Will the multiplier used for my OC make a difference ?

[sourthings]

Hi all,

I was wondering if there is any real world performance difference if I were to run my Q6600 at 8x450 or 9x400 for a 3.6 OC. I give this as an example, as I currently am running it at 8x430 for a 3.44 OC. If I were to match my OC with a 9 multiplier rather than a 8, does this give any real world difference in performance ? My memory is always running at a 1:1 ratio, so right now I have it at 860.

Thanks in advance.

 

[Idontcare]

Higher FSB will result in higher maximum theoretical bandwidth between the cores in your quad (more efficient multithreading performance) as well as higher bandwidth from your CPU to the RAM.

So yes, a lower multi combined with a higher FSB (8x450 vs 9x400) is more desired if the RAM latency at 450 (DDR2-900) at least matches the latency at 400 (DDR2-800). If it doesn't then you've lost some performance in terms of increased RAM access latency in exchange for an increased in theoretical bandwidth.

Increasing theoretical bandwidth (which will only come into use should you truly saturate the FSB with data handling, very rare on desktops) at the expense of increasing access latency (which will impact every single transaction the computer will ever make with the RAM) is not a good trade off from a fundamental perspective.

Loosening the timings while increasing the FSB is OK, provided the loosening (increases ram clock latency, but since ram is being clocked higher the latency itself can remain the same) is not so aggressive that the RAM latency actually ends up being higher.

In this constrained situation, higher FSB will result in a more efficient and faster memory subsystem for the CPU which will result in an overall faster system.

 

[sutahz]

any real world performance difference

No.

 

[The-Noid]

There will be a difference in real life albeit slow.

2-3% approximately would be a best case scenario.

 

[Markfw]

Well, at least speaking for something like F@H, increase can be major. Duvie is coming up with some real numbers, different FSB, 1:1 memory, same end speed. We already know its real and observable, just can't quantify until he posts.

 

[DSF]

Originally posted by: Markfw900
Well, at least speaking for something like F@H, increase can be major. Duvie is coming up with some real numbers, different FSB, 1:1 memory, same end speed. We already know its real and observable, just can't quantify until he posts.

I'll be interested to see that, as this is the kind of question most people answer without really having hard facts in front of them.

 

[myocardia]

Originally posted by: DSF
I'll be interested to see that, as this is the kind of question most people answer without really having hard facts in front of them.

Straight from the horse's mouth.

 

[Markfw]

Originally posted by: myocardia

Originally posted by: DSF
I'll be interested to see that, as this is the kind of question most people answer without really having hard facts in front of them.

Straight from the horse's mouth.

The test that I suggested to Duvie (and he is working on) is more relavent, at it compares a quad (not referenced in this article) at the same core speed, overclocked, but at different memory speed and multiplier. But those numbers suggest what we found, as much as 20% I think it what we found.

 

[Diogenes2]

Originally posted by: Markfw900

Originally posted by: myocardia

Originally posted by: DSF
I'll be interested to see that, as this is the kind of question most people answer without really having hard facts in front of them.

Straight from the horse's mouth.

The test that I suggested to Duvie (and he is working on) is more relavent, at it compares a quad (not referenced in this article) at the same core speed, overclocked, but at different memory speed and multiplier. But those numbers suggest what we found, as much as 20% I think it what we found.

Are your tests with the SMP console ?

Also, are you folding work units that require large amounts of RAM ?

 

[sourthings]

Thanks for all the replies. I think I will try hitting 8x450 then. It's odd, my OC at the same Ghz are always stable when I use the 8x multiplier but are flaky when I'm using the 9x multiplier. This on a p35c-ds3r mb.

 

[aussiestilgar]

You'll need to find the "sweet spot" with your RAM timings and your FSB frequencies. If you need to raise your timings to increase the FSB, this can be detrimental to performance. Either way you wouldn't notice any difference in real world apps.

 

[Markfw]

Either way you wouldn't notice any difference in real world apps.

Did you read my post above, and the one at F@H that proves my point ? Yes its DOES make a difference in some apps.

 

[aigomorla]

Originally posted by: Markfw900

Either way you wouldn't notice any difference in real world apps.

Did you read my post above, and the one at F@H that proves my point ? Yes its DOES make a difference in some apps.

If your talking about word and office productivity software, unless its a SQL server or some DB server, i highly doubt you'll see any difference between 3.2ghz @ 400fsb x8 vs 3.6ghz 400fsb x 9

Also gaming you probably wouldnt notice any big difference unless your videocard was absolute crap.

But as mark says, if you encode, or do lots of folding or distributing computing, you'll see differences.

