Editors Note: From time to time, SemiAccurate will be republishing some older articles by its authors, some with additional commentary, updates and information. We are mainly reprinting some of the oft referenced articles that originally appeared on the Inquirer. Some will have added content, but all will be re-edited from the originals as per contractual obligations. You may see some slight differences between the two versions.
This article has had some of the original links removed, and was published on Thursday, August 28, 2008 at 7:47PM.
HOT ON THE heels of it’s denials that anything is wrong with the G92 and G94s comes another PCN 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.
This PCN is very similar to the one linked above, and the formatting is almost almost exactly the same, so we won’t cover all the details, just the pertinent points. This one is much more important, it confirms that the problems are not confined to the 65nm products. Since Nvidia told us the last one was unimportant and refused to give it to us, we didn’t bother asking this time, we just took notes when they were shown to us at a recent conference.
It is titled “G92 GPU Desktop Products” with a subtitle of “Change Bump Material from High Pb to Eutectic Solder”, with a date of June 2008 and a number PCN0346A on it. Page 2 has the “PCN Submit Date” of June 13, 2008, “Planned Implementation Date” of July 28, 2008, and a “Proposed First Ship Date for change” of August 17, 2008. Short story here, if you have a G92 or G92b purchased before next week, you likely have a lemon. Remember, these are chip ship dates, not boards in stores.
The next few chunks, “Change Category” and others are the same, “Class 1”, give to everyone under the sun, and OMGWTFBBQ. That is kind of a ‘well duh’ thing, and is exactly the same as the G86 part PCN.
The big one is the affected parts list. It clearly states that not only are 65nm parts bad, but 55nm ones are as well. The entire list of affected parts is as follows.
Small batch, my arse
Lets see, what do we have here? It looks like they changed the bumping material on the 55nm parts a month and a day after introduction. Yup, no reason for that at all, nothing to see here either.
The next part is a description of what we already knew and told you about on the last PCN story. To use their words, “NVIDIA will transition from using high-lead solder (95%Pb/5%Sn) to eutectic solder (63%Sn/37%Pb) flip-chip bump material for the G92 product family. During the transition period NVIDIA will be supplying both high-lead and eutectic bump until inventory is depleted. No other materials are being changed.” This makes complete sense, and it is followed by a picture of a modern chip with the bumps and underfill pointed out.
The reasons are the same, supply and robustness, as is the impact statement. Same very curious wording. Nothing new, just bad news.
The “Implementation and Qualification Plan” however does have some new news. It says, “NVIDIA has previously qualified numerous products using eutectic solder bumps using the same bump suppliers, substrate vendors, underfill and assembly sites as this device. Qualification data is available upon request.”, information that backs up our previous assertions that this is quite widespread among all their 65nm and 55nm products. Qual data is available “Now” and samples on July 1, 2008.
Page 4 has the same diagram, and indicates that the eutectic bumps are marked the same way as the G86 ones, with a trailing R on the lot #. Because it is etched on the die, you have no way of knowing which one you have until you take it apart, pull the heatsink, clean off the thermal paste, and read the laser-wielding chicken scratchings. Most stores won’t let you do this, and NV is going to be mixing the dies up until they burn off inventory. This means you won’t be safe until long after the card is irrelevant, say later in Q4.
The “Recommended Action” and contact info is the same as the G86 PCN, and the Revision history has an Initial Release date of 06/13/08. There is no blank page 5 on this one, it is just the disclaimer that was on page 6 of the last one.
While Nvidia is playing these PCNs off as nothing to worry about, they are. The fact that their defective chip problem extends to the G92 line like we said earlier is bad enough. It pretty much confirms that the problem is the same as the “Small batch of EOL laptops parts only given to HP” that they warned about in July. The bigger problem is that it affects the newer 55nm parts as well. Those were supplanted in a number of days you could almost count to on your fingers and toes if you grew up in a small town in Appalachia, never a good sign. In fact, qual samples were available before the 9800GTX+ actually launched.
It is hard to understate how bad this is. Basically every 65nm and 55nm Nvidia part appears to be defective. It is not a question of yes or no, but how defective each line is, and what the failure rate for each one is. We are hearing of early failure rates in the teens percent for 8800GTs and far higher for 9600GTs, so this is not a quibble over split hairs.
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, and it looks like Nvidia isn’t keen on helping you figure it out either, that would cost too much.
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, I wouldn’t hold your breath for the right thing to happen.S|A
Latest posts by Charlie Demerjian (see all)
- You can now buy bare Snapdragon SoCs with 410E and 600E - Sep 28, 2016
- Spin Transfer Technologies talks about their ST-MRAMs - Sep 27, 2016
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016