13 Comments:

1. Thrax

   Cad

   
   3 Dec '04, 12:00 am

   Well done. Should be a great tool in our continued fight to easily narrow down problems for even the most clueless of users.
2. csimon

   Veteran Icrontian

   
   3 Dec '04, 10:38 am

   don't forget to dload the latest version 1.40!
3. Geeky1

   No comment.

   
   3 Dec '04, 1:30 pm

   I think saying 1 error is too many is a little extreme. I've had one or two instances where I've seen 1 error in a few hours of running the test. That's not out of the question. Errors don't happen often in good RAM, but they do happen occasionally. Every time I've come across a stick of bad RAM, it has had hundreds of errors at the very least in memtest...

   I think anything up to say, 1 or 2 errors in an hour long test is acceptable. Especially if it shows up in one run, but you re-run the test again later and it doesn't show up.
4. Brian Ambrozy

   The Icrontic Guy

   
   3 Dec '04, 1:44 pm

   No way, in a production environment (such as client computers or whatever) 1 error is unacceptable. That means that at some point, under some circumstance, you will get a memory error, and that is not acceptable to me. Good ram will run endless passes with zero errors. There have been times with I've run with the extended tests (12 tests) over an entire weekend and had zero errors - that's what I do with servers, etc.

   The new SM server that is being built - there is NO WAY I'd take it out of this office and bring it to the datacenter if there was even ONE error over a weekend-long RAM stress test.
5. Geeky1

   No comment.

   
   3 Dec '04, 1:58 pm

   You use ECC Reg. RAM in all of your client machines? Better start if you don't, seeing as how 1 error is unacceptable...

   Regular, run-of-the-mill, non-ECC, unbuffered RAM will have an error every now and then. I swear I've seen published error rates for RAM somewhere (similar to what's published for hard drives). One or two non-repeatable errors are not a big deal for a system using regular RAM. If you get the same error, in the same block of memory, every time you run the test, then yeah, it's probably on its way out. But if it's 1 error over a few hours, and it doesn't recur when you re-run the test, it's likely just a fluke.
6. Brian Ambrozy

   The Icrontic Guy

   
   3 Dec '04, 2:06 pm

   Nope, I 100% disagree with you. I handle hundreds of computers (standard, run-of-the-mill, non-ECC, etc.) for clients and I will not accept a single error, because if a single error causes a blue screen or a program crash or whatever, then that means we will have to go on site to diagnose/fix/whatever, and that costs me money.

   A standard home user is in the same situtation at a smaller scale - if they have an error, in the end it will cost them time and money.
7. Thrax

   Cad

   
   3 Dec '04, 4:56 pm

   Sorry Geeky, an error that shows up in any module is an excellent predictor for a cascading error that will spread to other parts of the memory in due time and contribute to the instability of a computer in the mean time. An error indicates something electronically wrong with the module; be it accessing with CAS/RAS, switching memory banks, or starting on a new thread, electricity is going places it shouldn't go. When that happens, the problem is bound to compound itself. You have one error, you will have more.
8. Straight_Man

   Playing with Virtual Painter

   
   4 Dec '04, 10:08 am

   Definitely, Thrax. I buy pretested and then stress test here. Although I DO go with a 24 hour test for client boxes, and host with Prime's company because he KNOWS computers. In fact, stress testing before deploying is a thing normally done with servers and those OEM'd boxes that any reliable OEM does not want to EAT his\her profit on as RAM dropouts are frequently caused by shorts and O\H results from shorts, and pre-ship stressing a system bundle is called burn-in, Geeky and others who might read this. One reason Dell does a 48-72 hour burnin of a configured system while that system is on a very thoroughly power protected circuit is to have evidence that the build is good before it is shipped.

   Most bad things show up in the first 90 days of use, and reliable OEMs want the real bad FUBARS to show up before shipping. IBM does the same thing, BTW.
9. Jimborae

   He ain't mad, just English

   
   4 Dec '04, 3:11 pm

   Prime I hate you, I thought my memory was stable @265fsb, turns out it errors unless I drop is down to about 245fsb, I've also spent the last 2 days sat in front of the PC finding the stable speed & settings.

   Seriously though, excellent guide and a great help to me. Well done
10. Brian Ambrozy

    The Icrontic Guy

    
    4 Dec '04, 5:47 pm

    Glad I could inspire hatred
11. Geeky1

    No comment.

    
    4 Dec '04, 6:35 pm

    Thrax: If that one error shows up more than once, yes, it's almost certainly an indicator of a failing stick of RAM. But if it shows up in one run of the test but isn't seen in subsequent runs, it doesn't necessarily mean anything.

    On a side note, a whole bunch of errors in Memtest doesn't necessarily indicate bad RAM. It could also be a failing motherboard. I had a friend with an Abit KX7 that was crashing all the time. Ran Memtest. Got hundreds of thousands of errors. But I took the RAM home anyhow, and threw it in my MSI Pro266TD (dual P3, DDR). Ran Memtest overnight. Zero errors. He's now using that same RAM in an A7N8X, and it's working fine.
12. TheLostSwede

    Where's the ****ing beers?

    
    4 Dec '04, 6:45 pm

    Prime,

    I agree that ram shouldn't produce errors. However, what happen if memtest is clear with your testing, but another test program, eg goldmemory produces errrors? I know for a fact that goldmemory is a much tougher test than memtest, but where is the limit? Are your limit zero errors in memtest but ok with an error in goldmemory?
    I know what you are going to answer but you know what i mean right?

    Edit// To stay on topic. I actually didn't like the article since nothing explains what modules in memtest is for mainboard, memory controller or memory. #5 and #8 is memory based, we all know that. If you are testing memory, all except 5 or 8 is more or less useless and time consuming. And as Geeky said, error in memtest doesn't have to be the memorys fault. A little explanation from the article writer in this area would have been nice. Just my 2 pesetas.
13. Brian Ambrozy

    The Icrontic Guy

    
    4 Dec '04, 10:08 pm

    Quoting Mackanz

    I know for a fact that goldmemory is a much tougher test than memtest, but where is the limit? Are your limit zero errors in memtest but ok with an error in goldmemory?

    No, an error in any memory test is unacceptable.

    Sorry you didn't like my article that much, but it was meant to be a quick-n-dirty guide (I said as much in the article) to testing your memory with Memtest. This guide was NOT geared towards power-users such as yourself. It is basically an instruction guide on how to download and run memtest. Of course people like you, me, geeky1, thrax, etc. don't need this article. This is for people who come here who are not computer experts at all, and we say "did you run memtest yet" and they say "I don't know how" or "I have no idea what you are talking about" or "I went to the memtest site and could not figure out how to download it". That's what this article was meant for. Something to point novices to, so they could download, create a bootable disk, and run memtest for basic memory diagnostics.

New Features on Icrontic: