Strange memory errors - thoughts?

[DiagonalArg]

Hi all. I've got a T520 and a T530, and I've been running memtest86+ like crazy on the two. I've got a pair of 4G sticks which behave just fine in either machine. I've also got a pair of 8G sticks (HyperX, KHX16LS9P1K2/16) that behave fine in the T520, but consistently gave me errors in the T530. The errors were at just over the 10GB point (actually, just past 10,000MB), so I swapped the two chips inside the T530, thinking that if it's one of the modules, the errors should move down to the 2GB area. That didn't happen, they remained just past the 10,000MB point. Uh oh, that seemed like a problem with the machine.

Anyone have any insight? Is my T530 defective? Any other testing I might do??

/DA

[rkawakami]

When the modules failed, was it a single error, a couple of errors or something that looked like it was catastrophic? And did the error occur right away or so time into the series of patterns (tests) that memtest86+ executes?

I would run memtest86+ on a single 8GB module in the T530. Let it run for at least three complete passes (assuming no errors). If it passes, then move it to the other memory slot and repeat. Do the same to the second module. You'll then have a table of how each module performs in each slot.

The KHX16LS9P1K2/16 is rated as DDR3 1600 which means it's exactly as fast as it's supposed to be for the T530 but faster than what the T520 requires (DDR3 1333). It might be that the module(s) are just at the limit of the DDR3 1600 spec so that when they are run in the slower T520, it doesn't cause any errors. The only way to know for sure is to try two other DDR3 1600 modules in the T530 and see what happens.

[DiagonalArg]

@rkawakami - thanks for your response.

When the modules failed, was it a single error, a couple of errors or something that looked like it was catastrophic? And did the error occur right away or so time into the series of patterns (tests) that memtest86+ executes?

Just a couple of errors. I ran 6hrs and got 1 error (at 10036MB) then I ran 13hrs and got errors at 3 locations (10040MB, one time, 10100MB, 3 times and 10101MB, 4 times)

I would run memtest86+ on a single 8GB module in the T530. Let it run for at least three complete passes (assuming no errors). If it passes, then move it to the other memory slot and repeat. Do the same to the second module. You'll then have a table of how each module performs in each slot.

I've been avoiding that because, for some reason, Kingston says they will not warranty modules that are run individually. Do you know why that is?

The KHX16LS9P1K2/16 is rated as DDR3 1600 which means it's exactly as fast as it's supposed to be for the T530 but faster than what the T520 requires (DDR3 1333). It might be that the module(s) are just at the limit of the DDR3 1600 spec so that when they are run in the slower T520, it doesn't cause any errors.

The HyperX are supposed to be modules that can be severely over-clocked. I'm not doing that, of course, but I don't believe they should be over-stressed. I did find someone else who had gotten these modules for a T530 and he said they worked; though I don't know if he ran memtest to check for isolated errors as I have.

The only way to know for sure is to try two other DDR3 1600 modules in the T530 and see what happens.

Pricey.

/DA

[rkawakami]

Single bit errors sounds like a memory problem, rather than something in/on your motherboard. Typically heating the modules (i.e, letting them run a long time) would cause data access times to get worse, along with DRAM cell refresh to get shorter.

I don't know why Kingston would only guarantee modules that are run in pairs, unless they were originally sold as a matched pair. In which case, you can return both and say they don't work together. As far as I know, a T530 does not have to have both sockets populated in order to run. According to the tabook, factory builds often came with a single 4GB module. I don't believe that would change when using an 8GB module.

As for the "pricey" modules, sorry but I can't help you there. Such is the landscape when dealing with the newest and oldest forms of semiconductor memory. The latest and greatest is expensive when considering price per GB, along with stuff that's old and slow that nobody makes anymore.

[DiagonalArg]

Ok, if I'm not going to fry anything, I'm going to test them singly.

One thing you might help me clear up. memtest is running now. On the t520 it says that those 2 x 8G modules (DDR3-1600, matched) are running at "10208 MB/s", whereas on the T530 the 2 x 4G modues (also DDR3-1600, not matched as they are different manufacturers) are running at "16377 MB/s". The ratio is 16377/10208 = 1.6. But the T520 is supposed to take DDR3-1333 and the T530 DDR3-1600, so that ratio is 1600/1333 = 1.2.

What's wrong here?

[rkawakami]

Note sure why that is at the moment. The figures you cite are bandwidth numbers; how much data can be transmitted in one second. That spec can be affected by a few things.

Look carefully at the memtest86+ screens on both systems. What are on the following lines (I'm running version 4.20 at the moment; your information may be in slightly different locations):

1) First line on left under the green MEMTEST86+ <version>
After the Intel processor type, what speed (Mhz) is it running at?

2) 6th line on left: IMC:
What's the BCLK rate?

3) 7th line on left: Settings: RAM:
What's the speed (Mhz) and is it indicating Dual Channel?

