Slashdot Mirror
Nvidia 55nm Parts Are Bad Too
JagsLive sends in a story (in somewhat inflammatory prose) from The Inquirer, which links to many others; they have been following developments in the alleged NVidia quality "fiasco" for some time. "Hot on the heels of its denials that anything is wrong with the G92 and G94s comes another PCN [Product Change Notification] that shows the G92s and G92b are being changed for no reason. Yup, the problems that are plaguing G84 and G86 are the same that affect seemingly all 65nm and now 55nm NVidia parts ... It is hard to overstate how bad this is. Basically every 65nm and 55nm NVidia part appears to be defective ... We are hearing of early failure rates in the teens percent for 8800GTs and far higher for 9600GTs ... To make matters worse, NVidia has a mound of unsold defective parts that they are going to bleed out into the channel along side of the (hopefully) fixed parts. As a buyer, you have no way of knowing which one you are getting ... Until NVidia comes fully clean on this fiasco, lists all the defective parts, and orders boxes clearly marked, you can't say anything other than just avoid them. Then again, since doing the right thing would likely bankrupt them, we wouldn't hold your breath for it to happen."
I stopped reading when I got to "By Charlie Demerjian."
Seriously, this guy is to NVIDIA as Jack Thompson is to video games. It's just not as common knowledge that you shouldn't take him seriously.
They say failure rates are "in the teens percent". Figure 20%, just for kicks. That means your chances of either card failing is 1 - (1 - .2)(1 - .2) = 36%.
For some reason that I don't understand, the vast majority of people have innate misconceptions of the rules of probability.
It's always a long day... 86400 doesn't fit into a short.
I've got one too and was wondering exactly the same thing. From what I can tell the 8600GT has an 80nm process size, so it should be safe. Which is good, I really like this card.
Give me Classic Slashdot or give me death!
I would say it's because lead-based solder actually works properly, but according to this story that doesn't seem likely to be their motivation.
Then again, since doing the right thing would likely bankrupt them, we wouldn't hold your breath for it to happen.
At that link, you'll find The Inquirer's (however flimsy and speculative) financial analysis of a full-scale Nvidia recall of the bad parts.
The Inquirer doesn't and has never claimed to be a fair and balanced news source, so they are free to put these sorts of quips on their stories. People there are pretty knowledgeable, and appear to have connections and sources in the industry, which is why people keep reading The Inquirer and don't really complain about stuff like that.
rohs has exceptions for very fine pitch stuff iirc.
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register
I'm not at all sure your criticism is based on the correct quotation source; cf: http://en.wikiquote.org/wiki/Friedrich_Nietzsche#The_Gay_Science_.281882.29 Now back to nvidia....
Yourself included, since card failure is an independent event. The chance of any card failing is - tadaa, 20%. Just like if I have 3 dice, the chance of rolling a number is 1/6. If I roll it again, the chance is still 1/6. It will always be 1/6.
I'm not sure what you are talking about with this unrelated dice example, but the GP is correct. The chance of neither failing is .8 * .8 = .64.
The chance of at least one failing is therefore 1 - .64 = .36.
36%, as the guy said. Where did you go to school again?
For your dice example, here is a more analogous one:
If I roll a six-sided die 3 times, what are the chances it will come up "6" at least once?
Chance of it not being six at all = 5/6 * 5/6 * 5/6 = 125/216 (~.58). The chance of it coming up 6 at least once is ~(1 - .58) or roughly 42%.
<Complete your profile by adding a signature!>
WRT to beef though, salmonella poisoning by beef is almost completely unheard of - chicken yes, beef no. Where this whack job got his numbers from is anyone's guess but they are wrong.
Think outside the... Hey, where'd the friggin' box go?
Charlie at The Inquirer has no credibility when it comes to nVidia.
From TFA, nVidia is changing from high lead to eutectic (tin) solder - for RoHS compliance - and has issues a PCN to that effect. Charlie has latched onto this as "proof" of his claim that all nVidia chips are faulty and overheat.
