Interpreting Memtest86+ Results

[nikki605]

Today, for the first time in a very long time, I had a BSOD crash on my T21. It was just idling, I wasn't even using it. After rebooting, I let it report the error to Microsoft. The status I got back was a memory error.

I've started running memtest86+ and so far have one error, but I don't know how to interpret the results - how do I know which of the 2 sticks (2x256) got the error. Both sticks have been installed since 2004.

Here is the one error so far:

Tst - 5
Pass - 1
Failing Address - 00007038428 - 112.5MB
Good - ffffdfff
Bad - ffffffff
Err-Bits - 00002000
Count - 1
Chan - (blank)

I guess I'm not even sure which memory stick is which - how the sockets are arranged.

Thanks for any help.

[rkawakami]

I haven't (yet) determined which slot corresponds to the failing address, so I simply test each module by itself. The "good", "bad" and "err-bits" values are in hexadecimal notation; digits are 0-9,A-F and represent value in base 16. The "F" is equivalent to decimal 15, or 1111 in binary. The "D" is equivalent to decimal 13 or 1101 in binary. Which is basically just a way of telling you that only one bit was detected as being in error. It's fairly rare to encounter a single bit error but it does happen.

My advice would be to pull out one of the 256MB modules and repeat the testing. Let it go at least three complete passes before considering it "good". Repeat with the other module. It shouldn't matter which memory slot you use as long as you use the same one for each test. If both pass, try the other slot. If both STILL pass then let me know and we'll go from there.

You can also put memtest86+ into a mode where it repeats only one of the tests in the sequence. Press the "c" key and follow the prompts to select test #5 (Test Selection, Select Test, Test Number). Press the "0" key to continue testing with the selected test.

[nikki605]

Thanks, Ray. I've taken out one stick. The one that's left is in the rear socket, closest to the LCD hinge. Since its getting late here on the east coast, I'll let the standard test run overnight, check it in the morning and post the results.

[nikki605]

I've completed the 4 tests - each memory stick in each socket, one at a time. The minimum number of passes on any one test was 6. The result - no errors on either stick, installed in either socket.

I'm beginning to think I may have had a seating problem or a dirty socket pin.

One other item of possible note: I said in my initial post that I've been running these sticks since 2004, which is true. What I forgot to mention (if it even matters) is that this is a different motherboard than the original one in 2004. My original mobo developed the BoD and I replaced it in April, 2007.

Just for chuckles, I'm going to install both sticks again and run memtest overnight.

Any other ideas or thoughts? Ray?

[rkawakami]

Just for chuckles, I'm going to install both sticks again and run memtest overnight.

Good thought to run them together... that was going to be my next suggestion.

Any other ideas or thoughts?

A couple, but your eyes may glaze over .

I said earlier that single bit failures are fairly rare. As semiconductor fabrication processes are supposed to be very uniform, to have only ONE bit out of the entire array of 2,048 megabits (256 megabytes x 8 ) fail at any time is a very, very, very small percentage. Most memory failures can be traced back to broken external connections (solder ball/bond breaking), electrostatic damage where one or more input or output pins are "blown up", or internal failures in the memory resulting in entire rows or columns of bits going bad.

One source for random, single bit failures which is very hard to pin down is alpha particle disturbance. This happens when a high-velocity alpha particle passes through the memory array and temporarily changes a bit of data. Where does that come from? Radiation from naturally decaying material, some of which can actually be contained in the plastic or ceramic compounds used to encapsule the memory chip(s). It does not permanently damage the memory; it just makes a zero turn into a one or vice versa. This is known as a 'soft error' because if you re-write the data, then all is well again. Modern memory designs and packaging have greatly reduced this type of error from happening but it still can occur.

Another, more likely source of failure, is from a refresh standpoint. Dynamic memory like SDR and DDR has to have the data periodically re-written in each bit. A memory bit can be thought of as a tiny electrical charge which is stored (or not) in a capacitor. The charge in a capacitor as small as what's used for a memory bit naturally "leaks" off. The act of refreshing simply is the process of re-charging the tiny capacitor. For the memory inside a T21 it uses what has become called "low density" memory. This is another way of saying that 4,096 refresh cycles need to be given to the memories every 64 milliseconds. It's possible that the one bit that failed doesn't have the same charge storage property as the rest of the memory. High temperature is generally the worst case condition for refresh; i.e., higher temperatures means the charge leaks off faster which means you need to refresh more often. Why it failed when the system was idle (and thus more likely at a lower operating temperature), is unknown.

Bad contact(s) between the module and motherboard can affect the signal properties being sent between the two, so re-seating them could have cleared up the initial problem. I don't think that your previous motherboard swap has anything to do with this error. If after your testing of both modules reveals no failures, you might have to chalk up this experience as "unknown failure".

[nikki605]

First - with both sticks installed and only 44 minutes into pass # 1, the test has reported 4-single bit errors. The the good and bad bit patterns and error bits for all 4 errors are exactly the same as in my original post. The failing addresses are different.

Now what?

Next - I read your "research paper." I didn't understand it, but I did read it. All I can say is

Since I did not get any errors when testing the sticks individually, what sense can I make of this result? I've left the test running, but I'm no closer to knowing which stick has a problem. Or could the motherboard now be suspect? I do not have any other memory sticks to swap. I would have to go buy one. I could try swapping the 2 sticks in their sockets, but if I still got errors, what have I proven?

[rkawakami]

I'm left with the last things that I can think of and which I left off of my previous analysis... power line noise and/or memory bus loading. Most likely you will get the same type of error but in different addresses if you swap the two modules around. The power line issue is that the modules are supposed to be provided with a "clean" 3.3VDC power source. However, due to all of the signals being generated in the laptop, brief overvoltages and undervoltages (i.e., noise) is present on the power pins. That can be enough to disturb reading or writing of data.

