RAM (8-8-8-24). So the approved RAM is most likely in fact an older model.

@jefro I really don't see why you're so insistent that you are right here. There is absolutely no reason I can't test each individual stick of RAM separately. I've tested configurations (in fact all possible that my motherboard supports) of RAM from 1, 2, 3, 4, 5 and 6 sticks and received results that return errors and results that run stably. If the system will boot and run with those RAM configurations, then the RAM can be tested that way.

And thank you again for stating the obvious again, which I've known from my very first post here. Of course some part of the memory system is what's making it fail. The question is which part? Your continuing posts are contributing absolutely nothing to answering this question. While my testing of each RAM stick separately has shown that the RAM itself is not the problem, but that it has something to do with the multiple channel configurations used on the memory controller, i.e. the memory controller itself (the CPU) or the path to the memory controller (the motherboard) or possibly the powering of these components is the problem. I could not have conclusively ruled out the RAM being bad without testing them individually.


Report •

#13
September 23, 2011 at 13:31:02
I tested by raising DRAM Termination, Ch-A (B & C) Data Vref and Ch-A (B & C) Address VRef to 0.82V (half of the 1.64 DRAM voltage) and still got errors (testing with all 12GB). I tested by raising QPI/Vtt voltage incrementally as high as 1.335V (any higher is warned against) up from 1.175V and IOH Core voltage as high as 1.26V up from 1.1V. I still got errors consistently here as well.

One thing I should note with the 12GB all 6 stick configuration, the errors happen consistently when the last 12G - 12G address range is tested. This is after it's tested 184K - 2048M, 2048M - 3583M, 4096M - 6144M, 6144M - 8192M, 8192M - 10G and 10G - 12G, so it's always the very highest addresses of the 12G range that return errors. This is consistent across all the tests as well. It's also consistently the same err-bits, namely 00004000 (actually a single bit). For the most part this is true with most other configurations, where the errors occur for the highest range of addresses (in the 10G range when 10GB, in the 8G range when 8GB, etc.). While there have been some exceptions to these observations, it is most often the case.


Report •

#14
September 23, 2011 at 19:20:38
You have gone about as far as you can without spending wildly to replace components (MB, CPU, RAM), I think you need to bring these observations to the motherboard mfg's tech support and possibly even Intel's tech support and see if they can agree on what the problem is. I fear they will probably blame each other or the memory, but it might be worth the try and they might both tell you to RGA them for testing and possible replacement.

You have to be a little bit crazy to keep you from going insane.


Report •

Ask Question