Thanks to all of you - my Prime95 stable 2606 MHz overclock.

All I can say is "uhhhh" - I never thought I could *ever* make it this high - I only started overclocking seriously about ~6 months ago, and I am using fairly inexpensive water cooling (I blew *all* my money on a video card). The famous L12 mod for NF7-S boards did the trick - before that I was stuck at 211 X 12 as my highest overclock (anything higher would freeze prime95 at priority 10, and voltage raises would also freeze things up) - now I am at 217 X 12 - rock stable with a 1.95 Vcore.

Here is a picture of it merrily primeing away (damn, it's hot though) - this failed after 59 minutes (~1.93 Vcore average - way too low, I am surprised it lasted almost an hour) - but I upped the Vcore to average ~1.95 actual, and it has been going for 4.5 hours so far. The problem now is the voltage fluctuation, and undervolting. The Vcore is set to 2.03 in the BIOS for a 1.95 actual. Vcore low hits 1.9 - High is 1.97.



I have met my goal and exceeded it. I don't know what to aim for now... I guess a higher FSB, and a VDD voltage mod is next - for (227 X 11.5) 2610 MHz. It is sad that the addiction wants more and more.

This isn't a boasting post - My most sincerest thanks go out to all of you who helped me do this (I admit that anyone with my hardware & knowledge could do this - but it is your knowledge that gave me all mine) by sharing information, and experiences with me. Wardog - your generous mailing of the conductive pen to me... puts a tear in me eye.

Anyways, whatever, (this is a pretty lame-*** post when you think about it) any comments are welcome, as well as any suggestions about soloutions for the Vcore fluctuation/undervolting, and the high temperatures.

Anyone who wants to lend me their Prometeia - please PM me and we can work something out

congrats!


I wish mine oce'd as half as your but its

Congrats.

To get high frequency and stability, voltage and temperature are the two key components to keep the CPU operating point within the stable multi-dimensional region.

1. Higher voltage gives more current (Idsat) to the transistors in the few critical logic paths (less than 10% of all the circuits) of the CPU so that those transistors can charge or discharge the load capacitance (CL) faster to meet the cycle time requirement between latches. cycle time = 1 / f, where f is the CPU clock frequency.

2. Higher Vcore means more power dissipation into the CPU

power = Vcore^2 C f

where C is the equivalent capacitance of the CPU

3. Higher power means higher CPU temperature T_CPU

(T_CPU - T_ambient) = power cooling_constant


To make the CPU run stably at high Vcore, the CPU temperature T_CPU has to be below certain level or as low as possible so that the temperature sensitive fluctuations of the many variables (such as transistor thermal energy, mobility, leakage current, threshold voltage, ...) inside a CPU over a long period of time would not bring the CPU operating point outside that multi-dimensional stable region.

The higher the Vcore, the lower the CPU temperature has to be in order for the CPU to stay inside that stable region. That is why if a Tbred B runs at 2.2 GHz, the CPU temperature can run stably at a much higher temperature (say 60 C). But at 2.5+ GHz, the CPU temperature has to be relatively lower, say 40 C.

To lower the CPU temperature, two possibilities:
- lower the ambient temperature or
- lower the cooling constant, i.e. better heat sink or better cooling means.

hitechjb1 - you and the "2.5 GHz+ air results" thread are mostly responsible for my learning. Thanks man.

Lowering the ambient temperature is the most effective cooling enhancer right now - but soon I will have another BIX radiator in parallel, and more fans and such (effectively doubling the surface area the water gets cooled by, and adding more "heat capacity" to my circuit). Right now the window is open, and load temperatures are at ~44 Degrees Celcius. It passes prime with the windows closed, but the temperature is scary high (mid 50's - which reminds me of me air-cooling days) - I want this chip to live for at least a year untill I have enough money for an Athlon64 FX -5X rig

(actually, might take more than a year for me to save *that* much )

i can't get my FSB over 220 in single channel and 211 in dual with my NF7 and 2500+, should i try the L12 mod? i thought it was only for tbreds but that SS says you have an athlon XP?

Yep, I have a Thoroughbred AMD Athlon XP 1800+ processor - the L12 mod worked like a charm on it. The socket-wire-pin-mod didn't work though - I had to connect pins physically with a conductive pen. Give it a shot. Wow, there are two wardog's on these forums

You can see pictures of the mod in the ABIT forums, in the L12 mod thread, through the Xtremesystems link.

i was reading on xtreme systems forums it's not for bartons and people who have tried it havent gotten a gain,

btw it's god not dog

hitechjb1 said:

To get high frequency and stability, voltage and temperature are the two key components to keep the CPU operating point within the stable multi-dimensional region.

