Intel must respond

The thing you must keep in mind with various benchmarks is that they are testing a SINGLE application. It's tough to make a SINGLE applications scale linearly with additional processors.

However, what quad-core CPUs will do for you is let you multitask MULTIPLE applications faster. If you're running a web-browser, downloading charts in one program and doing analysis with another and placing orders with yet another, having a dedicated core for each program will let you run the overall combination MUCH faster than with fewer cores.

And don't forget the future, virtualization. I'm running WindowsXP, Windows2003, Windows7, MacOSX, and three flavors of Linux on a SINGLE server with four quad-core CPUs. Having more cores available lets you run each instance of a VM much faster as you can dedicate more CPU resources to it. And virtualization lets you extract lineary-scaling from multiple cores & CPUs.
 
Quote from DannoXYZ:

And virtualization lets you extract lineary-scaling from multiple cores & CPUs.
More...

Not necessarily. If you hit memory bandwidth limits, more cores or virtualization wont help.

As always, system performance is a tough subject and it's even tougher to make accurate predictions.
 
Quote from dcraig:

Not necessarily. If you hit memory bandwidth limits, more cores or virtualization wont help.

As always, system performance is a tough subject and it's even tougher to make accurate predictions.
More...

Completely agree, the memory bus is the key and why SMP architectures flatten out after four processors.

Gotta move up to NUMA for any appreciable gains beyond four processors and even then, it's never going to be a linear increase due to the x86 architecture.
 
The problem with Intel's multi-core design is that each core shares the same memory through the FSB. Yes, even two cores on the same die goes through the FSB to access memory.

With AMD's design, each CPU has its own memory bus and memory. On my 16-core machine, the 8th VM runs just as fast as the 1st one.
 
Quote from DannoXYZ:

The problem with Intel's multi-core design is that each core shares the same memory through the FSB. Yes, even two cores on the same die goes through the FSB to access memory.

With AMD's design, each CPU has its own memory bus and memory. On my 16-core machine, the 8th VM runs just as fast as the 1st one.
More...
I am 99 delta that the nahalem has an almost identical design to the AMD design in this regard and AMD no longer has that advantage as of the nahalem.
 
You must log in or register to reply here.
Back
Top