| Thoran |
03-28-2006 01:38 PM |
Multiple processors in any flavor (dual core or true MPC) only provide benefits if the software and OS are designed to take advantage of them. Dual Core systems will provide some advantage to non-mpc software because Windows is able to use multiple processors... but in applications like games if the software is a single-thread design you will not see much of a performance difference.
Even when everything is designed for multiple processors you will not see double the performance because of limited access to resources. Typically one processor has control of IO... so things like disk and video access HAVE to go through that one processor. In addition, Intel processors all share memory... so memory performance becomes an issue (hence the larger caches).
AMD's mpc architecture when fully implemented has the potential to scale significantly better than the Intel architecture. On my Boxx system each Procesing chip has its own memory stack, so only the on-chip processors need to share memory. Even so, the architecture requires planning to work well... you need to make sure that as much as possible, data needed by processor 1/2 is available in the local memory block instead of in the memory of processor 3/4. There are OS level tools (affinity) that help with this, but software designed to do this will always work better.
The upshot is that just because you've got 4 processors will not mean your software is 4x as fast... unless it's also designed to scale well.
|
