Memory Frequency Multiplier

[b7rk]

Hello,

I am wondering how does changing memory frequency multipler effect my processing speed?
I am using GA-EP35C-DS3R + E8400 + Nirvana NV120 + Cruciall Ballistix PC6400 2x1GB.

It seems like using a lower memory frequency multiplier I can make my memory frequency to match a higher FSB, and I have a better overrclock. Does that mean I should always use the lowest frequency muliplier? (This seemed to have improved my SuperPi calcuation speed from 12.5 seconds to 11.5 seconds)

Thanks

 

[dagger]

Yes, 1:1 ratio is synced, and provides best performance. If ratio is higher, the ram would be running faster than cpu, yet the cpu can only access it on its own lower clock, thus, the extra is wasted. There's more going on than that, but that's the basic idea.

 

[iluvgillgill]

yes from the current situation by Intel 1:1 multiplier is the best to use.

 

[spuddyt]

strange colours.... pretty....

 

[Zorg]

IMO, 1:1 is overrated. Sure, if you can run the FSB to match the RAM at 1:1 then by all means run 1:1, otherwise it doesn't make much real world difference either way. I would say that you are better off with RAM at it's rated speed and not 1:1 unless the RAM will run with tighter timings to offset the difference.

The whole thing with 1:1 is that the MCH doesn't have to work as hard and thereby runs cooler and more stable at high OC's.

The core2 arch isn't heavily affected by the RAM bandwidth anyway, assuming it's not an Allendale.

 

[dagger]

IMO, 1:1 is overrated. Sure, if you can run the FSB to match the RAM at 1:1 then by all means run 1:1, otherwise it doesn't make much real world difference either way. I would say that you are better off with RAM at it's rated speed and not 1:1 unless the RAM will run with tighter timings to offset the difference.

The whole thing with 1:1 is that the MCH doesn't have to work as hard and thereby runs cooler and more stable at high OC's.

The core2 arch isn't heavily affected by the RAM bandwidth anyway, assuming it's not an Allendale.

You don't get any real performance increase with higher ratio. Higher clock rate ram cost more and tend to have looser timings.

 

[Zorg]

And your point?

 

[dagger]

And your point?

Why should people use higher ratio, which in turn force timings looser, when they can just keep it at the optimal 1:1 and get tighter timings? And if they buy more expensive new ram for the higher clock speed, it makes even less sense.

 

[Zorg]

Why should people use higher ratio, which in turn force timings looser, when they can just keep it at the optimal 1:1 and get tighter timings? And if they buy more expensive new ram for the higher clock speed, it makes even less sense.

Somebody is running around using the thumbs up/down buttons.

I agree if you don't already have the RAM and you know that you won't be OCing then there really is no reason to get the higher speed because the real world benefit, with the core 2, is not that much. RAM is pretty cheap though, and one never knows whether they will get the proverbial hair and decide to OC later, do they? Also, the timings of slower RAM are usually not as tight as faster RAM which is underclocked, I don't think that claim can be made across the board though, and also doesn't have much real world affect. As far as whether running DDR2 800 at 667 or 533 1:1 with tighter timings makes any difference I would have to say that it's also negligible.

So the question is, does it really matter if you are running 1333 FSB with RAM at 1:1 and tighter timings or 1333 FSB and RAM at 800 5:6 and looser timings. I've benched it and run it both ways and it is negligible. The benches show some difference, but I can't really "feel" it.

So really the "1:1", that everyone pushes as being the clear choice, doesn't make that much difference if any.

 

[iluvgillgill]

zorg i think you shouldnt recomment anyone set anything higher the 1:1. i can understand you as slower the ram it actually create a high latency by running slower speed with the default(high) timing. but even run at a slow latency the actual read/right figure doesnt change to a dramatic difference.

since most people are looking at a 400fsb OC so 1:1 will get them 800mhz anyway. so 1:1 is the right memory multiplier to go. no good to put unneccessary stress on the ram is there?

 

[Zorg]

zorg i think you shouldnt recomment anyone set anything higher the 1:1. i can understand you as slower the ram it actually create a high latency by running slower speed with the default(high) timing. but even run at a slow latency the actual read/right figure doesnt change to a dramatic difference.

since most people are looking at a 400fsb OC so 1:1 will get them 800mhz anyway. so 1:1 is the right memory multiplier to go. no good to put unneccessary stress on the ram is there?

