More P5B secrets uncovered

[EnJoY]

This is some of the best reading regarding fsb overclocking I've seen in awhile. I'm extremely excited to do some of my own testing on this and see where it gets me. This answers so many questions.

[RichUK]

Changing the multiplier in Windows also changes the NBCC on the fly. However, as far as I can tell EIST does not change the NBCC when temporarly throttling up and down on the CPU speed. Because the set multiplier really does change in the same sense as when it is manually selection via software/BIOS. That's about the best explaination I have so far. EIST is a completely different beast than manual multiplier selection and it shares about nothing in common with what we see in BIOS.

-FCG

So in theory if someone could some how manipulate EIST within the BIOS that is used, instead of EIST working on OS intervention. This could possibly provide some sort of solution to high overclocks?

Hopefully DFI + 965p = hella good board?

[JoeBar]

Very nice read bigtoe (aka Tony) m8...

[Tony]

What we need is EIST control as well as multi control, this way from bios you could see where the NBCC is and set the FSB accordingly to how you want the cpu to run.

I have a feeling once a few manufacturers see we have new needs from bios they will repsond with more tweaks, only time will tell.

On a side note will someone with a C2 rev 965 board run 600fsb and show us a sandra screenie of the motherboard spec...like Kris has shown, that way we can see if the new rev of the chipset is hitting some silly speeds. Also if they could test between 360 and 399fsb 1:1 for ram stability without masses of voltage to see if the new chipset is clocking and controlling the memory better we would have a better idea of what to expect from the newer boards.

[Dubz]

leave it up to Tony and FCG to bring some good reading to XS. Thanks alot guys.

[OnDborder]

same or very similar performance, but very different levels of stability depending on the default multiplier of the CPU.

say you have a E6400 that has a default multiplier of 8.

well, you take the default multiplier, divide it by the multiplier you set, and then multiply it by the FSB you set, and then you get the Northbridge chipset clock.

so 8/8 x 400 = for 400 mhz or 1600 mhz on the chipset clock.

but then just lower the multiplier to 7, and stability should change cause your stressing the chipset even more.

8/7 x 400 = 457 mhz or 1829 mhz on the chipset clock.

even though all you did was change the multiplier.

so the default multiplier is the leas stressful multiplier to use, but it's also the most limiting in most cases, cause it's easier to do 7x 500 then it is to do 8 x 500.

upon learning all this, I think now its way easier to spot problems when overclocking conroes, so thanks FCG

Thank you for simplifying it.

[MrDeeds]

I'm very interested in what this means for the x6800. Does this indicate that the x6800 underclocks the nbcc at higher multis. Could this also be the reason most of the 975x boards dont offer multi down on conroes other than the x6800.

[Revv23]

Thinking this over it seems that the e6400 would be an ideal chip, run at 7x multi and push NBCC towards 600 while shooting for 3.6ghz.


also, so what 965's are the new revision right now?

[aggybong]

From what I've seen, the only way to tell is to load up CPU-Z and check. The boards (at least the P5B-Ds) with the new C2 chipset are still revision 1.03G

[fscussel]

I have come to a question based on what you have discovered:

is it possible that E6400 will constantly do 4ghz by 500x8 whether E6300 fail to do 4ghz with 571x7, with a mobo that can do all the way up to 600mhz and a memo capable of 600mhz also? what I mean here is, does the E6400 overclock better (reach a higher final clock) than the E6300? since all the C2D seems to only be limited by the mobo and FSB that it can provide, they all should have the same TOP Limit if using the default multi.

[remorema]

i did a very simple test to verify if the graphics posted on bleedinedge were correct.

pretty soon i found that is not that simple.

the graphics showed that you need to run you chip @ 460+ to get the same memory bandwith as with 399 fsb.

so how do you explain this scores:

p5b vanilla running @ 395fsb


p5b vanilla running @ 405 fsb


I agree that they have to maipulate de NB strap in order to run at higher fsb's, but the difference is not that high.

as you could see i had better results @ 405 other than 395.

take your own conclusions isntead of just believing what you read.

[caater]

i'm still puzzled about that - how does the chipset know the default multiplier of a true unlocked chip (ES) if multiplier is changed normally in bios (not sure what is normal though )?
i'm aware that, at least asus p5b boards, are using EIST for selecting lower multipliers (just like asus dothan boards) - they boot up with def mult and then transparently change it to lower value.. effect is similar to crystalcpuid mult change etc.
but then again - a bios with "normal" (not eist) mult change and ES (or just unlocked chip), board still boots up with def mult and then bios changes it, right?
however, my experience with e6400 and e6600 (asus p5w - 975) were:
e6600 limit 8x443=nbclk=498
e6400 limit 6x473=nbclk=630

edit: disregard that eist stuff.. browsed some intel dev manuals and mult lowering is done using MSR..

btw, would be idea to try to up fsb with active eist.. if eist multiplier don't up nb speed.. ?

