CPU 700MHz Bus Speed Project

[wytiwx]

Hey, guys

I want to achieve my 700 FSB OC goal, now I get some questions, I wish the 700 FSB masters can help me.

1. I am going to use C2D E8400 or E8500 or E8600, how much FSB @ Vtt should CPU can do on air? for example, 650 FSB boot @ 1.20V Vtt?  or 670 FSB boot @ 1.30V Vtt?

2. What motherboard will be good enough for 700 FSB? I saw many OCer used EP45T-EXTREME, but it's really hard to find by now. DFI P45 is great board but it can drive you mad too.

3. How much MAX FSB @ Vnb should motherboard can do on air? for example, 600 FSB boot @ 1.20V Vnb? or 630 FSB boot @ 1.30V Vnb?

4. How to tweak Vtt, Vnb, Vpll, CPU Vref, NB Vref on motherboard? What is the main principle?

THX very much!

[TerraRaptor]

1. Hi. Normally, VTT won't give you any noticable gain - one cpu will do FSB 650 in 32m with 1.44-1.49v vtt, another one will hit the wall at 630 with same voltage and raising vtt to 1.7v won't break that wall. Also, some cpus have a very narrow range between full stability (pi 32m run) and valid - just 10-15mhz, others may have that range extended - like 640mhz 32m and 690mhz barely stable to validate (but yet 50 mhz difference). I would say C1 stepping is first category, E0 is second category. CPUs are also extremely sensitive to GTLs. So I would bin cpus for fsb validation if that is the goal. I had just a couple of good cpus that would break 690mhz, both were trying to post with air cooling @670mhz. Very important thing to consider - push Vcore for fsb. I needed >1.7v to get 700mhz although core clock was 4.2ghz.

2. Biostar t-power p45, giga ep45t-extreme, rampage exteme (8x serials). There is a chance with P5e3 x48, p5e64 evo. DFI is really annoying, ate all my patience yesterday - so I won't recommend it (though board is still worth mentioning here).

3. Boards run into a kind of wall with fsb, I wouldn't count Vnb as important factor - while it scales, who cares of volts? My safe limits are 1.82v and 45C idle temps.

4. VTT - normally every cpu has a sweetspot, 1.45-1.55v. Vref (aka GTL) should be 0.61-0.65 of VTT, depends on cpu. Vpll - keep it low, high values are killers for 45nm anyway. I have an impression that Vcore should be higher than vtt with a margin of 0.15-0.2v all the time. NB vref is a kind of unknown for me yet, but looks like p45 needs it to be x0.58 of vtt, x48 will need higher values (0.63-0.65).

[wytiwx]

3 hours ago, TerraRaptor said:

1. Hi. Normally, VTT won't give you any noticable gain - one cpu will do FSB 650 in 32m with 1.44-1.49v vtt, another one will hit the wall at 630 with same voltage and raising vtt to 1.7v won't break that wall. Also, some cpus have a very narrow range between full stability (pi 32m run) and valid - just 10-15mhz, others may have that range extended - like 640mhz 32m and 690mhz barely stable to validate (but yet 50 mhz difference). I would say C1 stepping is first category, E0 is second category. CPUs are also extremely sensitive to GTLs. So I would bin cpus for fsb validation if that is the goal. I had just a couple of good cpus that would break 690mhz, both were trying to post with air cooling @670mhz. Very important thing to consider - push Vcore for fsb. I needed >1.7v to get 700mhz although core clock was 4.2ghz.

2. Biostar t-power p45, giga ep45t-extreme, rampage exteme (8x serials). There is a chance with P5e3 x48, p5e64 evo. DFI is really annoying, ate all my patience yesterday - so I won't recommend it (though board is still worth mentioning here).

3. Boards run into a kind of wall with fsb, I wouldn't count Vnb as important factor - while it scales, who cares of volts? My safe limits are 1.82v and 45C idle temps.

4. VTT - normally every cpu has a sweetspot, 1.45-1.55v. Vref (aka GTL) should be 0.61-0.65 of VTT, depends on cpu. Vpll - keep it low, high values are killers for 45nm anyway. I have an impression that Vcore should be higher than vtt with a margin of 0.15-0.2v all the time. NB vref is a kind of unknown for me yet, but looks like p45 needs it to be x0.58 of vtt, x48 will need higher values (0.63-0.65).