[DiagonalArg]

@rkawakami - Ok, I've tested the modules individually, and I'm finding what you'd expect: one modules is bad in either slot of the T530, the other is good in either slot. So it's got to be the module. I've already called Kingston Tech support and registered a case. I'll have to wait until tomorrow to talk to the warranty department.

I'll get the info your asking for in your last post and get back in a bit. I am curious about these speed issues.

Thanks
/DA

[DiagonalArg]

Note sure why that is at the moment. The figures you cite are bandwidth numbers; how much data can be transmitted in one second. That spec can be affected by a few things.

Look carefully at the memtest86+ screens on both systems. What are on the following lines (I'm running version 4.20 at the moment; your information may be in slightly different locations):

1) First line on left under the green MEMTEST86+ <version>
After the Intel processor type, what speed (Mhz) is it running at?

I've got two different versions here. 4.20 (from an ubuntu 12.04 disk) and 4.37 (from the memtest86.com web site). Interestingly enough, version 4.20 gives more info than 4.37, including whether it's single or dual channel and the memory SPD information. (*edit: I stuck with info's as provided by 4.20.)

T520: is an i5-2520M @ 2.5Ghz.
T530: is an i5-3230M @ 2.6Ghz.

All modules are DDR3-1600. The 8G pair are identical manufacturer/part#s. The 4G pair are different manufacturers.

2) 6th line on left: IMC:
What's the BCLK rate?

3) 7th line on left: Settings: RAM:
What's the speed (Mhz) and is it indicating Dual Channel?

T520 with the two 4G modules: Memory: 10171MB/s, IMC as above (Intel Core i5 ...) BCLK is 99Mhz
RAM 664Mhz (DDR3-1329), CAS: 9-9-9-24, Dual Channel(!)
T520 with one 8G module: Memory: 6406MB/s, IMC as above (Intel Core i5 ...) BCLK is 99Mhz
RAM 664Mhz (DDR3-1329), CAS: 8-8-8-23, Single Channel
T520 with the two 8G modules: Memory: 10208MB/s, IMC as above (Intel Core i5 ...) BCLK is 99Mhz
RAM 664Mhz (DDR3-1329), CAS: 8-8-8-23, Dual Channel

T530 with the two 4G modules: Memory: 9070MB/s, IMC as above (Intel Core i5 ...) BCLK is 216Mhz
RAM 1729Mhz (DDR3-3450), CAS: 11-11-11-28, Dual Channel(!)
T530 with one 8G module: Memory: 9070MB/s, IMC as above (Intel Core i5 ...) BCLK is 216Mhz
RAM 1729Mhz (DDR3-3450), CAS: 9-9-9-27, Single Channel
T530 with the two 8G modules: Memory: 18140MB/s, IMC as above (Intel Core i5 ...) BCLK is 99Mhz
RAM 798Mhz (DDR3-1596), CAS: 9-9-9-27, Dual Channel

T530 Memory MB/s/T520 Memory MB/s:
Two 4G modules: 0.9
One 8G module: 1.4
Two 8G modules: 1.8

I would have thought it would be roughly 1600/1333 = 1.2, since the T530 calls for DDR-1600 and the T520 for DDR-1333.

(Notice that in my previous post, I had said that the T530 with the two 4G modules was running the memory at 16377MB/s. It seems I made an error(?could it be different if I swapped those 4G modules from one slot to the other?) In any case, the ratios are odd.)

Ok, I am now officially _tired_ of swapping memory modules!

[rkawakami]

I've got two different versions here. 4.20 (from an ubuntu 12.04 disk) and 4.37 (from the memtest86.com web site).

Ah, if you indeed did not mistype the web address, memtest version 4.37 from memtest86.com is basically a different program than version 4.20 (which I assume has MEMTEST86+ on the screen), which comes from memtest.org. By "different program", I don't mean it's simply a difference in version numbers; there's two separate programming teams that have released very similar memory diagnostic programs. Several years back development of this memory diagnostic was split into two forks, one retaining the original name of memtest86 and a second version with the "+". From what I remember, they selected different memory patterns to use. When I was working for Mosel Vitelic, a maker of DRAMS at the time, the Applications group ran a correlation between different memory diagnostic programs and our multi-million dollar Advantest production memory testers. It was determined that memtest86+ was doing the best job at catching memory failures than the other programs that were available.

I would recommend using only one of the versions of memtest when attempting to compare modules and systems against each other. There will be enough of a difference between the patterns used by the diagnostic, that it's possible (but rare) one program will identify an error, while the other one won't.

ref: http://memtest86.com/ <-- memtest86
ref: http://memtest.org/ <-- memtest86+
ref: http://memtest86.com/history.htm <-- brief history on the program development

[DiagonalArg]

@rkawakami - well, that explains the difference! Thanks for that history.

Though, as I clarified in the edit to my previous post, I stuck with 4.20 from the Ubuntu disk for all that data. That is mtest86+. So I think the data should be valid for looking into the speed issues.

/DA