Slashdot Mirror
Firms Get Away with Selling Untested DRAM
peppytech75 writes "Melanie Hollands in IT Manager's Journal reports that 'In recent months, some Asian DRAM memory manufacturers have been getting away with selling untested ("UTT") DRAMs. Disturbingly, the practice seems to be getting traction at the lower portion of the module business. This is being done mostly by Taiwanese DRAM makers, who are undercutting the tier-1 guys by selling untested and unmarked parts.' What's the solution here? Or is there an actual solution to what amounts to pirate companies issuing counterfeit parts?" (IT Manager's Journal, like Slashdot, is part of OSTG.)
Here is the obligitory Memtest86 post. It's a great program, and chances are that you might already have a copy on your Linux install CD depending on the distro. There are even kernel patches that allow you to avoid the bad bits if they are isolated enough.
Solution:
...as it seems products are rushed to market without significant testing. Take the Treo 650. They "tested" the device, but later found out (after release) that people who used it in the real world couldn't use the new file system because it didn't store things the same way.
DBA? Software Engineer? My company is hiring! Click
Especially when they're just reshipping returned RAM. That's when you find a new RAM supplier.
(true story - I worked at a computer store. Not that computer stores aren't guilty of reselling returned defective computer parts as new.)
Inconceivable!
Same here. I used to buy whatever was cheapest, but after the time that a series of flakey bugs was solved by switching to good quality DRAM, I'll never go back. I probably spent two days troubleshooting it, which at my hourly rate, is many times the amount I "saved" by buying cheap memory.
Blatant promotion: I've never had a bad stick from Crucial
Chip H.
If you RTFA, the author was saying that these unmarked and untested DRAM chips can later be marked as if they came from a Tier 1 manufacturer. These chips can then be sold for a premium, yet still less than the Tier 1 price. In that case unmarked and untested = pirated.
I just buy ECC RAM.
/proc/ram or /proc/mc/0, see the statistics and instantly find if the memory is seeing errors or not. Here I do see a corrected error or two sometimes, although very infrequently. But it's indeed very nice to know it's been corrected.
Sure it's more expensive, but it's great. If the computer does something strange I know that I can check
However, even if it's ECC I still wouldn't like at all knowing that it's not been tested. ECC has limits to the corrections it can make, after all.
Whenever we buy new RAM, mostly as part of new PCs, we run Memtest86. It's easy to do, it takes a while so do it overnight. There's so much that can go wrong with RAM, even with "good" RAM: it might not work together with the board, the SPD-timings might be off, whatever. Every once in a while we find some RAM that doesn't work for us and return it to the shop. We never had any problems at all to get it exchanged.
For hardware-sellers it's probably more expensive if they have to factor in a certain return-rate (and the overhead for that) so they will look to it that the RAM they buy is ok. That way market forces will work for the benefits of all of us: untested RAM will, in the end, be more expensive than tested RAM. It's much easier and cheaper to do RAMtesting factoryside than having it returned by millions of customers.
Of course that doesn't work if you buy your PC in a supermarket, but even for cheap PCs it's better to configure them yourself than buying crap. That way you can specify exactly where to save money and if anything breaks you get it fixed much quicker.
"By the way if anyone here is in advertising or marketing... kill yourself." -- Bill Hicks
The motherboard supports generating an interrupt when something happens. You can tell it to do that in case of a corrected error, uncorrected error, or never. I think Windows will BSOD when that happens, so I'd just set it to do it on an uncorrectable error. Then it will crash, but at least it will stop things before they mess up something important.
/proc you can monitor with detailed statistics about how many errors were corrected. IIRC, without specific support Linux will generate an oops, but continue if the board generates an interrupt. The patches can be told what to do in that case.
On Linux you have the ecc-linux(2.4) and bluesmoke(2.6) kernel patches, which will give you a file in
I suppose there must be some software to get all the features on Windows too, but I don't know where to get it.
