The inquirer has been carrying this story for a long time. My understanding is that the GPU and substrate bond (I assume they're using some sort of epoxy, or some sort of underflowed bonding material) was slightly off-spec. That means that it doesn't share the same coefficient of thermal expansion as the substrate, and the GPU.
Everyone knows that Heat-Cold-Heat stress shocking causes extreme wear on that bond, and it's apparently just a matter of time before the affected parts fail. It's a problem I work with every single day in the CCD fab I'm part of.a There was speculation for a while that Nvidia knew about this since the A1 and A2 revisions of the 8600 GPUs had different core voltages, which was rather curious.
Reader Comments (Page 1 of 1)
不笑的老Ｋ @ Jul 11th 2008 1:52AM
『是 G84 和 G86 共用的 ASIC 使用的黏著材料（之類的）在熱膨脹系數上和其他組件不相同』
恕我問個外行話，請問這是封測廠還是晶片商的問題，如果純粹是晶片商的問題，那 Nvidia 現在的股價再打個對折也不過份；如果是封測廠的問題，那搞不好事態還會更嚴重。
Roygbiv @ Jul 11th 2008 1:54AM
其實…我也沒有搞得很懂。下面是主站一位 nerdtalker 的回應，看起來應該能說明一些事情吧 XD
The inquirer has been carrying this story for a long time. My understanding is that the GPU and substrate bond (I assume they're using some sort of epoxy, or some sort of underflowed bonding material) was slightly off-spec. That means that it doesn't share the same coefficient of thermal expansion as the substrate, and the GPU.
Everyone knows that Heat-Cold-Heat stress shocking causes extreme wear on that bond, and it's apparently just a matter of time before the affected parts fail. It's a problem I work with every single day in the CCD fab I'm part of.a There was speculation for a while that Nvidia knew about this since the A1 and A2 revisions of the 8600 GPUs had different core voltages, which was rather curious.
Chris Han @ Jul 11th 2008 2:27AM
個人認知這件事封裝廠和晶圓代工廠要負的責任應該比Nvidia大很多, 尤其是封裝廠. 因為很顯然這是封裝材料和晶片本身熱膨脹係數差太多所產生的問題. 假如我沒記錯, 這些晶片所採用的是Flip-Chip封裝, 換言之, 熱膨脹係數差異產生的問題可能是晶片上的金屬凸塊係數和預期有差異, 但也有可能是封裝材料(那堆環氧樹酯之類的東西)的膨脹係數跟原始設計有差異. 這兩種可能狀況, 封裝廠都要負大部分的責任. 另一種可能的狀況就是金屬凸塊跟晶片本身脫離, 如此一來晶片代工廠就會被牽拖到.
不笑的老Ｋ @ Jul 11th 2008 4:43AM
翻閱了一些資料，G84、G86 的狀況看來像是 Die 與承載基板間接合的問題，這個部份看起來應該是封裝廠的負責範圍，但 Nvidia 對這類材料的 spec 肯定有作過認證，所以應該不會全部由封測廠負責，因此這個事件最有可能的結果是，晶片商與封測廠各打三十大板後全面回收解決，如此會是對市場生態影響最低，對終端消費者的傷害也最小的解決方案。
比較糟糕的情況是，如果肇事廠商不願意拿出「有誠意」的解決方案，而是採用更新 BIOS、驅動程式一類「拖命撐過保固期」的方法，那恐怕損失的不僅是提供給憤怒的消費者 RMA 的費用，還會連帶賠上自己多年來累積的商譽，更嚴重者，還會對目前市場上還在銷售的 G92 與 GTX200 系列都會造成負面影響，到時損失的恐怕就不只是當年該認列的一次性虧損，還要斷送消費者對其未來所有產品的信賴感。
以上，都假設這則新聞是真的，但基於市場健康與良性競爭，實在衷心期盼這只是個不實的謠言。