Overclock.net banner

Intel Overclocking Redefined - Interesting P965 / 975X Secrets

2 reading
61K views 119 replies 45 participants last post by  Rebound  
#1 ·
Alright, some high level guys over at XS have discovered something interesting with the P965 / 975X chipsets. You may want to read through it all here, or just take my summary:

Your north bridge has an internal clock speed and latencies just like your CPU and memory. The FSB of your north bridge can be found by dividing your original CPU multiplier by your set CPU multiplier and then multiplying by your FSB.

So if you are running a E6600 (266 * 9) at 400Mhz x 8 your NB FSB is:

(9 / 8) x 400 = 450Mhz FSB (1800Mhz Total)

Just like your memory may be able to run at 4-4-4-12 at 1000Mhz but needs to run at 5-5-5-15 at 1200Mhz, your north bridge has a series of latencies which it must adjust in order to maintain stability at its FSB. These latencies seem to play a far more significant role in system performance than memory latencies.

Intel has predefined specific latencies at specific NB FSB speeds. They are referred to as straps. There is a strap for when the NB FSB is 1066Mhz and under, 1333Mhz FSB and under, 1600Mhz FSB and under, ect. When you go from the 1066Mhz FSB strap to the 1333Mhz FSB strap, the north bridge's internal latencies loosen to allow for greater stability.

ASUS has redefined the NB strap so that the 1333Mhz FSB strap does not come into effect until 401Mhz FSB (1604Mhz). Other perimeters of straps are somewhat unknown.

Tony, the guy who pretty much figured all this out has since been hired by OCZ (they were impressed). He is now a lobbyist trying to get ASUS and other major motherboard manufacturers to give the end user the control of when straps start.

There are 2 ways to beat the NB strap:

1. Boot to windows in the 1066Mhz strap and then use Clockgen to increase your CPU's FSB. You can then get to a much higher FSB while maintaining the 1066Mhz strap simply because the BIOS does not adjust the north bridge's latencies in real time.
2. Get a X6800 or QX6700 (or even a ES chip). To the north bridge, you are always at a default multiplier with a Extreme Edition processor. This allows you to set a much lower or higher multiplier without the NB FSB being effected.

Look at these Super Pi 32M results:

Quote:


Boot @ 400x9 Boot @ 402x9 Boot @ 400 & CLG to 402

DDR2-800 --> 14:10.969 DDR2-804 --> 14:42.922 DDR2-804 --> 14:06.672
DDR2-1000 --> 13:35.765 DDR2-1004 --> 14:07.297 DDR2-1004 --> 13:32.562
DDR2-1200 --> 13:17.109 DDR2-1206 --> 13:51.891 DDR2-1206 --> 13:13.109



Quote:
516x7=3611MHz, DDR2-1032 --> 13:55.625 400x9=3600MHz, DDR2-1000 --> 13:35.765

Quote:
ok, I have came to a conclusion of my own... when you use clockgen or another "windows" program to change the clock, apparently you cheat the board in a way that it doesn't apply the next strap, so you keep the best of both worlds, a faster strap and an fsb/clock increase... that has lead me to another point: probably when rising the FSB with clockgen you will reach a crash/limit must earlier than by doing that using the bios... so, let's say, if you can reach 500mhz by setting it in the bios, you will only reach 430mhz when using clockgen, that's because of the faster strap used when rising from windows (clockgen)... someone with hardware could test that !! I don't have my C2D rig yet...

Quote:


Boot @ 410x8 Boot @ 400 & CLG to 410x8

DDR2-1026 --> 15:22.390 DDR2-1026 --> 14:39.859

Image


Image

I will be adding to this as I learn more.
 
#3 ·
did you read this

i read up on this a while back, its very interesting stuff, just through this in with what you have and make a faq or something like that for this site

edit: for this same reason is why people hit apparent fsb walls at 395 or so and have to jump over 400 to like 402 or so with certain mobos
 
#4 ·
so 450 x 8 (which I currently run at) is no faster than 400 x9, perhaps slower?

And to notice an improvment 460 x 8 would be my best option?
 
#5 ·
What seems to be the best at the moment is stock multiplier and a 400Mhz FSB or greater than 460Mhz FSB (at stock multi).

You see, you can find the speed of the NB by taking your default multi and dividing it by whatever you have it set by and then multiply by your FSB.

So - 9/8 * 450 = 506.25Mhz (2025Mhz FSB for the NB)

It seems like the FSB strap will come into effect at a lower FSB when using a lower than default multiplier.

The 1333Mhz strap is suppose to start at 333Mhz FSB. However, ASUS manipulated the strap to allow to go to 400Mhz FSB before the 1333Mhz strap latencies kick in.
 
#7 ·
thread

yeah if you read to the bottom of that thread it shows ( i think) that a higher fsb say 500 or so coupled with a decrease of 1 from the stock multi gives you the best results, this can give yousome really good stuff with the 6300/6400

i would test more but my mobo has locked multi's
 
#8 ·
Rep +1 Nice. Very Nice. I did notice a lack of superpi progress after about 3.6Ghz after looking at the 1M superpi bench thread.

