AMD Phenom versus Core Quad question

[thilanliyan]

Take a look at the "Multi-core Efficiency" screenshot of Sisoft Sandra:
http://www.cpu3d.com/review/59...oard/overclocking.html

Why is the core interconnect (bandwidth and latency) so slow in the Phenom...doesn't it have some direct core-to-core communication (ie. crossbar)? Doesn't the Core Quad communicate over the FSB in which case it should be slower? Am I missing something?

 

[SlowSpyder]

I thought in real world apps the Phenom actually scaled a bit better then the Intel quads? The only thing I can think of is that I belive the Phenom cores more or less have to communicate through the L3 cache (could be wrong on that).

I think this link shows that they do. Since the L3 runs at a slower clock speed then the cores do that it causes some problems.

 

[Idontcare]

I have yet to find a single real-world application that responds to bandwidth metrics as Sandra does. Graysky and I even tried, twice, looking for the imrpovements that Sandra was showing us we should be seeing.

I mean is it just me or has anyone else found Sandra memory bandwidth tests entirely pointless in every aspect of the terms?

 

[myocardia]

Originally posted by: Idontcare
I have yet to find a single real-world application that responds to bandwidth metrics as Sandra does. Graysky and I even tried, twice, looking for the imrpovements that Sandra was showing us we should be seeing.

I mean is it just me or has anyone else found Sandra memory bandwidth tests entirely pointless in every aspect of the terms?

DC projects, especially F@Home, benefit from increased bandwidth. It/they are about the only thing I've ever seen/heard of that benefits, except for SuperPi, which is only another benchmark.

 

[Cookie Monster]

Im certain all core 2 quads communicate via FSB. This could potentially cause stuttering in cases where the FSB becomes the bottleneck, mostly in situations like heavy multi threaded environment where the cores needs to communicate with other cores, a good eg is a sync lock. Its almost (but not quite) as if having two dual cores in a dual socket system, that share the system memory.

Um, unless one knows how sandra measures or quantifies "multi core efficency", i dont think its such a big deal to be concerned with those results.

 

[thilanliyan]

Originally posted by: Cookie Monster
Um, unless one knows how sandra measures or quantifies "multi core efficency", i dont think its such a big deal to be concerned with those results.

I know...I don't care too much about those synthetics...heck I don't even care about having the fastest system (eg. I bought a Phenom ) but I figured the direct core communication of the Phenom would show up in a case like that which is supposed to show the benefits of direct communication (at least you'd think it should).

 

[Cookie Monster]

Well, there are other ways to see the "native" advantage of a phenom.

Try this for instance. You will need a friend with a core 2 quad system. You will need to download the "wallchanger" software, which allows you to automatically change wall paper at given intervals set by the user. Now load up say all your favorite wallpapers and have them changed at random times. Now fire up a 3d intensive app, like crysis. Play the game. Now see if there are any stutterings experienced while playing. Ive seen the results myself out of curiosity. Core 2 quad system stutters each time the wallpaper is changed where as on a phenom system, you wont notice anything.

 

[Phynaz]

Wow, that has got to be one of the most scientific tests I have ever heard about.