It's not that strange that 3 generations are affected, since it took about 2 years for the first signs of problems to surface. That's the problem with these issues... the hardware doesn't break down immediately.
The same can be said for the XBox360.
I'd still say that this is more of an Nvidia specific issue. At least such a wide spread, all encompassing defect. I believe around 3 generations of GPUs and chipsets are affected because of their underflow material of choice, and solder composition.
It's not a one-way street.
The nVidia engineers have to be informed and advised by the manufacturing company about the possibilities of their manufacturing process and an indication of the differences between materials.
I was responding to your earlier point that nvidia has no idea what the manufacturer is putting in the chips. It's absolutely not possible to design an advanced GPU without knowing exactly how all the materials behave so yes, nVidia has expertise in the area combined with TSMC's - but ultimately nVidia chooses the design regardless of TSMC's advice or test results. That's why they ended up with high lead/substandard underfill in that generation, and single via's instead of double via's in this generation.
Yes, nVidia makes the final decision, but they do so based on TSMC's advice.
I think we can assume that nVidia's engineers use logic to arrive at their design decisions, so they wouldn't purposely choose a material if they know beforehand that it's not going to work.
So I have problems with people trying to paint a picture that nVidia did it on purpose, or that nVidia's engineers are incompetent.
There was probably some kind of miscommunication between TSMC and nVidia. Yes, ultimately nVidia is responsible, and nVidia took that responsibility. But that's not the whole story, it lacks various nuances.
I wish I had something that has died for no reason that I could do this to. I got a motherboard with blown capacitors, dead PSUs, a motherboard with huge black burns from lighting... booo no science for me!
I still cannot get behind your idea that nVidia doesn't know the properties of the materials involved.
What I meant is that nVidia goes by what TSMC tells them the properties are. After all, TSMC is the one that develops the fab, manufactures nVidia's GPUs, and gives nVidia feedback on the production process. The materials don't operate in a vacuum, so nVidia can't really study the properties without TSMC's help. Even if nVidia did have its own fab, it wouldn't really help them, as their fab may not perform the same as TSMC's.
As my previous post should indicate... yes they use logic to make decisions, and as you say, there are logical reasons for choosing high lead bumps.
Hindsight is always 20/20.
If nVidia knew beforehand that a large part of their GPUs would die before the warranty was over, and it would cost them millions of dollars to service them, do you really think they would have chosen this material? Ofcourse not.
I can only say that I appreciate it when companies take their responsibility and replace broken hardware free of charge. Intel did the same with the Pentium fiasco back in the day, also a problem on a very large scale.
I was disappointed to find out that AMD did nothing about the Barcelona TLB bug other than a performance-sucking bios workaround. They should also have taken their responsibility and replaced the CPUs with a fixed stepping free of charge.
nVidia actually does choose the substrate material, it's like choosing cloth or leather seats - even if the fab is outsourced the materials choice and design layout is entirely nVidia's. The bumpgate issue was actually a combination of poor bump layout combined with poor choice of substrate material. Both are engineering decisions nVidia made.
Fwiw everyone I know who owns a 8600-8800 has had theirs die now and had to be RMA'd (my buddy's 8800GTS just died last week was the last one to go), and these are the only video cards we've had to RMA since we started with 3dfx Voodoo and Riva TNT cards.
I find it amusing that you lump together a near catastrophic physical failure which affected almost all of nVidias chips with the TLB bug something that only affected a minority of users. Now thats effective focus switching.
What I meant is that nVidia goes by what TSMC tells them the properties are. After all, TSMC is the one that develops the fab, manufactures nVidia's GPUs, and gives nVidia feedback on the production process. The materials don't operate in a vacuum, so nVidia can't really study the properties without TSMC's help. Even if nVidia did have its own fab, it wouldn't really help them, as their fab may not perform the same as TSMC's.
As my previous post should indicate... yes they use logic to make decisions, and as you say, there are logical reasons for choosing high lead bumps.
Hindsight is always 20/20.
If nVidia knew beforehand that a large part of their GPUs would die before the warranty was over, and it would cost them millions of dollars to service them, do you really think they would have chosen this material? Ofcourse not.
I can only say that I appreciate it when companies take their responsibility and replace broken hardware free of charge. Intel did the same with the Pentium fiasco back in the day, also a problem on a very large scale.
I was disappointed to find out that AMD did nothing about the Barcelona TLB bug other than a performance-sucking bios workaround. They should also have taken their responsibility and replaced the CPUs with a fixed stepping free of charge.
Right, if nVidia knew beforehand they wouldn't have done it - that's a failure of engineering. They get test parts back from TSMC before they enter full scale production, they get to test every aspect of the GPU they get back, thermal expansion, stress, hotspots, voltage leak, everything. They failed to recognize the stress induced on the parts. If they felt the properties of the high lead bumps was misrepresented or below the given specifications they would have sued TSMC. We just saw how AMD chose to double up on vias after their test chips came back from TSMC. They know what they're working with.
Phenom TLB bug, Pentium FP bug, also missteps, and it would be nice if they offered exchanges for those effected but realistically 99% of users would never encounter the TLB bug and didn't need the fix, and most users wouldn't actually be effected by the FP inaccuracy either.
It's not that simple.
As I said, it took about 2 years until the first GPUs started failing 'in the wild'.
You can't really test for those kinds of situations.
Reminds me of the anecdote about the Space Shuttle. NASA still uses original 8086 chips. One of the main reasons is that they know how reliable these chips are. They could use newer chips, but nobody knows how long those last, especially in outer space conditions. Not even Intel can tell you how reliable their latest CPUs are. It's one of the biggest mysteries in the world of chip manufacturing.
But I guess we're not going to agree on how much engineering goes into these parts
It isn't actually.
It's a common problem with many companies when they switched to ROHS regulated materials.
http://www.humus.name/index.php?page=News&ID=283
This is a nice and funny thread. Poor video card. That baby was a winner 2 and half years ago...
At least you have a ubber modded HSF.. Alto stock will do just fine. gl