But Germany is another case after all, as is the whole EU: Here every seller of technical equipment has to give a 2 year warranty to every piece of hardware he sells. So if a RAM trader buys those untested RAMs and sells them within EU borders, he has to replace every single RAM that fails within 2 years for free (except he can prove the customer is at fault).
In fact according to the law he has to give the money for the original RAM back, but he can instead try to 'better afterwards' a.k.a. repair or replace the faulty part if the customer agrees.
So there are enough specialized labs out there that can perform RAM testing in large numbers, because the RAM traders need them.
Is it really worth saving the £3-£5 by getting cheap unbranded RAM? As the saying goes, you get what you paid for.
Pardon the US prices
Crucial CT6464Z40B 512meg pc3200 $60.00 shipped
Lowest bid 512meg PC3200 $30.00 shipped
Lowest bid 1024meg pc3200 $65 shipped
What do you get with the brand name? Lifetime warranty, assurance of compatibility, known reliability. Good resale value, esp with odd chip types no longer made.
What do you get with the lowest bid? Half the price, might carry a lifetime warranty but then again they are labeled poorly so you have no clue who would even honor it. But who cares it's half the price. Grab bag buying, don't know what it is till you get it, and might not work in your board, but you can also sell it local for what you paid for it.
I have one system with crucial, one with generic.
Worth the headache? Depends on whether you can deal mucking about. But hardly a few pound difference, it's a 100% difference.
There is no sanctuary. There is no sanctuary. SHUT UP! There is no shut up. There is no shut up.
Langham Act. Putting someone else's name on something they didn't make ("false designation of origin") IS defined as Piracy. That's why selling fake Oakleys or fake Rolex watches are also called Piracy. It's also forgery, fraud and all that, but since there's not much new under the sun to make illegal, our congresspeople just sit around and spend our money thinking up new ways to make illegal things MORE illegal. After all, how can the police harrass the citezenry if they can't throw 300 charges at someone and hope one of them stick?
In this case a bunch of losers are printing "tier 1" manufacturer labels on their ram chips, and damaging those manufacturers' reputations as well as financial (guess who gets the returns when those chips, often with a lifetime warranty, end up bad?)
Of course, if this is for your games machine or something you upgrade every few months anyway, doesn't matter. But if you think that memory might stick around for a while and get used in a business critical application...well, I wouldn't, that's all.
And yes, I do buy Crucial memory. Given my dislike of rebuilding things late into the night or being stuck without working hardware, it is extremely cheap insurance.
Panurge has posted for the last time. Thanks for the positive moderations.
...is to not buy cheap-ass no-name RAM. Spend the extra 30 bucks and get some damn Crucial or Mushkin, ffs.
- A.P.
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
I was surprised at just how far companies like Kingston have to go to honor their lifetime warentee. I worked for SGI a couple of years ago and I was using a old beat up (8 years obsolete and it still performs decently!) Personal Iris 4D/35 when after a power failure it failed to boot complaining about bad memory. So I pull the thing apart and find that it has an enormous board with 16 SIMM-like slots. I pull out the offending module and notice 2 things:
1. It is obviously some sort of custom memory module unlike any I had ever seen before, and hasn't been manufactured in years and years.
2. It has a Kingston Memory sticker on the front.
So, I decide to see just how good the "lifetime warentee" is. Amazingly enough, they send me an RMA label right away and within days I have a brand new memory module and the system is back up and working perfectly! I was truely amazed that they were still willing to honor their agreement (I've had many bad "lifetime" warentees before where the "lifetime" is defined as 1 year or other BS) without complaint or hesitation.
I read the internet for the articles.
The major risk as I see it is a batch of modules gets into a major user (think IBM, H-P, and/or Dell) and fails (probably in Asia). The user goes publicly ballistic over the combination of faulty material and the supplier's inability to control the quality of its material. The press runs with it and the unlucky DRAM supplier's stock gets hammered. Some time afterward, it emerges that all the DRAM suppliers have this risk and then they all go down.
