More P5B secrets uncovered

[sierra_bound]

Ah okay, I see what you're saying.

Maybe it's ClockGen and SetFSB that are screwed up. I generally try to avoid using either. Booting at around 390-395 with 2:3 ratio has given me the best results.

The other strange thing, which you may have noticed, is that SuperPi times get slower, the more times you run it. After each 32M run, I usually reboot. I found that I get better times that way.

[Revv23]

Ah okay, I see what you're saying.

Maybe it's ClockGen and SetFSB that are screwed up. I generally try to avoid using either. Booting at around 390-395 with 2:3 ratio has given me the best results.

The other strange thing, which you may have noticed, is that SuperPi times get slower, the more times you run it. After each 32M run, I usually reboot. I found that I get better times that way.

But you need to clockgen to keep performance up for higher fsb's...

It sure would be nice if asus would just let us control it.

[sierra_bound]

But you need to clockgen to keep performance up for higher fsb's...

It sure would be nice if asus would just let us control it.

Well, that only works up to a point. It's very hard to use ClockGen and go from 400 to 460 and beyond. Also it's a pain in the butt to do that every time you boot into Windows. There's only one tweak I will do after I get into Windows. Otherwise I run the bench as soon as possible. When I'm done, I reboot.

Again, I'm just talking about benching. People are becoming a little more savvy. Awhile back, everyone seemed obsessed with doing 500+ FSB. But many of them later realized that it wasn't necessarily faster than running 400+ with a memory ratio.

[Concorde Rules]

Ok, so what do I do now?

Setup as per sig.

Voltages are:

1.25V CPU
2.25V Mem
1.45V FSB Term
1.55V NB
1.6V SB

What should I try doing?

1. Just keep upping the FSB in clockgen until its unstable?
2. I also want to try lowering the chipset voltages, any suggestions with that?
3. I could probably get to 500mhz on the mem at 5-5-5-15 timings, 4:5, worth it? Any performance benefits over 4-4-4-12 420mhz?

Cheers guys I always feel I must have the tightest timings possible, Im used to AMD64 lol

Cheers

[sierra_bound]

I would lower FSB Termination Voltage. You don't really need 1.45v. Try 1.4 or even 1.3. Also try setting NB and SB voltages to Auto.

Regarding timings, my guess is 4:5 with 500MHz memory speed and 5-5-5 timings would be faster. But you'll have to run some tests to confirm that.

My feeling is that if you have RAM that does DDR2-1000 or higher, then you should use that speed. Otherwise you've wasted your money. Again, that's just my opinion.

[K404]

This guide is great. Posted at BleedinEdge as well.

This may seem really nooby, but does RAM multiplier have any effect on this? or is the 4:5 on the P5B DH just crap?

What i`ve managed:

440x7 (4:5) 550
440x8 (4:5) 550
445x8 (4:5) 556

445x7=NO, 450x8=NO The NBCC for these would have been 572 and 508 resp, though Tony said that is irrelevant now?

2:3 works better, though I have to choose RAM performance OR CPU performance here... my CPU and RAM options mean I cant get the best of both at the moment.

The other thing i`m finding is that 395->402 FSB is a no-go for me. I tried 390 and clockgen past 400...BSOD. If I throw more volts at it...will that solve it?

Kenny

[STEvil]

That's correct in general, yet not always. Perfect example is to compare 3-3-3 with 4-4-4 at FSB lower that 400MHz (e.g. 360 or 370) It's a perfect example, that P5B Dx is weird, to not say a bit screwed up

What I was trying to say, is that with NB strap change memory ratios change (1066MHz FSB strap @ DDR2-800 is 2:3 and at 1333MHz is 5:6). But from experimenting, boot @ 400 and Clockgen to 450 vs straight boot @ 450, would result in different memory speed, yet it doesn't seem so, judging by level of stability. That's why I'm not so sure if 401Mhz results in a 1333FSB strap or rather only NB timings get loose.

Anyone checked to see that the booting at different speeds is not also affecting CAS latency?

