Lowering divider to match fsb 1:1

[BTRY B 529th FA BN]

Hi, I was just thinking. I have DDR3 1333 and a Q9300 FSB 1333. If I take my memory and lower it to 667 to match my fsb 1:1, will i be able to overclock my fsb from 333 to 400 without opening the timings? As far as I've gotten without opening my timings @ 333 strap, fsb 333, & memory at 1333, is 360 @ 1440 2.7Ghz before i get memory errors (BSOD) Again the question, will i be able to open up my fsb to atleast 400 without opening up my timings?

DDR3 1333 6.6.6.18
Q9300 2.5Ghz
Asus P5K3 Del.

 

[n7]

DDR3 isn't meant to be run 1:1.

Let me repeat that for all the misguided 1:1 fanbois.

DDR3 isn't meant to be run 1:1.

The whole 1:1 ratio crap has gotten old, & thankfully, is soon to be completely outdated.

Yes, the benefits of running faster than 1:1 are generally small with DDR2, but Intel knows very well that extra bandwidth helps a little...& every little improvement is a good thing

If you are running your CPU @ defaults & your DDR3-1333 RAM @ defaults, you are running 1:2!

Your FSB is 333 MHz quad pumped.
Your RAM is 667 MHz double data rate.

I'd stick with that ratio if you can, though if you want to drop down one divider to maintain 6-6-6, that shouldn't be too bad.

Going to 1:1 w/ DDR3 would be silliness. (Though you're welcome to try it & see if you can somehow match performance of running say 3:5...)

You don't buy DDR3 to run it at half speed.

You'll have to bench to see which ends up being faster between 1:2 looser timings vs. the next lower divider (i believe it'd be 3:5) with tighter timings...

 

[BTRY B 529th FA BN]

I really don't know what i'm doing, hehe

 

[n7]

Eh, you're on the right track.

Take a look at your RAM dividers.

Since i prefer to take RAM outta the equation when finding the max clocks for a CPU, it wouldn't actually be a bad idea to run 1:1 while doing that, as you'll know the RAM isn't causing any limitations.

But once you find out how far you can get the CPU, then you'll want to bring the RAM back up in speed.

 

[BTRY B 529th FA BN]

Ok. Cause I did try it at 2.9 with a fsb of 387 and it gave me the blue screen. I am actually confused as to what the bsod actually is. I've seen it but it says something about memory erros so i assumed it was the memory. And the one time, I loosened up the timings (AMD machine not the Intel) it booted ok with no Prime95 errors. So I assumed it was the memory limiting me.


So we are talking about the same thing. You are saying take the memory divider and set it to 1066 and start ramping up the cpu speed?

 

[n7]

Well, you need to isolate things when OCing.

Find maxes for things separately.

If you don't have a clue what your CPU max is, or mobo max, or RAM max, & you are OCing hard, it's going to be very hard to determine what's giving your issues.

And btw, unless you have a crappy CPU cooler, your mobo is going to end up being what holds back your overclock, which makes things very interesting.

Anyway, if i understand things right for your mobo, 1066 divider is likely 5:8, which would work.

That'd give you lots of leeway to work on a higher FSB.

Keep in mind you'll need to up vcore, but likely also NB & VTT to get to around 7.5x450, which your mobo can hopefully do.

 

[BTRY B 529th FA BN]

Ok, I think I am seeing its the MB that is holding me back due to its stated FSB of 1333, am i right?; I am using a Zalman97 btw

http://www.newegg.com/Product/...?Item=N82E16813131181= mobo

 

[n7]

No.

Your mobo "supports" 1333 MHz FSB CPUs, amongst others.

Supports & what it can overclock to are very different things.

I was just mentioning the mobo will hold back the OC most likely, since very few current mobos can stably hit over 450 MHz with a quad, that's all.

That's the drawback with the Q9300...it's crippled by the 7.5x multi it has.

But that said, even at 3 GHz, that's a beast of a CPU...you don't need huge overclocks for it to be fast...it's fast at stock

 

[BTRY B 529th FA BN]

you wanna trade quad machines?

hehe

 

[n7]

Nope.

