I got this little note:
I was thinking what if, and this is a big IF, what if nVidia’s new chip NV30 somehow proves to have a fatal flaw in it? I mean like boards dying after two weeks use, or if it starts showing graphics error in any app/game after 10 minutes or if the drivers crash like crazy and they can’t fix it for several weeks.
I’m saying, WHAT IF? What if a great problem occurs, and can’t be solved for a long period of time, say 4 weeks+? Since the NV30 is already delayed…
Will this affect nVidia greatly, or merely dent their rep? Will OEMs shy away from nVidia chips after such a failure, based on the pretense ‘nVidia chips only crash/break/die/whatever’ even if the problems only relate to the NV30? ATI will probably celebrate in secret, but how would this affect them?
I hope I give you give you inspiration to rant away now, I like reading your rants. Sorry, I meant to say ARTICLES of a more non-review nature. J
As I recall, you have several times talked about this ‘what if’ situation concerning AMD and the Hammer. Here’s your chance for something new. 😉
This gentleman has previously expressed some displeasure over my Radeon 9700 articles, and I guess he isn’t exactly in love with a few of my Hammer comments.
Under normal circumstances, I would have explained that SOMETHING has to start the process of me wondering about a product.
Sometimes it better to be lucky than good. It just so happens that just the day before, there IS a piece of news about nVidia that does need to be addressed.
Does nVidia Have A Yield Problem Now
You can read something to the contrary here:
“The yields on 0.13-micron NV30 chips aren’t high at all unfortunately; currently yields are between 10 and 20%, meaning that for every 10 chips made, only 1 or 2 are actually functional. Once again, to put things into perspective, neither AMD nor Intel would ever produce and ship silicon with yields as low as 3 – 4 times that.”
Unfortunately, the author the page before also said this:
“During NVIDIA’s quarterly conference call, it was publicly confirmed that NV30 had yet to tape-out at the time. For those not familiar with the manufacturing terminology, to “tape-out” is to produce the first functional silicon based on a processor’s design; prior to being taped-out, a processor is only functional in the form of simulators.”
The conference call was at the end of July. You can’t make the things until they tape out. The article says tape-out didn’t occur until last week. So TSMC is one week into doing this. That means first silicon or close to it. Guess what? First silicon isn’t usually very good. For instance, six months ago, AMD was crowing about getting functional silicon out just a month after first silicon, and the author nodded approval.
Even worse, the article took it as a given that TSMC’s yields couldn’t get any better.
This is like giving the final exam to students the first week of the college semester and basing the student’s final grade solely on that score.
Oh boy, wonder how many would accept those rules?
Of course, you teach the class, give the class the opportunity to learn about and get better on the subject, then you give the final exam.
It’s just the same at a fab plant. That 100 days or so is meant to get everything worked out; get everything functional, get the bugs out, get the yields up.
Could nVidia End Up With Problems?
That’s a different question with a different answer. The answer to the second question is “Quite possibly.”
Finding that initial yields is no smoking gun doesn’t mean nVidia gets a clean bill of health, though. It’s just too early to know one way or the other. There are other grounds to be concerned about the NV30.
Implementation of .13 micron technology has generally proven to be a headache which has usually lengthened the period of time to reach post-tapeout fabrication. Look at AMD. They rushed, and they’ve apparently had to do a do-over.
The problem is not how you’re doing after week one. You expect to be doing badly at that point. The potential problem is finding out eight weeks down the road that you’ll need more than thirteen weeks to get it done right, and the powers-that-be tell you you now have less than thirteen weeks to get it right.
Time is not what nVidia has right now if they want to get NV30 into the stores for Christmas. There’s going to be tremendous pressure to get this out the door and cut corners.
It’s not how well or badly first silicon does, it will be how well or badly the GPUs made three months from now will be. If they’re not quite where they’d like to be then, (and nVidia couldn’t afford to produce these chips at a 10-20% functional yield rate, they get less than $100 for even the highend GPUs), then maybe some of the standards will get relaxed a bit, maybe some bugs won’t get fixed, maybe some rather marginal chips will get through.
What might that mean? It might just mean the GPUs will be binned out of necessity, and you won’t be able to overclock the GPU on a low-end GF5 very much. It could just as well mean some bad video cards.
It is a matter of concern. It’s something that needs to be looked at before the card comes out and monitored after it comes out should the GF5 show up sooner rather than later.
But the GPU is hardly damned yet.