Say set at 3-3-3-x at 400, but boot at 401 and you get 4-3-3-x? Setting CAS4 is faster at some FSB's than others too, right? So maybe CAS latency also effects what strap you get at certain FSB's?

[sierra_bound]

This guide is great. Posted at BleedinEdge as well.

This may seem really nooby, but does RAM multiplier have any effect on this? or is the 4:5 on the P5B DH just crap?

What i`ve managed:

440x7 (4:5) 550
440x8 (4:5) 550
445x8 (4:5) 556

445x7=NO, 450x8=NO The NBCC for these would have been 572 and 508 resp, though Tony said that is irrelevant now?

2:3 works better, though I have to choose RAM performance OR CPU performance here... my CPU and RAM options mean I cant get the best of both at the moment.

The other thing i`m finding is that 395->402 FSB is a no-go for me. I tried 390 and clockgen past 400...BSOD. If I throw more volts at it...will that solve it?

Kenny

The 4:5 ratio has never been great on this board. It wasn't until bios 0609 that it became acceptable.

395-402 is a no-go with 2:3 or just no-go period? At 2:3, that would put your RAM speed above 590.

[$liM]

Hi

what fsb can you usually reach, when boot up from windows at 400mhz and than overclock with clockgen or setfsb! mine finished at 475mhz. At 480mhz win freeze.

sliM

[bachus_anonym]

Anyone checked to see that the booting at different speeds is not also affecting CAS latency?

Say set at 3-3-3-x at 400, but boot at 401 and you get 4-3-3-x?

Not that we can tell, no, but I don't think this is happening. Besides, performance-wise CAS3 vs CAS4 itself does not seem to make a big of a difference, except for a higher voltage requirement for CAS3. Loosening TRCD to 4 gives you a bit bigger penalty, though.

Setting CAS4 is faster at some FSB's than others too, right?

I've been playing a bit here and I don't think in case I described few posts up (3-3-3 vs 4-4-4 @ below 400MHz FSB) CAS3 is the problem but it appears that it's TRCD=3. Once you go 3-3-3, performance is worse than 4-4-4. But boot with TRCD=4 (or even change it in Windows, interestingly enough) and you're back on track. But above 400MHz, things get different again

BIOS on P5B Dx is really weird

[K404]

The 4:5 ratio has never been great on this board. It wasn't until bios 0609 that it became acceptable.

395-402 is a no-go with 2:3 or just no-go period? At 2:3, that would put your RAM speed above 590.

Hey Sierra. I just tried more voltage on the MCH and its not a problem now. 390x9 in BIOS, and clockgen to 410. (4:5) All good Sorry bout that.

K

[Concorde Rules]

Well I did some testing today.

399x7 and 400x7, 1:1, 4-4-4-12 gave the same results so I can say thats the 1066 strap.

At 401 there was a big change in B/W and CPU scores.

I tried the 4:5 divider at 400FSB, 5-5-5-15, benchmarks were better but simply could not get it stable

So back to 4-4-4-12 @ 400 and clockgening up as far as I can go with 4-4-4-12/5-4-4-13.

[Tony]

I go see Asus tomorrow, im going to plead with them to open up the bios for us, you need to know that they have to reverse engineer a lot of the settings so its a time consuming process.

MCH latency is the key here, relaxing it allows the board to clock real high but as you see there is a loss in performance...there will always have to be a balance so performance rises linear.

[Concorde Rules]

I go see Asus tomorrow, im going to plead with them to open up the bios for us, you need to know that they have to reverse engineer a lot of the settings so its a time consuming process.

MCH latency is the key here, relaxing it allows the board to clock real high but as you see there is a loss in performance...there will always have to be a balance so performance rises linear.

Why not ask them to allow us to set the strap?

Shoudn't that be easier to do, the BIOS does it automatically anyway atm?

As for my clocks, testing at 430x7, 1.25V CPU, 1.15V mem, 4-4-4-12 timings, 1.4V FSB Term

CR.

