3 ms·
Desktop applications aren't really ready to take advantage of that many cores yet - parallelism is hard, and the vast majority of desktop users probably can't k
by bdonlan 14y ago
Desktop applications aren't really ready to take advantage of that many cores yet - parallelism is hard, and the vast majority of desktop users probably can't keep even two cores saturated most of the time. There's also thermal dissipation limitations, plus with that many cores you need to have a lot of bandwidth to memory to keep them saturated. And, of course, the more circuitry you have the lower your yield, so the cost of a $BIGNUM-core CPU would be quite high. So there's no real reason to add tons of cores to a single CPU.
Meanwhile, on the server side of things, there's only so far you want to go with multiple cores before it's worth thinking about just buying more than one server. Multiple servers has advantages when it comes to reliability, helps scale out resources other than CPU (eg, memory bandwidth, hard disk bandwidth, or NIC bandwidth), and of course this forces you into an architecture that scales out much further than just adding more sockets to the same machine does. And, of course, using the same basic product lines (modulo clock rating and the like) as desktop chips reduces cost too. It's not really worth adding tons of cores to the same package, when you can always buy multiple CPUs if you really _really_ need tons of cores on one machine.