Update: Asus 680i Observations Part 2: Large Pics Warning....

[Kunaak]

Since I've had more time to mess with the 680i lately, I've found a few more interesting observations that may help some people with these boards.

first off though, I want to say, I am not the expert on these boards, people often ask me questions about this board, that I just have no answer for...
often people have troubles that I simply havent had, so I can't do anything more then give general advice.
so what works for me, isn't always gonna help you, since we all have varying degrees of experience, and very different hardware.
these are just my latest observations, on a very unusual chipset...

first off, some pics to show 500+ isn't impossible with these boards.



this is a straight boot shot.
no tweaks for the super pi test, other then whats in bios.
I am not much of a tweaker when it comes to Super Pi, as I am not usually looking to break records, except my own personal ones.
no clockgen used, not cause I wouldnt use one... I just havent found one for these boards yet.



just a bandwidth shot.
booted at cas 5-4-4-10-10 TRC and rest at auto.
TRC and TRAS were tweaked with Memset, just alittle.



this is my most tweaked shot, with me trying to go for a personal best in Pifast, with this particular processor.
unfortunatly I am just at the limit of what this CPU will do with only 1.6 volts... I need to go back to bios 0302 where the Vcore actually worked...
bios 0702 works mostly, except the Vcore doesnt go above 1.58 volts anymore... seems asus just screws up this board more everytime they release a new bios.
my advice, stick with bios 0302 if you got it.

oh yeah, about PI Fast.
theres 2 major versions out, version 4.1 which is the version supported by Hexus currently. Version 4.3 is out, but its alot faster then version 4.1 by about 5 seconds.
if you ever see a PiFast score thats unusually high, chances are they are using version 4.3 to get the upper hand.



this is just Aquamark3 at 4.1 ghz, nothing too interesting about this shot, just alittle 3D for the hell of it.

8800 GTX at 660/1060.
Mushkin PC6400 at 1240 or 1250 mhz (forget what it really was).
cas 5-4-4-10, TRC 10, rest at Auto.
E6400 at 4.1 ghz.

Now, Observations.

[Kunaak]

heres the best tip I gained for myself recently.

Dividers and FSB.

see, I've always noticed this board likes even numbers.

meaning 2000 FSB (500 FSB's) will often boot, if you use a even number for your ram.

meanning 2000 by 1000mhz ram, will boot.

2000 x 1139 mhz, will not boot.

I've always suspected this board has some very strong dividers, and some very weak dividers, and I think I've narrowed it down to 2 dividers that just seem to work really really well for me.

the 5:6 divider, and the 9:11 divider.
you can only get these from the Unlinked mode, but you probably figured that out already if you seen the inside of these bioses and saw you can't manually select what divider you get when entering in a unlinked ram speed.

in particular, the 5:6 divider is extremely strong, and for me, offers great stability....
but theres a huge draw back.

to do successful 500 FSB's with the 5:6 divider, you need ram that can do 1200 mhz very easy on these boards.

(FSB x 1.2 = Ram speed if using the 5:6 divider)

I will give you an example of how strange the dividers effect your stability.

8 x 500 = for 4 ghz.

unlinked.

set my own ram to 1200 mhz = boot, runs solid, even with tight latency.

same speed and FSB, but set ram to 800, and any latency = no boot... ever.

why? I have no idea... if it boots at 500 and 1200 mhz, why wouldnt it boot at 500 and 800 mhz ram? I have no idea... it just wont for me.
latency and voltage dont matter... it just seems theres a "weak" divider that just doesn't wanna work at that setting.

when I started to see this, I started really experimenting...

500 FSB - 800 mhz ram = no boot.
500 FSB - 1000 mhz ram = works fine.
500 FSB - 1100 mhz ram = boots, but very touchy.
500 FSB - 1200 mhz ram = solid works great.
500 FSB - 1250 mhz ram = solid, works fine, just need alittle voltage.
500 FSB - 1300 mhz ram = boots, but unstable, not usable for long term (haha... I cant honestly expect 1300 mhz to be stable)

now heres the monkey wrench...