Thank you so much!

For point 1, Is C1 stepping better than E0 stepping for 700 FSB?

For point 2, I can use EP45-UD3 to boot 650 FSB easily, but I failed to boot 560+ FSB on DDR2 EP45-EXTREME with the same CPU and memory and the same tweak method, how could it be possible?

And how could I know if cpu hit the FSB wall first, or mobo NB hit its FSB wall first?

[TaPaKaH]

1. It really varies from board to board. The same CPU might do 650 32M air in one board, but fail 600 POST in another 650+ capable board. In my 14 years on 775 I couldn't figure this out, so I would normally just pick a dozen best MHz/volt cores on air, put all under LN2 and see what happens.

2/3. Do you expect to bench at 700 FSB with decent efficiency, or are you after the 750 FSB record?
For the former, you would have to use Garbage Extreme or P5E3-Premium (P5E64-WS Evo can also be an option), but those generally don't go above 710-720 no matter how well you bin or cool the NB.
For the latter, EP45T-Extreme seems to be the one to go for. However, I was never able to get one above 600 and always found Biostar TP45HP / Tpower I45 to be super easy for raw FSB. Maybe ASUS P45 and/or Gigabyte P45 UD-series boards can easily do 700+ too, but this has not yet been verified. DFI P45 can potentially do 700+ too, but those boards are a pain to work with - DFI's shitty engineers just couldn't figure things out on their own, passed this down to the customer and claimed 'it's the ultimate tweakers' board".

4. Sweet spot for VTT is usually 1.40-1..50, for PLL is anywhere between 1.25 and 1.60 depending on CPU. Note that only Gigabyte boards can set PLL below 1.50 in the BIOS, for the rest you will need a hardware mod. From my experience, P45 usually scales up to 1.60-1.65 VNB and X48 usually scales up to 1.75-1.85 VNB, but of course there are counter examples.

[TerraRaptor]

38 minutes ago, wytiwx said:

For point 1, Is C1 stepping better than E0 stepping for 700 FSB?

Analyzing first 15 subs give 4 subs with C0 and 11 submissions with E0. So E0 should be better - just keep in mind that you likely miss best cpu if you will only bench it for 32m stability - E0 has higher range of semi-stable fsb clocks.

42 minutes ago, wytiwx said:

For point 2, I can use EP45-UD3 to boot 650 FSB easily, but I failed to boot 560+ FSB on DDR2 EP45-EXTREME with the same CPU and memory and the same tweak method, how could it be possible?

P45 is having cooling issues imho due to open die - stock coolers are of sub-par quality and may not sit tightly on die. I had DFI P45 that wasn't able to accept anything over 1.45v - then I glued IHS on NB and it became way better. So cold NB seems to be the key for P45.

I have also run into strange issues with 32m stability with ddr2 boards - they won't do over 618-620mhz 32m 1:2 (I have tested Rampage Formula and DFI P45-T2RS). Pretty strange both boards hit the same wall (and cpu was definitely not the limit - at least 630mhz with 6-7 ddr3 boards). Probably something needs tweaking in mchbar (use baredit etc).

1 hour ago, wytiwx said:

And how could I know if cpu hit the FSB wall first, or mobo NB hit its FSB wall first?

Just my experience - my best board is capable of running large waza (3-4GB) @666mhz. But even 1M goes bluescreen/"not exact in round". I think that it is cpu issue if board is capable of waza but fails tests stressing cpu-nb link.

[TASOS]

5 hours ago, wytiwx said:

Hey, guys

I want to achieve my 700 FSB OC goal, now I get some questions,

......

 

 

If you are aiming for 700 FSB , start binning cpu's and boards also.

A really good cpu needs a really good mobo to pair with.