Click to expand...

amen...

To make the CPU run stably at high Vcore, the CPU temperature T_CPU has to be below certain level or as low as possible so that the temperature sensitive fluctuations of the many variables (such as transistor thermal energy, mobility, leakage current, threshold voltage, ...) inside a CPU over a long period of time would not bring the CPU operating point outside that multi-dimensional stable region.

The higher the Vcore, the lower the CPU temperature has to be in order for the CPU to stay inside that stable region. That is why if a Tbred B runs at 2.2 GHz, the CPU temperature can run stably at a much higher temperature (say 60 C). But at 2.5+ GHz, the CPU temperature has to be relatively lower, say 40 C.

Click to expand...

my bro OCThis taught me the similar thing! although he didn't use such terminology and equations but it had the same meaning. My bro OCthis says he agrees with your statements so well it just gets exciting.. I like it as well. This is very good statement...

Me thinkx you need to meet with Ol'man (another fantastic member). Ya two would fit good in discussing overclocking and cooling. Oh wow oh wow im excited..

Aw, you stopped the Prime95 printout too soon. Almost all my overclock failures that don't fail within the first few minutes fail during the most stressful of the tests, the 896K FFT.

Hoot

GRRRRRRR.

Well, I have run Prime95 for almost 24 hours perfectly - so stopped, and decided to bench in 3Dmark with the new speed.

3DMark freezes at this clock speed, and indeed anything aroun 2600 MHz. I am really crushed - as this overclock is useless to me if it isn't perfectly stable. I am wondering why Prime95 passes, but 3DMark fails.

Back to 25XX Mhz I go.

Maybe RAMsink on my mosfets will help, maybe not...

If anyone has any adice on how I can get this to be 3DMark stable, please help me!

EDIT: Hoot, it passes prime95 on priority ten with ease, but only with a 1.9-2.0 Vcore range (fluctuation is crazy), Vcore average 1.96. Just saw your post now

Ramsinks on the mosfets will almost certainly help. In all my overclocking experience, if the processor is held at a relatively good temperature and hits a brick wall, source power quality is somehow at fault.

Do you have any cooling on the mosfets currently felinusz? Just putting a fan over them might help somewhat. It helped me...of course, I've got to be hardcore and put some heatsinks on there too...winder break project!

I currenly have no mosfet cooling...

I am really mad about this - full prime95 stability - but 3DMark freezes with any FSB over 211. memtest86 passes wity flying colors, ran it all last night.

Guess I will try mosfet cooling, and a VDD volt mod

felinusz said:

I am really mad about this - full prime95 stability - but 3DMark freezes with any FSB over 211. memtest86 passes wity flying colors, ran it all last night.

Click to expand...

I have had the same problems. 3DMark is much more sensitive to high FSB than Prime95 is.

Try having some good/cooler air flow over the MOSFET of the Vcore regulator. When using air cooling, with enough CFM fan (such as TTSFII over SLK- series), the fan overhang over the HS gives this added benefit of cooler air breeze over those MOSFET's.

Prime95 passes and 3DMark01/03 fails may indicate FSB, memory module, memory controller (in NB), video related, some drivers (remote chance) related issues.

A few things to try (if you had not):
- Try 1 module of memory at a time, you can then find out whether a particular memory module is limiting the 3DMark stability, and/or the NB is limiting dual channel operation
- For rev 2.0 chipset motherboard (such as NF7-S rev 2.0), I found that the chipset Vdd and the chipset cooling (especially NB) are not that crucial as in the first rev 1 case. Even with good stock fan and bios Vdd of 1.7V should be sufficient for FSB at your current level (220 - 230 MHz, of course everyboard is different).
- Put a small passive heat sink over SB on top of good air flow may help.
- You may try Vdd mod which is easy to do (I tried w/ a fixed resistor but NOT helping at the level of 230 MHz).
- Make sure good, cooler air flow over the NB and SB chips.
- I suppose your Vdimm is already at 2.9 V (bios max). I suspect Vdimm is limiting my FSB at 227-230 MHz, yet to try Vdimm mod myself (don't know when).

A Southbridge Heatsink is definetely an option that I will try - and mosfet sinks will be added soon as well - as soon as I can get them (going to order them tonight).