But keeping the FSB same with different multi's usually yield close performance. The 3.6ghz of course would be faster, but unless you fall in DC computing, or the list i gave you above, i highly doubt you'll notice any difference.

 

[sutahz]

Originally posted by: Markfw900
Did you read my post above, and the one at F@H that proves my point ? Yes its DOES make a difference in some apps.

Your post above doesn't show us anything. It states that you suggested to Duvie to do some tests that you state show 20% increase in performance. That's 20% increase in one specific application (F@H). sourthings never said if he was part of F@H nor any other DC project. When someone asks "real world benefit" I think of office, games, video play back, small encoding tasks maybe.

 

[Markfw]

Originally posted by: sutahz

Originally posted by: Markfw900
Did you read my post above, and the one at F@H that proves my point ? Yes its DOES make a difference in some apps.

Your post above doesn't show us anything. It states that you suggested to Duvie to do some tests that you state show 20% increase in performance. That's 20% increase in one specific application (F@H). sourthings never said if he was part of F@H nor any other DC project. When someone asks "real world benefit" I think of office, games, video play back, small encoding tasks maybe.

I said he DID them, he just hasn't posted his results. And for the 894,441 people running F@H, it is a real world apps benefit. Just because you don't some use softwareware doesn;t mean its not a real world app. When I get my new PC-8000 on wed (coming back from RMA) I will run them myself. I just referenced Duvie, since he already did it.

 

[aigomorla]

LOL mark... dont get so worked up. Were from a different league of ocers as others are.

You and are DC OCers which is different from Gamer OCers, and also World Record OCers.

Trying to tell one class the benfits of such overclock to another is almost completely useless and only results in fights.

Trust me ive had my share in this area.


All they need to know is that if your Indexing/encoding/Folding/WCGing/SuperPIing then the added MHZ does play a big role. If not, they wont see squat of a difference.

 

[Markfw]

You can lead a horse to water, but you can;t make him drink...

If they won't believe me, they won't get the benefit, their loss. I will post the numbers later this week, 389 (9x), 438(8x) and 500(7x) fsb is my plan, all@3.5 ghz. F@H SMP client, 2 instances.

 

[sutahz]

"The test that I suggested to Duvie (and he is working on)" does not say "I said he DID them, he just hasn't posted his results."
F@H is a application, true. It exists in this world and people run it. But at the end of the day all you get is a higher electric bill, F@H giving you so many points and hopefully a warm fuzzy for doing your part.

Originally posted by: aigomorla
You and are DC OCers which is different from Gamer OCers, and also World Record OCers.
Trying to tell one class the benfits of such overclock to another is almost completely useless and only results in fights.

Good advice, it says what I've said but differently so hopefully Markfw900 will understand. I'm not saying that a higher FSB will NEVER net any gains, but for your avg. person, it won't net any real gains (and your avg person doesn't run F@H nor any DC software). So when someone asks about higher FSB and different multi's and if they will see any differences in the combinations, and they don't specify what programs they use....
"If they won't believe me, they won't get the benefit, their loss." What benefit does one get from running F@H? So not earning as many points in a day is their loss? What did they lose?
Your focusing on one progam.. and where in my posts did I say you were wrong about a higher FSB not helping F@H?

 

[Thor86]

Originally posted by: sourthings
Thanks for all the replies. I think I will try hitting 8x450 then. It's odd, my OC at the same Ghz are always stable when I use the 8x multiplier but are flaky when I'm using the 9x multiplier. This on a p35c-ds3r mb.

What are your nbcore voltages? You may need to raise this to at least 1.6 if not 1.7 for 450fsb on C2Q ocing.

Also, I could care less about F@H, but for real-world gaming performance, nothing beats 1:1 memory ratio Intel or AMD. I tried 4:5 and 5:6 ratios, and they cause my games to stutter at the same cpu clocks.

 

[lopri]

I would think this never-ending argument over 'real-world' benefit of FSB is kind of a tautology. It's just like 'real-world' benefit of CPU vs GPU in gaming. And we all know how it goes. If an application (and its configuration) is GPU-bound, it's GPU-bound. If it's CPU-bound, then it's CPU-bound.

If F@H is FSB bound, then it's FSB bound. And I have no reason not to believe it if that's the case. (see where we're getting at?) You can check it real quick with Everest Cachemem test. No matter what you do with your CPU/memory, the memory write speed (unless your memory runs slower than FSB) will always be the same per FSB.

Obviously there is a situation where FSB is a 'bottleneck'. But how 'real-world' will depend not only on the rest of CPU-memory configuration but also on subjective view on 'real-world'. (F@H can be as 'real' to some as Office 2007 is to others)