[kemist]

I thought he used Memtest. From my experience the bandwidth it displays is much more sensitive to memory settings than windows based stuff. Anyway im sure Big Toe knows what hes talking about, hes very respected in the community. I wouldnt go throwing around inflammatory comments such as "take your own conclusions isntead of just believing what you read"

[Tony]

i did a very simple test to verify if the graphics posted on bleedinedge were correct.

pretty soon i found that is not that simple.

the graphics showed that you need to run you chip @ 460+ to get the same memory bandwith as with 399 fsb.

so how do you explain this scores:

p5b vanilla running @ 395fsb


p5b vanilla running @ 405 fsb


I agree that they have to maipulate de NB strap in order to run at higher fsb's, but the difference is not that high.

as you could see i had better results @ 405 other than 395.

take your own conclusions isntead of just believing what you read.

I ran tests outside of windows, if you are going to counter what i published please test the same way I did.
Run in 10fsb steps from 270 up, 1:1 4-4-4-10 2.4V and show us the bandwidth results in memtest 86+

I ran from 270 to 510 with a 6400 using the 7 multi.

Of course the vanilla P5b may not act the same as the deluxe either...you have to consider this also

[OnDborder]

I thought the trick is to use a lower multiplier than default. Right??

This ought to be a "sticky".

Maybe I was lucky when running the fsb up. I didn't have any problems in between 330-400.. Maybe extra cooling helps?

[TridenT]

very nice! ill try for sure to see what i get with my 6600.

[aggybong]

I ran tests outside of windows, if you are going to counter what i published please test the same way I did.
Run in 10fsb steps from 270 up, 1:1 4-4-4-10 2.4V and show us the bandwidth results in memtest 86+

I ran from 270 to 510 with a 6400 using the 7 multi.

Of course the vanilla P5b may not act the same as the deluxe either...you have to consider this also

But shouldn't the results from Memtest86 be mimicked when running memory tests in Windows? If not, then none of this means much when it comes to memory bandwidth. For what it's worth, I did a tiny bit of testing myself with the 0614 BIOS and Memtest86:

8x399 = 4986
8x400 = 5000
8x401 = 4481

[caater]

about these sandra screenies.. is it possible that sandra is just buggy and cant read correct cpu speed? if sandra shows 600 fsb, is cpu speed normal or not?
if i change mult with crystalcpuid, sandra will always show bootup cpu speed and when lowering mult, sandra "adjusts" its fsb speed accordingly because it detects new multiplier, but can't read correct fsb and clockspeed.

[Tony]

But shouldn't the results from Memtest86 be mimicked when running memory tests in Windows? If not, then none of this means much when it comes to memory bandwidth. For what it's worth, I did a tiny bit of testing myself with the 0614 BIOS and Memtest86:

8x399 = 4986
8x400 = 5000
8x401 = 4481

it all depends if Everest actually measures anything, or calculates bandwidth by bus speeds etc.

memtest measures it as far as i know

[aggybong]

So would Sandra be better for that?

[remorema]

I ran tests outside of windows, if you are going to counter what i published please test the same way I did.
Run in 10fsb steps from 270 up, 1:1 4-4-4-10 2.4V and show us the bandwidth results in memtest 86+

I ran from 270 to 510 with a 6400 using the 7 multi.

Of course the vanilla P5b may not act the same as the deluxe either...you have to consider this also

how many of us work 24/7 out of windows ??

i guess that tests should be done where we are actually going to seek for maximum performance.

but since any of you doesn't trust everest, i'll post exacly the same settings as i posted with everest, and then we'll see if something changes.

p5b 395fsb


p5b 405


As you can see, different programs equals to different results.

so i keep saying, test your own hardware and take your own conclusions.

i doubt that you have to reach 460+ to equal mem bandwith that you achieve with 399/400 whatever.

[Super Nade]

Superb information guys. Thank you for all the hard work!

[gundamit]

Also if they could test between 360 and 399fsb 1:1 for ram stability without masses of voltage ...

Am I reading too much into that quote by thinking the only way through the 360 to 399 fsb "no mans land" is "masses of voltage"? Not an ideal solution, but until something else shows up its all we have. I've put more than the prescribed volts into my sticks at those speed to push back the fsb barriers at 1:1. My DFI 975 grinds to a halt in the 380s. The P5B-D I have does't really like less than 402 much. Thanks to all who are trying to get this all sorted.

[aggybong]

These are with a P5B-Deluxe running the 0614 bios. ALL MEMORY TIMINGS ARE UNCHANGED (4-4-4-12-4-35). I rebooted and set all values in the BIOS, no clockgen or any other Windows program were used to manipulate settings.

Memtest86 Results:

8x399 = 4986
7x399 = 4986

8x400 = 5000
7x400 = 5000

8x401 = 4481*
7x401 = 4324*

*These two don't match up like the 400 and 399 results do.

These two have the same CPU speeds and memory timings, but different FSBs and multipliers:
8x400 = 5000
7x458 = 4938

SiSoftware Sandra 2007 Results. The pictures also include MBench\CPU-Z numbers.

8x399: 7231 / 7242
8x400: 7261 / 7303
8x401: 6788 / 6802
8x425: 7142 / 7115

7x400: 7167 / 7168
7x458: 7605 / 7575
8x400: 7261 / 7303 (Posted again for comparison to 7x458)

Notice how in Memtest, 7x458 and 8x400 have nearly the same number, but with Sandra and MBench, things are much different.

[Tony]

Great post, I never thought of using Mbench for the access latency but I will now

See if anything happens from 467 up also, thats where the next strap in theory should kick in.