На последок о multicore и multiprocessors:
How Dual Core Works
As I mentioned previously, dual-core architecture puts two slightly slower cores than today's high-end chips onto a single die. By using both cores, the chip delivers overall performance greater than the highest performing x86 processors today. The singular advantage is greater throughput with lower power consumption and lower cost.
Single-threaded applications, such as those found on desktops, do not derive the full measure of improved performance because they run on only one core. However, by providing a second core for the operating system to perform its internal functions, this configuration still delivers benefits to the single-threaded application, which now can run without being swapped in and out by the OS. In other words, on every desktop there are always multiple threads active at once, some created by the operating system (like Linux, Solaris, or Windows), and others by applications. On a dual-core system, two threads can run in parallel, giving both the operating system and applications plenty of processing resources.
That's just the start, of course. The real bonus occurs when multithreaded software, such as workstation and server applications, run on the new processors. Now the individual threads can run in parallel and, combined, do far more than a faster single-core processor. Hence, server applications will see the greatest lift from dual-core chips, and this is the reason why AMD is targeting servers in its initial roll out. This is where the benefits are most important and most evident.
Even though multicore is conceptually similar to multiprocessor, it is important to point out that the architectures are distinct. Let's examine this difference.
Multiprocessing vs. Multicore
What distinguishes technologies dual core and multiprocessing is the extent to which processor resources are duplicated rather than shared. On multiprocessors, of course, you have two separate processors, so everything is duplicated. At the other end of the scale are technologies such as Intel's Hyper-Threading Technology, where some parts are duplicated and most parts are shared in a single processor. This high degree of sharing limits the delivered improvement in HTT performance.
Dual-core chips are closer to multiple processors, both in terms of design and performance. This is because they mostly duplicate processor resources (two separate cores, after all) and share very little—primarily they share connections to the rest of the system. The critical resources such as cache are indeed duplicated. The result is performance that is significantly better than a single processor, although less than dual-processor systems. For this reason, AMD correctly positions these dual-core processors as significant upgrades to single-core chips, rather than as single-chip replacements for dual processors.
And as an upgrade, the dual-core Opteron chips have an impressive feature beyond the greater performance: The physical package in which they ship plugs directly into the socket used by single-core Opteron processors. As a result, upgrading consists of replacing the chips and updating the BIOS (for Rev E processors and later). Such easy upgradeability is unheard of in the processor industry. When was the last time you remember being able to upgrade to a different processor architecture without having to buy a whole new system?
The upshot is AMD's dual-core Opteron processors give you a heck of a lot more performance for the same price and the same level of power consumption as single-core Opteron chips. If you want more information on this technology, go to http://multicore.amd.com/.
