> that due to multi-core systems slowly changing to many-core systems, a > networking model is very scalable and with so many things to break, the > fault tolerance will become a must. This could allow then for computer
i'm not sure i follow along. since i've been paying attention, processors have ranged from ~6e4 transistors to 2e9 transistors. over than range, processors have not gotten 1e5 times less reliable. as a rough first-order guess, reliablity has remained steady. it's also important to remember that unless you have a z/os mainframe, fault-tolerance within the box is a pipe dream. i've never seem a white box pc with redundant motherboards. and motherboards are up near the top of component failures. on the other hand if you want to scale to 1e5 commodity-style systems like google and others do, you're right, fault tolerance outside the box becomes interesting. but that doesn't relate to gpu computing. - erik
