Bus speeds

[GiGaBiTe]

I was having an argument on IRC about the bus speeds on a Core 2 Duo vs. a Core 2 Quad and what the performance impact was per core on having the FSB split between 2 cores vs. 4 cores.

The main processors of interest were the E6600 (2.4 GHz x 2 @ 1066 MHz FSB) and the Q6600 (2.4 GHz x 4 @ 1066 MHz FSB). The conversation lead nowhere on IRC as most of the people there were "QUAD BEATS DUO, GTFO" so I was wanting the opinions of people that were not unintelligible.

Basically my reasoning was that if the FSB was 1066 MHz on a C2Q that the bus would be split four ways to 266 MHz and some change vs. the C2D being split two ways to 533 MHz. So I was thinking that each core would have somewhat of a performance hit from not being able to communicate with the rest of the system as fast as the C2D would.

I had an idea for a benchmark of sorts to test this theory out. Basically get a system with a C2D 6600 and also have a Q6600 on standby so that one set of tests can be run after the CPUs are swapped. Basically to test if the FSB would make a difference, run the tests on the C2D first, then swap to the Q6600; but when you have the Q6600 in the system, tell the benchmark tools to not use two of the cores (so that the bus speeds remain the same) which I think can be done with affinity in the windows task manager.

I'm really not looking at which processor is faster, I just want to determine what the effects of a bus split 4 ways instead of two has on performance.

 

[systemlord]

You have way to much time on your hands there bud. Have a look at Tom's VGA charts as they have both Core 2 and Core 2 quad, this should give you the info you need. Most people OC their quad core's anyways which should make them as fast as Core 2 Duo.

 

[GiGaBiTe]

Please re-read my post, the point of the benchmark was to test the impact of bus speeds being split four ways instead of two, not to see how much better the C2Q is over the C2D. Also, it's called being investigative, not having way too much time; you should try it sometime.

Those VGA benchmarks wouldn't work in this test because of the obvious fact that two of the cores on the C2Q weren't prohibited from contributing to the test. For an even playing field, each processor needs to have only two cores running so that the other cores on the quad don't interfere with the results.

 

[zenmaster]

It's not like each Core get it's own dedicated 266mhz pipe.

Simply put, the Q6600 will match the E6600 on tests that cant use more than 4 cores. On tests than can, it will win.

Even for programs that can only use 2 cores, it will still likely win in real world cases since you will likely be doing other tasks while waiting for these long tasks to complete. Or you may simply have other tasks going on in the background.

 

[systemlord]

I understand what your saying but even if you do find out if the quad core using only two core versus the E6600 is actually slower or not, what will you do when you find that out?

 

[4745454b]

What you are talking about only matters IF the chip is doing work that will fill the bus. As was pointed out, its not like each core gets its own section of the FSB to use. If a core needs to use the FSB, to get a piece of data from ram for example, then it sends the request. There is no guarantee that the bus will be full or empty. If core0 sends a request for ram, core1 might be working on something, and not needing the FSB at all. Core2 would be receiving its previous ram request, which would use the FSB. Finally, core3 could be sitting there idle, as no task has been assigned to it.

Only if you are doing something that requires all four cores to use the FSB will this become a problem. Even then there might be some issues. If the work is simple and requires no real CPU work, then the bus will stay filled for a long time. But if the work is more complex, requiring many CPU cycles to complete, then some cores will probably finish before others, and will find the FSB empty. This is the strength of the C2Q, as its able to request more data to work on faster then the C2D, as it has more available cores to perform work.