As far as I am aware, the 360 GPUs had faulty solder connections (due to poor underfill choice by ATI that couldn’t withstand the temperature) between the chips and the interposer, not the interposer and the board, shown by the fact that a lot of red ring 360s show eDRAM errors (i.e. can’t communicate to the module on the same interposer, ruling out poor board connections). Microsoft even admitted this in a documentary they made (link), where they said it wasn’t the board balls, it was the GPU to interposer balls. A similar underfill choice is also why there are slightly higher failure rates of early Wiis, although nowhere near as bad as 360 due to the low power of the GPU on there.
It’s hard to say for certain whose final call it was to do this underfill (it’s a tossup between ATI’s design engineers and the packaging partner they chose to work with to get the TSMC chip into a final product), but at the end of the day it was ATI’s responsibility to validate the chip and ensure its reliability before shipping it off to Microsoft.
Agreed, thermals were increasing faster than most manufacturers could handle. Only real exceptions in this time I can think of were IBM (because they had to, PowerPC G5 was such a power hog it pissed off Apple enough for them to switch architectures) and Intel (because they also had to, Pentium 4 was a disaster).
As far as I am aware, the 360 GPUs had faulty solder connections (due to poor underfill choice by ATI that couldn’t withstand the temperature) between the chips and the interposer, not the interposer and the board, shown by the fact that a lot of red ring 360s show eDRAM errors (i.e. can’t communicate to the module on the same interposer, ruling out poor board connections). Microsoft even admitted this in a documentary they made (link), where they said it wasn’t the board balls, it was the GPU to interposer balls. A similar underfill choice is also why there are slightly higher failure rates of early Wiis, although nowhere near as bad as 360 due to the low power of the GPU on there.
I thought tsmc chose the poor underfill.
It’s hard to say for certain whose final call it was to do this underfill (it’s a tossup between ATI’s design engineers and the packaging partner they chose to work with to get the TSMC chip into a final product), but at the end of the day it was ATI’s responsibility to validate the chip and ensure its reliability before shipping it off to Microsoft.
I always heard that was tsmc’s decision.
I don’t know how much of it was ATI’s fault or the fab’s, but my understanding is that no one had experience handling that amount of heat.
Agreed, thermals were increasing faster than most manufacturers could handle. Only real exceptions in this time I can think of were IBM (because they had to, PowerPC G5 was such a power hog it pissed off Apple enough for them to switch architectures) and Intel (because they also had to, Pentium 4 was a disaster).