There's an issue that it seems nobody is addressing : "availability". When you're working with a multicore architecture, a faulty processor lets you down until you get another processor. If you have a multiprocessor architecture, a faulty processor cuts your processing power to a half, but you can keep working.
I'm working in a small datacenter, and the multiprocessor approach was very useful a couple of times. If your cannot afford to stop working, maybe you have only one choice.
There's an issue that it seems nobody is addressing : "availability". When you're working with a multicore architecture, a faulty processor lets you down until you get another processor. If you have a multiprocessor architecture, a faulty processor cuts your processing power to a half, but you can keep working.
I'm working in a small datacenter, and the multiprocessor approach was very useful a couple of times. If your cannot afford to stop working, maybe you have only one choice.
AMD Opteron designs can handle cores going out and I think Intel's latest can as well.
It means nothing as far as operating systems and apps are concerned (ignoring licensing issues).
Performance-wise, it means little unless you're trying to run too many CPUs with too little memory/system bandwidth. This wasn't a problem with one core per socket, but with multiple cores and multiple sockets, the limit is easier to reach.
For example, that's the reason you don't see many quad-socket Intel servers- the FSB architecture breaks down with so many cores trying to talk to memory at once. Hypertransport works much better, and Intel's new QPI is a copy of it.
Comments
I'm working in a small datacenter, and the multiprocessor approach was very useful a couple of times. If your cannot afford to stop working, maybe you have only one choice.
There's an issue that it seems nobody is addressing : "availability". When you're working with a multicore architecture, a faulty processor lets you down until you get another processor. If you have a multiprocessor architecture, a faulty processor cuts your processing power to a half, but you can keep working.
I'm working in a small datacenter, and the multiprocessor approach was very useful a couple of times. If your cannot afford to stop working, maybe you have only one choice.
AMD Opteron designs can handle cores going out and I think Intel's latest can as well.
Performance-wise, it means little unless you're trying to run too many CPUs with too little memory/system bandwidth. This wasn't a problem with one core per socket, but with multiple cores and multiple sockets, the limit is easier to reach.
For example, that's the reason you don't see many quad-socket Intel servers- the FSB architecture breaks down with so many cores trying to talk to memory at once. Hypertransport works much better, and Intel's new QPI is a copy of it.
Holy thread resurrection batman!
I didn't even look to see how old it was.