Yes and No. In a multicore CPU, some subsystems are common to all the cores. The "computing" work is physically separated in the cores, some communications (bus) are common. I confess I have no exact idea of this design on the global performance. However, I suspect that on modern 64bit CPU's that embed enough first levels caches, the bottleneck of the computer is probably more located at the HD, RAM or network levels... It was always the case in my own experiences.

Just one tip: if possible, keep one core free for OS and JVM operations...

Although u might have already tried it, but I would recommend checking http://stackoverflow.com/ for that kind of questions. In my opinion, they are one of the best in coding.

Andreas and Kamran, writing parallel code (threading) means NOTHING if  the compiler (the JVM here) doesn't support it.  Long time ago, the catch was that Sun's JVM would only do threads and parallel programming on Sun hardware, the Intel implementation was intentionally or not, single thread. I do not know the situation now, after 10 yrs, but I believe this is what you need to check first: which JVM gives you threads implementation on INTEL architectures: is it Oracle's, IBM's, the openJDK, etc, etc.

Just for the record, parallel programming works fine since years on JVM's (not only Sun/Oracle) on Intel based platforms. I'm currently running a 12 CPU Intel based java code now on a Linux Box, using Oralcle JVM 7.xx. Beside the fact that my program is computer intensive and that the fans are running loudly, everything works fine.

Another point: there is no difference in handling Cores and CPU's (see original question) 

Thanks all for your support. Just a clarification, if I use threads on multi-core processors e.g. quad core processor, would the code be executed physically parallel?

Yes and No. In a multicore CPU, some subsystems are common to all the cores. The "computing" work is physically separated in the cores, some communications (bus) are common. I confess I have no exact idea of this design on the global performance. However, I suspect that on modern 64bit CPU's that embed enough first levels caches, the bottleneck of the computer is probably more located at the HD, RAM or network levels... It was always the case in my own experiences.

Just one tip: if possible, keep one core free for OS and JVM operations...

parallel programming is a very broad topic and there are at least two ways of thinking about how multiple cores will execute code in parallel. the first one concerns shared memory architectures. in this model, multiple cores share the same address space - thus, threads can communicate via memory references (e.g. MPP and SMP machines). another model is based on isolated address spaces (e.g. Clusters) where threads on different machines communicate using message-passing. both mechanisms are implemented in java, and a given application may use either (or mixed) approaches in order to scale performance of a parallel application.

As Kamran and the other researchers pointed out, by default the JVM will handle the allocation one its own and might use only core but as pointed out u can force it to do so as Kamran mentioned. 

If you are not sure why you need to pin a thread to a core, don't attempt doing it. OS scheduler knows what to do. In most cases, you won't even understand if there is any difference.

If I rememeber correctly you can patch the linux scheduler to do CPU affinity i.e. keep a process on a certain cpu.  Maybe by now after a couple years this is already mainstream and you just need to choose your scheduler policy?

If you are going to parallelize some operations based on some sequence of elements, I would suggest have a look to the Java 8 Stream API. By using that you can benefit from all the available cores without writing any single line of multi-threaded code. The API itself is based on some functional programming ideas, such as lambdas and monads.

I have on processor with 4 CPU's, I just would like to do a program that use 4 threads, 1 thread by CPU.

some help ?

I suppose you mean "one processor with four cores" ?

Anyway, the scheduler is OS based, and the JVM relies on the OS scheduler it is runnig on. Normally, if you have four threads, each one implementing a similar task, they will all run on a different thread. If you threads are running very fast, it could be that the first one has finished before you launch the last one for instance. In such a case, not all cores are used.

A basic discussion can also be found at http://stackoverflow.com/questions/1896065/how-to-ensure-java-threads-run-on-different-cores

You can obviously put your threads in a queue and "join" your threads if you need the results of all of them before running your subsequent code.

Ok, but I'd like to do experiments, i.e. my program have x seconds with one core, n seconds with 2 cores, y seconds with 3 cores and z seconds with 4 cores.

How to do that in a very easy way ?

Not easy in fact. You can use "CPU affinity", but this is not Java specific. The idea is to limit the number of CPU/cores used by your JVM. You'll find ressources on the web for Windows and Linux. To my knowledge, it is more difficult on Mac OS

Otherwise, you just can try your program with one thread, two threads,... instead of one core, two cores... If well written, your program should use as many cores as threads you define (at least until #threads