RAM Testing Procedures By Manufacturers

M

msny

2[H]4U
Joined
Sep 5, 2001
Messages
2,197
So many of us come here, or hardware forum, and seek advice about RAM
problems, BSOD's, shutdowns, other isseus, ect.

So we all jump in and say try the Memtest/Prime95 tests, and we do.
We then find errors, as we test our sticks.

I've built several computers over the years and had my share of RMA's.
I've used all brands, and my latest bout was 8gigs (2 x 4) bad out
of 16gb. Another RMA for G. Skill, now there all good.

I've also had sticks with no issues at all, they test just fine, and run
for years.

So I have a few questions for manufacturers, how do you test before
they go out the door? All, random, few, by lot?

Am I just unlucky, and the failure rate is low?

Are there external factors, like MB's themselves that contribute
to a sicks failure?

I'm just curious, because they all, for the most part, have lifetime
warranties.

Thanks for your input.
 
I'm having trouble finding that information myself. :(

This video from APC (Australian PC mag):

http://apcmag.com/inside_the_kingston_memory_factory.htm

says Kingston makes 3 grades of DRAM modules:

1) System Specific -- for major brand PCs, 100% tested
2) ValueRAM -- "competitively priced, but not 100% tested"
3) HyperX -- "the good stuff", 100% tested

Maybe that explains the 65% failure rate for the 512MB PC3200 Kingston ValueRAM I tried, several years ago. OTOH 100% of the Kingston ValueRAM RMA replacements I've known about worked fine, so it's possible they substituted HyperX or System Specific.

Spectek, which apparently takes wafers that were rejected by their parent company, Micron, has its own video that says their modules are just sampled for final testing.

There are also memory test farms, that is, companies that test modules, and I saw a photo of one of them using an ordinary motherboard and Gold Memory. It wouldn't surprise me if MemTest86/86+ is sold to such companies, too.

It seems that most companies use fairly cheap testing equipment that costs only a few thousand dollars, and most of the price may be due to the inclusion of a mechanical device to automatically load and unload modules. Or they may do final testing with just ordinary motherboards, perhaps running UltraX PHD RST Pro cards, but often at just room temperature (Corsair indicated they test at 50 Celcius, but that's way below the 70C rating for first-rate modules and the 85C rating for the DRAM chips themselves). Only one company, KingMax, clearly used first-rate equipment, an Advantest machine, like the kind I once saw on Ebay for well over $100,000, used.
 
Last edited:
I think your may be onto somethong.
I've seen nothing on testing procedures at all.

Wheres the corsair guy, when you need him?
 
When I got my Radioactives from Mushkin, I was told by them that they use Memtest86+ for testing purposes. Not sure what more they do or use, but at least there's something we can get our hands on.
 
i would think it would be like a production line, the modules are put into some device that sends some signal through it and based on a number it passes or fails, much like LCD's tv's, each one is connected specially to a system that runs some tests and and it passes or fails, not like someone sits and connects cable and watches some channels tv by tv.
 
MrGuvernment said:
i would think it would be like a production line, the modules are put into some device that sends some signal through it and based on a number it passes or fails, much like LCD's tv's, each one is connected specially to a system that runs some tests and and it passes or fails, not like someone sits and connects cable and watches some channels tv by tv.
Click to expand...

This. I know an EE who worked on designing memory test machines - basically a faster automated memtest, although there are other signal quality checks etc. you can't do in software. I do not know if this is used by companies that do not produce their own chips, or what % of chips undergo testing.
 
I wonder what the cost benefit analysis looks like for the manufacturers, in terms of increasing testing vs. having to incur the cost of a lifetime warranty
 
joelthepc said:
I wonder what the cost benefit analysis looks like for the manufacturers, in terms of increasing testing vs. having to incur the cost of a lifetime warranty
Click to expand...
It's obviously more expensive than the cost of gaudy heat spreaders.
 
Industry wide, what percentage of RAM goes out the door as bad?
 
If you mean goes out the door as in bad and needs to be destroyed, then probably very little. I would surmise that the company just downgrades its rated specs and sells it as lower quality. If you mean putting defective RAM on the market... well I have just gotten my sixth kit in a row that appears to be mis-speced / non-functional and I am not happy about it.
 
bholstege said:
This. I know an EE who worked on designing memory test machines - basically a faster automated memtest, although there are other signal quality checks etc. you can't do in software. I do not know if this is used by companies that do not produce their own chips, or what % of chips undergo testing.
Click to expand...

For testing out the initial capability of a RAM, I would think that they would do as you say above, namely, determine the operating capability of the RAM ic's in a variety of situations (including thermal, electrical, possibly magnetic, etc). I would also assume that they separate the tests to two specifics, the testing to operation (ie determine what parameters provide the best operation (bandwidth) of the RAM), as well as failure analysis (determine what parameters provide the limit conditions prior to reaching the catastrophic failure limit). If the company is doing industry standards, I would assume they are using an acknowledged analysis technique such as Six Sigma to determine the quality of the chip.

However, after all of this is done and the basic capability and robustness is determined, I would expect that the engineers are tasked to find a method of quickly doing some fast tests that basically will determine if an IC is a go or no-go. This is because to keep costs down, one needs to improve yield (wrt manufacturing time). Then they probably contract companies such as Aetrium to manufacture a gravity based tester that would test the chips to their requirements.

http://www.aetrium.com/model_55v6.html

They will probably build a binning architecture tio then bin the IC's depending on their capability.

Then the survivability of the stick of RAM would depend on the accuracy of the tests they done, the aggresive nature of where they specified the IC performance (ie. is it 3 sigma from the failure limit or less), the capability to tighten their processing capability (making sigma a smaller number), and their correct nature in making sure their QC quick test is representative of the real quality of the IC. If done right and if the chips are spec'd conservatively, then the survivability is probably quite high. If not, then probably the number of RMA's will increase.

BTW, the performance of a chip does not neccessarily correlate with the durability of a chip IMO. Oh and I would think much of the RAM survivability does not depend on the stick manufacturer like Kingston, GSkill or Corsair but rather the chip manufacturer such as Infenion, Samsung, ProMos, Elipida and Micron.
 
You must log in or register to reply here.
Back
Top