how does prefetching work?

[stealthy123]

I don't understand how they work at all and more specifically what did intel do to make the prefetching on core2 so impressive?

 

[veek]

There's a great article at BeHardware that provides quite an indepth look at the Conroe architecture.
I highly recommend reading it. The beginning of the article talks about the transition from Netburst
and contrasts the two architectures.

http://www.behardware.com/articles/623-1/intel-core-2-duo-test.html

The issue of prefetching is discussed on page 6 which you can access from the review index pull-down
menu option "Caches, memory, and prefetch."

Here's a direct link to that page:

http://www.behardware.com/articles/623-6/intel-core-2-duo-test.html

Hope this helps

 

[chrisf6969]

prefetching works to eliminate latencies associated with getting things from system memory, by predicting what you're going to need then already having it in the L2 cache.

Here's a STUPID layman example!

You train your dog to get the newspaper and slippers in the morning when you wake up. But it sucks b/c you have to wait at your bedside for hime to go get the slippers from downstairs & the newspaper from outside. ( = latency / waiting)

Well if your dog was "smart" and had prefetching, it would fetch those things for you, before you woke up, so soon as you opened your eyes they would be waiting for you next to your bed (L2). (less latency)

 

[hokatichenci]

Not a bad example... but to be more realistic it would be something like, there isn't one paper, but say theres a few dozen from different parts of the world. Imagine the case where every day you try the next region (assume theres an ordering). A smart dog would be able to either pick up on this or be programmed to just bring you either the next one, or a bunch of ones in the area it thinks you'll ask for next.

 

[Met-AL]

prefetching works to eliminate latencies associated with getting things from system memory, by predicting what you're going to need then already having it in the L2 cache.

Here's a STUPID layman example!

You train your dog to get the newspaper and slippers in the morning when you wake up. But it sucks b/c you have to wait at your bedside for hime to go get the slippers from downstairs & the newspaper from outside. ( = latency / waiting)

Well if your dog was "smart" and had prefetching, it would fetch those things for you, before you woke up, so soon as you opened your eyes they would be waiting for you next to your bed (L2). (less latency)

Woof!!!!