,HP or dell for not testing these products before shipping , i very much doubt that IBM would fail to run a memory test before shipping a server though.
If this hapens then its the fault of the companys such as IBM
It's virtually impossible for UTT memory to get into the supply chain of the tier one computer manufacturers because they deal directly with the tier one memory manufacturers (think Samsung and Micron). The modules that they purchase from those manufacturers are designed specifically for that company and designed for a specific memory chip. The modules must meet specifications provided by the customer before they'll buy any of them - and those specs include testing.
The tier one computer manufacturers don't go out shopping for memory and just buy a bunch of modules. They're all very specific about what memory modules they purchase and work very closely with the memory manufacturers to make sure that the memory that they need is what they get. Becaues of that, they don't have to run memory tests on the memory that they buy - it's been done at the memory manufacturer and the computer manufacturers have a very, very close working relationship with them.
I know because I'm an engineer at one of the tier one memory manufacturers and I design memory modules. The risk of UTT memory showing up in an IBM, Dell or other tier one system is zero.
But Prime95 confirms that no bit anywhere in nearly the complete memory space ever spuriously changes. I have seen plenty of memory that passes metest86 that fails prime95.
Based on my experience, Corsair will replace memory that fails prime95. Mushkin will NOT (despite a "lifetime" warranty); they basically told me that memory can't be expected to be 100% perfect all the time and that prime95 was too strenuous; if it passes memtest86 there will be no replacement. My other modules (from Geil, Samsung, and a few old no-name sticks) have always been perfect. IMO it's unconscionable to sell untested ram given how hard it is to return.
Are there any products on the market that can identify a DRAM vendor by using the data scramble patterns? Such a tool might be useful to flush out crappy DRAM.
DRAM is just a bunch of capacitors on a chip. When the chip is powered down for a while, the voltage on the caps leaks towards ground. When the DRAM is powered up, all caps are at ground. Discharge to this state can be accelerated by exposing the die to light.
Here's where it gets interesting: just because a cap is at ground on the die does not mean that you will read a zero out of the chip. With modern folded bitline architectures, half of the cells will read out as zero, and the other half as one. The pattern of 1/0 forms a definite pattern, called the "data scramble" which is a function of the chip architecture, and which will differ from vendor to vendor. Provided that few cells have been overwritten by the PC bootup, you can recover the scramble pattern and possibly identify the vendor.
Remember your old Commodore 64? Power it up, cold, and POKE 53265,59. That will slam the video chip into graphics mode. See the pattern? It's not random. That's the data scramble.
Two DRAM chips having different data scrambles are definitely not the same design. The converse is not true: two DRAM chips having identical data scrambles might be made by the same vendor, but there is a slight chance that two different vendors just happened on the same pattern. I don't know how much variation there is in scramble patterns, but this might be a useful way to trace chips to vendors.
The more technical explanation for scramble patterns: the sense amplifiers in a DRAM chip are essentially differential. The inputs to the sense amp are two bitlines. Each bitline is connected to a different physical column in the memory array. Between cycles, the bitlines are pre-charged to VDD/2. When a row of DRAM is read, one bitline is connected to the cell capacitor and receives an offset charge while the other bitline is held at the reference. The sense amp then "pulls apart" the bitlines, driving the higher one to VDD and the lower one to ground. Depending on which bitline a zero-charged capacitor is connected to, the sense amp can swing one way or the other. The exact connection depends highly on the cell geometry and fabrication process.
Past the sense amp, more fun happens. DRAMs are so dense that the signal from the sense amp requires one or two more levels of amplification before being suitable to drive to pins. To diminish crosstalk effects, the data buses are "twisted" like twisted-pair, which creates further address-dependent inversions in the pattern.
The combination of cell geometry and data bus twist create a vendor-unique pattern. It's unlikely that two vendors with two different designs will happen on the same scramble pattern.