More physical processors and less dies per processor are generally able to be more efficient for a number of reasons, but more expensive and wasteful of power, and typically are biased towards server implementations and opteron.
The more-cores-approach â they can have more cache per core, and in the case of the Opteron, they have more memory controllers per core.
Its a matter of memory starvation, memory locking, caching and memory latency/bandwidth.
On the opteron, now struggling in raw performance per die, scales nicely because for every die you have a dedicated memory controller and the advantage of local memory in a NUMA system.
Intel implements a shared cache for every 2 CPUs, even on quad core, so a 12MB cache quad core is two 6MBs, 6MB shared for 2 CPUs.
AMD implements a cache per core regardless, but its a lot smaller. This isnt a good thing.
Now, the intel, not having a memory controller per physical chip, has a shared memory FSB. So if you take a dual socket system, and put 8 cores in there (2 quad cores), youâll have a really find system with massive memory starvation, as you get the dual channel/interleave but you go through the intel MCH to get to memory, so thats 8 CPUs lining up behing one MCH. The reason for the giant intel caches.
However, intel with the MCH, can get to new memory technologies a lot faster than AMD since its only changing the MCH and not the CPU.
In âreal world/officemarkâ or gaming benchmarks, it is unlikely that you would see the benefit to using two dual cores vs one quad core. On the intel side it may not matter at all since the cache size per 2 cores can be made the same and there is still one MCH per two sockets.
Id get a single quad core and rejoice in the simplicity.