I don't disagree with you and you obviously didn't read my posts.

 

[iluvgillgill]

I don't disagree with you and you obviously didn't read my posts.

well i dont see anything wrong of further stress the point to the OP.dont get offended thats what i say.

 

[Zorg]

I'm not offended. I will post the portions of my previous posts that you didn't catch, because you must have missed them or you wouldn't have made the post that you did.

 

[iluvgillgill]

i think we both got abit mix up with the two posts we are in

 

[Lupiron]

Haha, you guys are funny!

I would like to chime in that the only thing really mattering is running your RAM with the FSB. So 1:1, or in sync.

Why? because if you use 1066 RAM with a 400 FSB, you sure the hell are not gonna get 533 Mhz worth of data per channel through that 400 Mhz FSB.

So I would guess that 1st you try and tighten the timings as said above, but barring that, why not run your ram at a slower rated speed with less voltage just for GP?

As in, wouldn't your 1066 RAM @ 2.1 v. Be better off running at 800 Mhz @ 2.0 with even the bonus possibility of also getting tighter timings, especially if having it at 1066 and 2.2 volts does nothing over the lower setting.

Thats just my guess.

--Lupi

 

[iluvgillgill]

good morning lupi. so you got internet now or now?dont see you online anymore.

 

[Lupiron]

Been moving!

 

[iluvgillgill]

alright. you settle down now?lol

the OP seem lik sleeping now!lol

 

[Zorg]

i think we both got abit mix up with the two posts we are in

I wasn't mixed up. Go back and check the other thread, I figured out the VID thing.

 

[Lupiron]

Get him, Zorg!



--Lupi

 

[Zorg]

I would like to chime in that the only thing really mattering is running your RAM with the FSB. So 1:1, or in sync.

Why? because if you use 1066 RAM with a 400 FSB, you sure the hell are not gonna get 533 Mhz worth of data per channel through that 400 Mhz FSB.

So I would guess that 1st you try and tighten the timings as said above, but barring that, why not run your ram at a slower rated speed with less voltage just for GP?

As in, wouldn't your 1066 RAM @ 2.1 v. Be better off running at 800 Mhz @ 2.0 with even the bonus possibility of also getting tighter timings, especially if having it at 1066 and 2.2 volts does nothing over the lower setting.

Thats just my guess.

--Lupi

All I said is that the "1:1" timing doesn't make any major difference. Everyone has been posting like it's the Holy Grail. It doesn't really get you anything except the reduction of voltage. I say go either way It doesn't matter.

If you are going for the record then the reduced load on the MCH and consequent heat is a great idea, otherwise it don't matta.

Also if you want the RAM and the FSB to be truly in sync, then you would need QDR RAM not DDR.

 

[Zorg]

Get him, Zorg!



--Lupi

how about you? :lol:

 

[Lupiron]

I agree partly! As long as your FSB is at least equal to your memory speed is what matters. But it is a waste to have any ram and a 266 Mhz FSB.

That part has been proven again and a again. Take your 800 Mhz Ram and have a 266 FSB with a 2:3 divider and test away.

Then simply raise the FSB, leaving the speed of the RAM alone. It's obviously a chunk faster.

You can always test it for yourself. Take some 1066 and use a 266 FSB and sync it up and test speeds, then do it again at 400 synced. Then again at 400 FSB with that 2:3 divider for 1066. The speed only jumps up when the FSB is raised as well.

Other than that part, doesnt matter what you do, as long as you run the highest safe FSB with it. That's where the performance lies.

Interesting theory, but they prolly will, if that would matter. Since qdr 1600 is FSB 400. Same thing, yes?

But it would be different if all 4 channels was linked for Quad data rate, yes. But everyone would need 4 stix!

--Lupi

 

[nachowarrior]

AMD... integrated memory controller for efficiency, cost effectiveness, and to create less arguments among hardware enthusiasts.

that's their new slogan, the head PR rep over there read this thread... and the 90 other ones like it.

oh, and the person that posted hasn't even re-posted yet. But when he does come back, here's my suggestion. Use a scientific method to figure out what makes the programs you run the most faster/more efficient. something like fraps for games, and a stopwatch for other programs and loading times.

 

[Lupiron]

Yee Haw!!

Nacho!

--Lupi