I prefer my "old tech" machine using old 65 nm tech quad, w/ old DDR2, & old P35 action

I'll be sticking with this for a while.

 

[BTRY B 529th FA BN]

Well, I dropped the mem divider to my next lower choice of 1066, raised the fsb to 400 which gives me a Rated fsb 1600 (what does this mean?) Dram freq is @ 640 6.6.6.18 1t and so far prime95 is stable running with 3Ghz

In reference to my fist post, what is holding me back if I run it at memory divider of 1333 - the mobo or memory? I am thinking its the memory and if it is and I upgraded to 1600 memory then used the 1333 divider setting would I be able to ramp up the memory speed/bandwidth without loosening the timings? I'm sure the mobo would default the memory speed to 1333 and still function normally.

 

[n7]

Based on what you've mentioned, the RAM was holding you back.

But you can loosen timings up.

There's nothing wrong with 7-7-7 or 8-8-8, at least certainly not while trying to find the CPU/mobo's max.

 

[BTRY B 529th FA BN]

Great! Thanks for the help

I've found running over 2.7Ghz will put 2 more degree's on the cpu's temp so i just might run it at 2.7 till i maybe get a liquid cooling setup.

Still after about 5 days of the new machine, I'm still feeling the 'awe' effects from it. Now if I can sustain my current satisfaction with the memory I have and not upgrade to a bigger hole in the pocket, I'll do just fine. Thanks again n7!

 

[j0j081]

sorry to interupt the thread but just out of curiosity then what would generally perform better?

an e6750 at 3.2 ghz 1:1 w/ram 800

or e6750 at 3.6 ghz w/ocd ram

Right now I run my cpu and ram matched because I thought increasing the cpu speed a bit but throwing off the divider wouldn't really gain anything performance wise. Is this not true?

 

[BTRY B 529th FA BN]

if you can get your memory to stay at 1:1 with your fsb @ 900 you'll do ok, but as far as changing or lowering the divider as I have done with my AMD machine (not sure it this holds true on the intel side) i didn't see any improvements. i didn't bench but it didn't feel any faster or i didn't see more fps

that brings up a question i have... when did cpu's start quad pumping the fsb?

 

[sutahz]

Originally posted by: n7
DDR3 isn't meant to be run 1:1.

Let me repeat that for all the misguided 1:1 fanbois.

DDR3 isn't meant to be run 1:1.

The whole 1:1 ratio crap has gotten old, & thankfully, is soon to be completely outdated.

Yes, the benefits of running faster than 1:1 are generally small with DDR2, but Intel knows very well that extra bandwidth helps a little...& every little improvement is a good thing

I'm waiting for the real-world benches...
So if it's soon to be outdated, what about the here and now? He's got a motherboard and ddr3 from here and now, is 1:1 bunk or not?

 

[n7]

If those who have DDR3 want to heavily underclock their RAM, i'm not stopping them :laugh:

But how people haven't figured out that 1:1 isn't all that i don't know.

Notice how Intel's 266 MHz FSB CPUs are designed to work on mobos' that support DDR2-667/800 on Intel's own generic boards?
That's not for OCing (can't do that on those boards)...that's for faster than 1:1 RAM, as in 4:5 or 2:3.

DDR3 is even more bandwidth oriented.

I wouldn't say everyone needs to be running 1:2, but that's the ideal ratio for DDR3.
333 MHz FSB + 667 MHz (DDR3-1333) RAM.

Running 333 MHz FSB with DDR3 underclocked to 333 MHz (DDR3-667) is certainly doable, but it's also foolishness, & would completely defeat the point of DDR3.

Intel designing for bandwidth is nothing new...this has been the case for years.

I'm not trying to say the benefits are huge; they are not.

But albeit small, they exist, & the whole z0mg 1:1 is teh bestest argument is generally made by those who don't understand that bandwidth is king on Intel, & has been for a long time.

The main exception to this rule is nForce chipsets, which offer 1T.
Due to that, you will indeed see better performance at lower speeds, or 1:1, with tighter timings.

Realize that Intel doesn't design for nForce chipsets though; they design for their own, & their own offer strongest performance when bandwidth is a-plenty.