The voltage ranges and the rest of the tweaking settings (along with the cooling) will be needed most , for the last top 10-15 fsb

Your best chances are with E0

 

 

 

[wytiwx]

1 hour ago, TaPaKaH said:

1. It really varies from board to board. The same CPU might do 650 32M air in one board, but fail 600 POST in another 650+ capable board. In my 14 years on 775 I couldn't figure this out, so I would normally just pick a dozen best MHz/volt cores on air, put all under LN2 and see what happens.

2/3. Do you expect to bench at 700 FSB with decent efficiency, or are you after the 750 FSB record?
For the former, you would have to use Garbage Extreme or P5E3-Premium (P5E64-WS Evo can also be an option), but those generally don't go above 710-720 no matter how well you bin or cool the NB.
For the latter, EP45T-Extreme seems to be the one to go for. However, I was never able to get one above 600 and always found Biostar TP45HP / Tpower I45 to be super easy for raw FSB. Maybe ASUS P45 and/or Gigabyte P45 UD-series boards can easily do 700+ too, but this has not yet been verified. DFI P45 can potentially do 700+ too, but those boards are a pain to work with - DFI's shitty engineers just couldn't figure things out on their own, passed this down to the customer and claimed 'it's the ultimate tweakers' board".

4. Sweet spot for VTT is usually 1.40-1..50, for PLL is anywhere between 1.25 and 1.60 depending on CPU. Note that only Gigabyte boards can set PLL below 1.50 in the BIOS, for the rest you will need a hardware mod. From my experience, P45 usually scales up to 1.60-1.65 VNB and X48 usually scales up to 1.75-1.85 VNB, but of course there are counter examples.

Hello sam, nice to see.

I just want 700 FSB validation, no for bench or WR.

Can you share some tips about Biostar TP45 HP or TP I45 for raw FSB?

[Noxinite]

Honestly I'm not sure why you guys says the EP45T-Extreme is hard to clock, I've had three boards now and all do >680 and high FSB 2:3 SuperPi 8M easy. So far I've found that clock skews don't really help for valid (or 8M), but more for stability (32M) (maybe why you guys get large gaps between valid and stable on some cpus?). The GTLs, PLL and silicon quality are what seem to dictate FSB. But even if you use the same setting for each CPU you get reasonable results (see below for ambient).

Edited February 23, 2021 by Noxinite

[wytiwx]

13 hours ago, Noxinite said:

Honestly I'm not sure why you guys says the EP45T-Extreme is hard to clock, I've had three boards now and all do >680 and high FSB 2:3 SuperPi 8M easy. So far I've found that clock skews don't really help for valid (or 8M), but more for stability (32M) (maybe why you guys get large gaps between valid and stable on some cpus?). The GTLs, PLL and silicon quality are what seem to dictate FSB. But even if you use the same setting for each CPU you get reasonable results (see below for ambient).

wooo, what a nice list!

Can you shared the tips of ep45-extreme for high FSB?

[Noxinite]

To clarify the FSB quoted is for SuperPi 8M on water.

On the EP45T it seems to simply be high nb ~1.64V and VTT around 1.5-1.6V as well as getting the NB gtl exactly right, CPU gtl seems generally to be okay on auto (as the older bioses do not give access to this and it still clocks fine).

As for FSB batch variation; in that list there are 3x Q842B449 which did 650, 640 and 620, and the 2x 3831A468 both did very high fsb - so you could argue it either way.

Edited February 24, 2021 by Noxinite

[wytiwx]

15 hours ago, Noxinite said:

To clarify the FSB quoted is for SuperPi 8M on water

Is CPU under water or NB?

[Noxinite]

8 hours ago, wytiwx said:

Is CPU under water or NB?

CPU water, NB stock cooler.

[wytiwx]

On 2/25/2021 at 9:40 PM, Noxinite said:

CPU water, NB stock cooler.

That's awesome, then I guess CPU can do 700 FSB validation if CPU can bench Pi 8M @ 680 FSB

[wytiwx]

After some works, finally get 700 FSB validation

https://hwbot.org/submission/4703644

 

Platform can do 680 FSB Pi 1M all by air cooling. PL value reading is wrong, it's 11 actually, not 1 

Edited March 15, 2021 by wytiwx

[wytiwx]

After re-binned my P45 mobo and idle E8 CPU for weeks, I found a holy new and amazing world of raw FSB freq. You can refresh your mobo and CPU wish lists after I submit some results and VCRs, just wait a minute  :P

[Noxinite]

Waiting patiently...