I attributed it to the same principle that drag racers succumb to. A 500HP car runs in the 11ish range in the 1/4 mile but a 1000hp car is only in the 9ish range. Twice the power, only about 1/4 the speed increase.

Now I know it might just be the latencies of the strap.
 
#9 ·
Quote:


Originally Posted by lightsource
View Post

Other than super pi, does it have an effect on any realword situations?

Yes, this results in significant performance and stability changes in your system.

Quote:


Originally Posted by Ihatethedukes

Rep +1 Nice. Very Nice. I did notice a lack of superpi progress after about 3.6Ghz after looking at the 1M superpi bench thread.

I attributed it to the same principle that drag racers succumb to. A 500HP car runs in the 11ish range in the 1/4 mile but a 1000hp car is only in the 9ish range. Twice the power, only about 1/4 the speed increase.

Now I know it might just be the latencies of the strap.

That is a good analogy, however it has nothing to do with this. You see, what you are referring to is the law of diminishing returns. The more you put into something the less you get out of it.

What effect that is happening in the P965 / 975X is the north bridge is increasing its latencies after specific FSB levels in order to remain stable. At stock settings (Intel defined) the north bridge will increase latencies at 1066Mhz, 1333Mhz, and 1600Mhz FSB levels within the northbridge. However, ASUS has reconfigured its motherboards so that you can go to the 400Mhz FSB level while maintaining the 1066Mhz FSB latencies within the north bridge. However, once you reach 401Mhz FSB, the 1333Mhz FSB strap kicks in and you get significant north bridge latencies.
 
#14 ·
This newly discovered info also shows why 410Mhz FSB is more stable than 390Mhz for a lot of people.

@ Joeking78. I am going to wait for my P5B (gets here tomorrow) before I test thing. However, there are plenty of people who have run this on their system with the exact same results.

From my interpretation, 400Mhz FSB with the default multiplier and as high of a memory speed as possible will be the best performance for people who can't go higher than 500Mhz FSB. For people with X6800 and QX6700 the multiplier part is null, because the NB detects any multiplier as default.

Edit: There is a big push to get the BIOS engineering to give the NB strap control over to the end user. That would allow up to define where the NB changes latencies.
 
#17 ·
Quote:


Originally Posted by Ihatethedukes
View Post

It also shows why the eVGA 680i has so many 'dead spots' in its FSB range.

EVGA 680i works entirely different than P965 / 975X

@ Lightsource: Yes, 400 x 7 is better than 437 x 7. Just do some Super Pi tests with it. Clockgen will be inherently less stable because you are working on the lower strap, but that is the best way to go to get 437Mhz (although it seems there is no chance at getting more than 10-15Mhz FSB from clockgen).
 
#18 ·
Quote:


Originally Posted by pauldovi
View Post

EVGA 680i works entirely different than P965 / 975X

@ Lightsource: Yes, 400 x 7 is better than 437 x 7. Just do some Super Pi tests with it. Clockgen will be inherently less stable because you are working on the lower strap, but that is the best way to go to get 437Mhz (although it seems there is no chance at getting more than 10-15Mhz FSB from clockgen).

All right, what about in F@H? Faster WU's?
 
#20 ·
Very interesting topic, good find.

I ahve one small kinda off topic question. I never realized before hand that dropping to a lower multiplier will OC the northbridge. This makes me nervous as I was going to go form a multiplier of 9 to 8, and run maybe 430 x 8 as this will get me ddr2-1075 which I can do at 4-4-4-12. Would this push my Northbridge too far, would I need better cooling? I know its off topic, but were talking about the northbrige anyways, so whay not.
 
#21 ·
Quote:


Originally Posted by Intervention
View Post

Very interesting topic, good find.

I ahve one small kinda off topic question. I never realized before hand that dropping to a lower multiplier will OC the northbridge. This makes me nervous as I was going to go form a multiplier of 9 to 8, and run maybe 430 x 8 as this will get me ddr2-1075 which I can do at 4-4-4-12. Would this push my Northbridge too far, would I need better cooling? I know its off topic, but were talking about the northbrige anyways, so whay not.

No, it is very much on topic.
Image


From my understandings of things, you would have a 484Mhz NB FSB (1935Mhz NB FSB Total) That would put you in the 1600Mhz NB strap, which would give you poor performance compared to 400Mhz (9 multi). You want to try to stick to a 400Mhz FSB, stock multi, and as high of memory speed as possible. Tony has been working hard to get ASUS to release a BIOS where we can manual set the NB strap. That way you could set the 1333Mhz strap to say 1936Mhz and be great with your setup. Although that would require some increased voltage in the north bridge to keep it stable.
 
#24 ·
Paul, is there any way you could explain what a NB Strap is? I have googled all over and can't find what it is/does.

*Sheepish nod*