Deathknight - how did you overcome the problem? Was there anything in specific that you managed to do, that fixed the problem? For now I will try running with a lower FSB, and a higher mutliplier - say 208 X 12.5 - and I will see how that works out. I want to keep this clock speed (obviosuly), and I guess working on it from a MHz perspective, then getting the FSB upwards afterwards is my best bet.

The lack of airflow over the mosfets is a concern as well, but I cannot see how to rig up a fan over them, without cutting a blowhole. Right now I run with the side off.

felinusz,
OCThis told me to copy and paste the following message for you since my bro is in jail(OC Jail) rite now..

He sez.. "If you got Abit NF7 v2.0 board and the Tbred 133fsb chip, you gonna have hard time getting to the high FSBs(200~230) to maintain "maximum stability" in 3Dmark and 3D games(especially the intensive graphic games, not the old simple outdated games) unless you get lucky as hell.. I'm talking some serious luck, major luck is the one that will get you some serious FSB boost with stability. If your 3dmark 01' fails at 210's FSBs, you WILL fail in 3d games.. even if you think it's doing good.. it will fail soon or later WHACKKKKK.. There are many super dud and medium dud NF7 v2.0 boards flowing around lately. These are the ones cause even more problems at high FSBs... The early release of rev2.0 boards seem to have better qualities, just simply made better, internally and did better in FSBs with the tbred chips.. but of course, still worse than the older generation NF7/s..

Yes, SB cooling *can* be *absolutely necessary* when you Vdd volt mod. Vdd volt mod will jack up temperatures in the board and not only the NB heats up, but SB heats up too, and the entire board becomes unstable too.. At this point, the board is extremely unstable, it is so unstable that it's.. rediculous.. But yes, SB cooling(effective cooling) does help, but will still be unstable than the board that is not Vdd modded and at slightly lower FSB. And no, cooling the NB and SB does not solve the problem. There are something else that's screwing you up.

And no, I'm not saying this will happen exactly to all NF7 v2.0 boards. They have variations of course. But you'll find many of these behave this way. Some of the v2.0 boards did not need any cooling on the SB and still FSB oc better than the other. As for the NB cooling, I can't say it'll work for all v2.0 boards but from my findings out of so many v2.0 boards, they didn't need any cooling on the NB at all.. except the standard factory HSF with white stinky goop.. I also like what the gentleman hitechjb1 has said about not seem to be getting FSB boost by extra cooling the NB on the rev2.0 boards compare to the older revision NF7/s boards. We have similar comment here.

You want to do some serious cooling? I'm talking supercooling the NB.. and Vdd volt mod to death, the rev1.0, 1.1, and 1.2 boards will love you and kiss you... this is no joke.



That's what OCThis sed.. OK, I'm tired of doing this.. He keeps bothering me..

Congrats on the OC. When I get my system together, I plan on using my Tornado(if it isn't sold) to OC. I want to Prime95 and 3DMark it so that I put "the heat on". I will first burn-in the CPU and see if that helps with the OC. The highest I ever took the CPU in my sig is to 2.1Ghz because I had a crappy fan on my SLK-800A. Hopefully, when I get my SLK-900U w/92mm Panaflo, this will all change. I hope to get mine around 2.5 or maybe even 2.6. If I could reach 2.7, that would be amazing.

Pikachu_Mommy - LOL,

That's what OCThis sed.. OK, I'm tired of doing this.. He keeps bothering me..

Click to expand...

You've got voices in your head man!

AND THEY'RE TELLING YOU TO OVERCLOCK

|..|,

What is OCJail anyways?

Thanks for the very informative post though - although it doesn't look too promising (my board is one of the newer ones - I will check the NB stepping tonight to confirm), I am saddened by this . And you are right too - I checked, and the FSB's where 3DMark freezes, the ATi Demo's freeze as well. I will try 208 X 12.5 = 2600 Mhz even - hopefully this will be stable, and benchable, as it is below my "limit" of 211 for 3Dmark.

Give OCThis some thanks from me.

Well, it is stable in Prime95 priority ten, and in 3DMark @ 211 X 12.5 = 2635 MHz. Using BIOS 18, for some reason 10 allows higher prime stable FSB, but fails 3DMark over ~200. For some reason BIOS 18 shows my temperatures as being hogher than 10 did, but everything is stable, an fast as the devil - so I am happy. I will get some screens up tonight!