You going to return the favour and post any tips?

[wytiwx]

10 minutes ago, Noxinite said:

You going to return the favour and post any tips?

Will do. I learned many things from you guys

[Gigelz]

i am also waiting

[TerraRaptor]

16 hours ago, wytiwx said:

just wait a minute 

?

[wytiwx]

Sorry to keep you waiting so long, need some time to process my VCRs.

We all know, EP45 UD3R/UD3P are good boards to Core 2 Quard CPU for 500 FSB stability, 

but few OCer would like to use EP45/EP45T UD3 series for 700 FSB goal, they felt not so good, so as I did.

Since Mr. GRIFF made 708 FSB by EP45 UD3P, I begin to pay more attention to these UD3 boards. 

After severals weeks, I found that UD3 boards were really good.

Now, The most important point I want to say is, 

When CPU and NB are both with LN2 cooling, at least 50% of E0 stepping C2D E8X00 CPU can do 700 FSB CPU-Z validation, by EP45/EP45T UD3 series.

If you can't make this, the first thing you suspect should be NB, or BIOS tweak, not CPU.

The important BIOS options on UD3 boards for raw high FSB (in fact, they're almost the same as on the other P45 boards):

Vcore: 1.40V to 1.60V (no need LLC)

Vtt: 1.40V to 1.60V

Vpll: 1.35V (air) to 1.59V (LN2)

Vnb: 1.46V to 1.70V (get a IHS on NB chip! make sure NB cooling is good!)

Vmem: your mem choice

CPU Vref: 0.61x to 0.62x

NB Vref: 0.578x to 0.591x

CPU Clock Drive: 800mV to 1000mV

CPU Clock Skew: 50ps to 100ps

Performance Enhance: "Standard"

Mem strap: 1:1

CPU features: all to "disable"

Other BIOS options remain "auto"

Mem for DDR2 UD board: Micron D9GMH

Mem for DDR3 UD board: Elpida Hyper

VGA: PCI card for 720+ FSB

Edited April 9, 2021 by wytiwx

[wytiwx]

The first VCR: 

E8600 + EP45 UD3R, 710 FSB post on screen, all by air cooling!

In fact, can do max 715 FSB boot on air ;D

and somethings big are coming soon  

[Noxinite]

Very, very nice. Probably can say blanket all P45 is somewhere around that range for NB Ref as on the EP45T-Extreme it looks like I used x0.565.

Suprised that such low cpu clock skew worked fine, I normally need 150ps minimum (EP45T-Extreme again).

Also, I never found anything other than 900ps to work well for cpu clock drive.

As for air post, I think I managed to get to splash screen on old (dead) EP45T-Extreme with 720 all air. Rip, need to have another go at resurrecting it.

[wytiwx]

1 hour ago, Noxinite said:

Very, very nice. Probably can say blanket all P45 is somewhere around that range for NB Ref as on the EP45T-Extreme it looks like I used x0.565.

Suprised that such low cpu clock skew worked fine, I normally need 150ps minimum (EP45T-Extreme again).

Also, I never found anything other than 900ps to work well for cpu clock drive.

As for air post, I think I managed to get to splash screen on old (dead) EP45T-Extreme with 720 all air. Rip, need to have another go at resurrecting it.

My EP45T-Ex boards all worked too bad on air by now, never saw 660+ FSB post, maybe I should rebench too

[ground]

Have you tried any of the really low end gigabyte p45 boards too? I wonder if they can get anywhere close to 700 as well...

[ludek]

I had once gigaboom GA EP45 UD3LR. I have done 624 mhz fsb on my old old e8200 c0 stepping, all air. I was very proud, but have no validation, only screen. It was back in 2009 or 2010, when I was starting my fascination on OC. It was before I found hwbot (polish overclock site with only screenshot to validate)

 

Wytiwx, thanks for sharing the knowledge!

 

Edit:

I have found this! LINK
it was 1.7 Volt, stock NB AIR cooling

Edited April 9, 2021 by ludek