[$liM]

Hi

what fsb can you usually reach, when boot up from windows at 400mhz and than overclock with clockgen or setfsb! mine finished at 475mhz. At 480mhz win freeze.

sliM

can anyone test this please because i dont reach 500fsb when i boot from bios. I would like to know wether maybe my E6300 couldnt handle the high fsb or my board steps out. But i guess its my cpu

[Pt1t]

can anyone test this please because i dont reach 500fsb when i boot from bios. I would like to know wether maybe my E6300 couldnt handle the high fsb or my board steps out. But i guess its my cpu

try it with 6X coef.

[$liM]

try it with 6X coef.

already done, no change

[Fr3ak]

Why not ask them to allow us to set the strap?

Shoudn't that be easier to do, the BIOS does it automatically anyway atm?

As for my clocks, testing at 430x7, 1.25V CPU, 1.15V mem, 4-4-4-12 timings, 1.4V FSB Term

CR.

Its not just about the strap. There are a lot of internal timings, we have no clue about yet. Messing with them around could yield in a minor oc, but superior performance.

[K404]

Well I did some testing today.

399x7 and 400x7, 1:1, 4-4-4-12 gave the same results so I can say thats the 1066 strap.

At 401 there was a big change in B/W and CPU scores.

I tried the 4:5 divider at 400FSB, 5-5-5-15, benchmarks were better but simply could not get it stable

So back to 4-4-4-12 @ 400 and clockgening up as far as I can go with 4-4-4-12/5-4-4-13.

If you`re clocking past 400 using Clockgen, with a 4:5 divider, thats more work for the chipset. Try upping the volts, if should work. 4:5 is fine up to 440-445 for me.



@Tony... look forward to reading about the outcomes from the meeting

[FELIX]

Finally I found 4 registers wich change invariably in same values
between "under fsb400" and "over fsb400" on 965 chipsets:

(in 32bits values)
under 400:
offset 030->000F5A06
offset 034->01002810
offset 03C->23034417
offset 1F0->00008000

over 400:
offset 030->00025A00
offset 034->36802800
offset 03C->842B4407
offset 1F0->00009000

You can check this with it (final version):
MchbarEdit.zip

[ssabripo]

I've been reading this thread trying to figure out where the issue is with my setup...I cannot get anything Orthos stable past 467 now, unless I got 1.375 in Vcore and that takes me stable up to 3.304ghz...from there up, nothing is orthos stable!

I booted up at both 399 and 401fsb, then clockgen'd to 480+ fsb, and still nothing. SuperPi seemed ok up to 3.4, but anything above 466fsb in orthos is not stable.

the funny thing is, at 466 i'm 24/7 stable....perfect. But at 467 and above, Orthos starts to give me problems....I believe it has something to do with when the straps change:



What else can I try? should I drop to 4:5 and 400 fsb and see? But that would deteriorate my performance in terms of CPU speed and also 3dMarks, no?

any help/guidance is appreciated.

[Ace-a-Rue]

i can run AQ3 and 3D2001 when the machine is not orthos stable...

you can't argue with stable platforms even though Orthos says the opposite.

[Tony]

Jeez I should neaten that graph up a little ...LOL

I thought I would link you to some of my ramblings, have a look at my blog post here we really are up against it here untill we get bios files that allow us to tweak and the the boards do what we want and not what the bios engineer wants.

[El Snorro]

Jeez I should neaten that graph up a little ...LOL

I thought I would link you to some of my ramblings, have a look at my blog post here we really are up against it here untill we get bios files that allow us to tweak and the the boards do what we want and not what the bios engineer wants.

impressive post tony...
But I dont think intel will release this info, and it's just waiting for the Rd600

[ssabripo]

Jeez I should neaten that graph up a little ...LOL

I thought I would link you to some of my ramblings, have a look at my blog post here we really are up against it here untill we get bios files that allow us to tweak and the the boards do what we want and not what the bios engineer wants.

thanks Tony...good post.

unfortunately, it doesn't help me at this juncture...I'm stuck at 466 and would really like to reach 480 stable, even if just for the thoery of it all, to learn where the bottleneck is