What Charlie doesn't explain is how switching from high-lead solder (5/95 Sn/Pb) to eutectic solder (63/37 Sn/Pb) - which has the lowest melting point of all tin-lead solders - is supposed to help if the chips are overheating. Nor does he explain how changing the solder material has any relationship to changing the underfill material on some mobile chips (other than they were both PCNs.) But hey, why let facts get in the way of a conspiracy theory/page hits?
I remember reading something not entirely dissimilar in Robert X Cringely's "Triumph of the Nerds". Might or might not be apocryphal; I don't have the book to hand. Apparently Intel (IIRC) were having problems. The amount of defective parts they were getting was going through the roof, and they were pulling their hair out trying to get to the root of the problem.
Finally they traced it down to the guy responsible for receiving the deliveries of the silicon wafers. Apparently he was taking out the wafers and putting them down in his desk- quite dusty and very definitely *not* up to clean room standards!- to make sure Intel was getting what they'd paid for.
"Slashdot - News and Chat Sites Deviant". (Click "homepage" link above for details).
Contrary to your belief that 'these kinds of problems are subtle and might be missed during a decent period of testing' it can be EXTREMELY difficult to find these kinds of problems. Beyond your wildest imaginings difficult.
Having worked on high performance hardware/software systems as an engineer I can tell you from first hand experience that the situation is more like there are 999,999,998 ways for things to go wrong and about 2 ways you can get it right, and those 2 ways are not AT ALL obvious. Usually the types of problems you encounter HAVE no obvious cause and no obvious solution and mostly can't be reliably replicated. They can stem from the very most subtle differences between two boards or systems. A cap that happens to be a bit out of spec and a slightly less than perfect solder joint can combine to create an error that happens 1 out of every 100 billion times an operation is performed.
Now, combine that with the fact that you have a dozen vendors slightly varying implementations of a given board design, PCs of all different types and quality levels running at different speeds with different CPUs in them, running a plethora of different versions and subversions of OS and drivers and applications, and the real miracle is you can make a board that works reliably at all.
Any attempt to make a really seriously bullet proof product that would virtually never have problems is simply infeasible. There is a law of diminishing returns involved. At a certain point you have to say "Well, we've tested it in 10 dozen different systems under 6 different OS versions with 128 different apps, and we get N number of crashes/malfunctions per hour of runtime." and then you call it a day. You could spend 10x more time and money on QA and reduce the failures to N/2, but you also won't sell much product when multiply your NRE by a factor of 10...
Plus such perfection will be for naught because MS will release BrokenOS patch "friday the 13th" 2 days later and you'll STILL be encountering the higher error rates. Same goes for new motherboards, games, etc. It is just a loosing proposition.
All you can realistically do is what they do now, test the heck out of it as best you can afford to, ship it out the door, and try to address any issues that come up later as quickly and painlessly as you can.
This is the kind of reason why military and aerospace grade hardware costs 2000x more than electronics with similar functionality with civilian retail/commercial specs. They REALLY do have to be certain things work exactly right or people die, and it is WAY expensive.
"Malo periculosam, libertatem quam quietam servitutem." -- Jefferson
Yes there is an exception in RoHS for lead solder that has a high melting point. However, the official RoHS rule is that while lead solders in general are prohibited, there is an exception allowing for the use of lead solder that contains at least 90% lead. The idea being that solder with at least 90% lead melted at a higher temperature and was at least somewhat safer if disposed of improperly. Otherwise, potentially there may also have been no replacements for high lead content solders that performed as well when the first RoHS directives were drawn up in 2003. Currently (2008), however, there are lead-free solders that would work, but the lead free solders are more expensive than lead based solders (by roughly three times). Using a lead-free solder with a significantly different composition may also require a new packaging design and another extensive round of qualification, too. I am not totally sure how this would be done.
It get worse, the new solder to be used by nVidia mentioned in this Inq article states that it will only contain 63% lead and 37% tin, making nVidia based cards with this solder not saleable to consumers in the EU according to RoHS directives. The replacement 63Pb/37Sn solder has a somewhat better tensile strength and a lower coefficient of thermal expansion than the older 95Pb/5Sn solder, which may be why nVidia chose this route to fix the problem. Whether nV will be selling very many products in the EU with this fix and whether this will correct the problems, is another issue.
Impersonating Tycho from Penny Arcade since before there was a PA.