The memory bus loading issue is that the electrical signals can be slightly altered when you have two modules installed. Either something on the motherboard has weakened to the point where they aren't fully working OR there's something on the module(s) which is causing the signals to degrade. As it's easiest to try different modules than replace the motherboard (again), that would be my only suggestion.

As an aside, what is/are the modules in question? By that, I mean manufacturer and part numbers.

A thought occurs... you can try using a pencil eraser to clean the edges of the modules. There could be some oxidation on them and rubbing an eraser across the contacts can sometimes cure weird problems like this.

[nikki605]

Pass # 1 completed with no more errors. However, pass # 2, which is running now, has picked up 4 more errors. All 4 are the same single bit errors, just at different addresses.

Interesting you should ask about the part numbers. Both sticks (I bought them at the same time) have the same labels, one on each side. Here are the complete labels, not just the FRU.

On one side is an IBM label:
256MB PC100 CL3 3.3V S N (32Mx64)
OPT: 33L3069 FRU: 33L3070

On the other side is a Micron label (the large M which I think is pre-Crucial):
MT16LSDF3264HG-133E4 PC133S-333-542-Z
US BZABT0G010 200249
256MB, SYNCH, 133MHz, CL3

I have read a lot of posts on the forum about what speed memory can be used in the various T models. If I am reading the labels correctly, it looks to me that these are Micron PC133 memory sticks that IBM has re-marked as PC100 sticks. Yes? No?

Anyway, since I had to power down the T21 to get the label information, I am going to re-install them swapping sockets and restart memtest. I fully expect to still get errors, but, as you pointed out, probably in different locations.

I guess I'm going to have to break down and by new memory stick(s). What do you think?

Oh - I did the pencil eraser thing on the pins. I too have had success with that in the past.

EDIT: After erasing the pins and swapping the sticks between sockets, I am still getting the same, single bit errors. Usually 4 per pass, in test # 5 (whatever that one is).

[hitokage]

I thought I could help out some.

(the large M which I think is pre-Crucial)

Micron doesn't brand OEM RAM with the Crucial name - as far as I know anyway.

MT16LSDF3264HG-133E4

Spec sheet for anyone interested.

I have read a lot of posts on the forum about what speed memory can be used in the various T models. If I am reading the labels correctly, it looks to me that these are Micron PC133 memory sticks that IBM has re-marked as PC100 sticks. Yes? No?

Yes it is PC133, however it is low density (according to the spec sheet), and shouldn't be a problem - IBM did give it their rubber stamp of approval.

I guess I'm going to have to break down and by new memory stick(s). What do you think?

Here's my opinion - with memory testing software if the memory fails the softwares tests it's bad, but if it passes it still could be bad. Since it only fails with both modules installed it could be one of the other possibilities rkawakami has mentioned. The easiest way to test this would be to swap both modules with another set and rerun the tests as this should eliminate any source of problem coming from the motherboard. Of course this only works if you have another set of modules - you may end-up buying another set whether or not what you have is bad or not just as part of troubleshooting.

Steve G.

[nikki605]

Thanks for the info. As I stated earlier, I don't have any additional sticks, so I would have to find and buy them to test.

Guess I'll start checking out prices.

[RealBlackStuff]

As an afterthought, if they were originally PC133, they may have not been up to the standards of PC133, and were thus down-classed and sold as PC100.
In other words, they were probably inferior to begin with.
A new set would be highly recommendable.

[nikki605]

I'm typing this post on the TP with both memory sticks installed. I booted the laptop into XP some time ago and so far, it is running fine, although I expect it to crash again.

I'm beginning to think that I may have had this problem for quite some time, but it hadn't surfaced until now. As Ray stated, single bit errors are not common. It probably takes just the right value to be written to just the right memory address for the failure to show up.

I'm thinking of going with the Crucial memory in the sticky at the top of this forum. They are PC133 but will only be run at PC100 speed (CL2). I wonder why Crucial has different SODIMMs for the T20, T21 & T22? They all use the same motherboard. Wouldn't the memory modules be the same too? A quick check on eBay only showed one 33L3070 up for auction and the prices in eBay Stores is too high.

[hitokage]

As I stated earlier, I don't have any additional sticks, so I would have to find and buy them to test.

Sorry, I must have missed that bit.

As an afterthought, if they were originally PC133, they may have not been up to the standards of PC133, and were thus down-classed and sold as PC100.
In other words, they were probably inferior to begin with.
A new set would be highly recommendable.

Micron doesn't sell modules like that, and chips that fall into that category get resold to third party memory module manufacturers (assemblers). Remember IBM marked them as being PC100, not Micron which means there were fine as PC133.

I'm thinking of going with the Crucial memory in the sticky at the top of this forum. They are PC133 but will only be run at PC100 speed (CL2). I wonder why Crucial has different SODIMMs for the T20, T21 & T22? They all use the same motherboard. Wouldn't the memory modules be the same too?

Those numbers are "stock numbers" - there's another thread where that is brought up. Also, PC100 can be CL2 or CL3 just like PC133, and CL2 will normally function in machines/motherboards that require CL3 - at the CL3 speed of course.

Steve G.

[nikki605]

After a little more searching, I came across this FAQ from the memtest86+ forum website.

http://forum.x86-secret.com/showthread. ... 601&t=2807

It had some good information, including this statement from a paragraph that discussed the same point Ray made about errors that only occur when both sticks are installed. It said, in part "...relax the memory latencies..."

I don't think however, that this is a user selectable option in a TP. At least, I don't remember seeing any user options in the BIOS that deals witrh memory latency timings.