1200 was stable at 500...

but try 400 FSB - 1200 mhz ram... and it wont boot.

again, theres some welll just touchy dividers there that just dont seem to like various combinations of FSB and ram.

but heres where it gets truely strange.

to break a 500 FSB easily... you need to have your ram really fast too.
500 FSB + low latency, low speed ram = glitchy, tempermental, doesnt wanna remain stable for long...
but 500 + high speed medium latency ram = very easy to get stable, assuming of course the ram can actually remain stable at high speeds.

Plainly put...

this board already has a very very high learning curve to it, that much was obvious from day 1, however, now beginning to see that theres various dividers and FSB combinations that just wont play nice, can make overclocking with this board, successfully a whole lot of trial and error, which alot of people just either wont take the time to do... or will give up in frustration with.

a 400 FSB is rather easy, regaurdless of experience level.
but 500+ is definatly gonna take some patience, unless you just happen to get a miracle board tha will do it nice and easy.

so basically, be patient, and try combinations with your ram and FSB that seem like they may not normally work.
sometimes you will be surprised...

[Kunaak]

TRFC and D9.

D9GMH in my opinion is best overall all around ram I know of.
its highly resilliant to huge voltages, will do rather high MHz on the RD600, 680i, and the 965 chipsets, which I can confirm cause I have all these boards here, and have tried.
its great for low latencies, and just has a great range of latency and MHz to it. I've used about 5 sets of different D9 so far, and I've always been impressed with the variety of OCs one can get from it.

well, the D9GMH, seems to actually alittle unusual on the 680i chipset... and I cant say I know why... cause it doesnt do this on the RD600 or 965 chipsets.

it doesnt like high voltages... nor does it need it.
thats where it gets truely strange...

see, D9GMH, on the 965 chipset, will typically need about 2.5 volts or more to get 1200 mhz... and that unstable usually, or short term stability.

and it will need nearly 3 volts in most cases to do things like cas 3 at 1000 mhz.

but all that goes out the window on the 680i.

680i, I can get 1250 mhz at 2.25 volts easily stable enough for benchmarks and such.
1200 mhz at 2.25 volts very solid, for day to day use.
thats incredible...

but heres where it gets weird even again.
normal D9GMH seems to need about 3 volts usually to get cas 3 at 1 Ghz.
atleast on the 965 chipset.... but on the 680i... only 2.5 volts or close.

but theres a problem...
D9GMH, atleast in the 3 sets I still currently have, in this board, simply dont like voltages above 2.4 volts.

thats truely strange for a set of ram, I myself have taken to well above 3.4 volts and know has a huge resilliance to high voltages.

so D9GMH just seems to act very strange on this board...
it doesnt need much voltage, and doesnt seem to need it.
works great at 1T, but just plain old wont boot if I try 2.8 volts on it with this board.

so basically...

with how just about every company out right now uses some form of D9, especially D9GMH, chances are high if you have quality ram, you might have some luck with this tip...

if your having troubles with your D9 and your trying to get it stable by adding voltage...

do the opposite, and keep the same speed... just lower the voltage.
believe me.... if I thought I could do 1250 mhz easily at 2.3 volts on my P5B deluxe, I'd think I was dreaming.
when I first started trying the 680i, I did exactly what I always did on the 965... cant get it stable... add more voltage.
but with the 680i do the opposite...
cant get your ram stable?
go lower...
it sounds weird, but I have 3 sets of ram here, that all do the same thing...
less voltage - gets me better stability.


ok, I didnt forget the TRFC.

the TRFC on the 965 chipset, is limited to about 42 tops... which people like to say is slow.
so the tendency is to try to get lower, at 35, then 30 and so on...
but honestly, TRFC really isnt a ram setting thats ever gained me much in any benchmark or game, its more a psychological number...
42 just sounds really really high, and as overclockers, we dont like things that sound slow, haha...

but on 680i.... if 42 sounds slow to you, wait till you see TRFC of 100+... cause its there and you can use it...

now, coming from the 965 chipset, my natural tendency is to wanna take the TRFC to 42 or less... just like my P5B deluxe.
while this will work... your gonna find out something real quick...

you will get alot more "freezes" by manually setting the TRFC, then if you just let it be at Auto.

Auto TRFC tends to set the TRFC at 66 if your using 1200+ mhz on the ram.

stability is alot easier to maintain, at any ram speed, if you let the board figure then TRFC instead of trying to manually force it to a setting you want.

remember...
sure, it gives you bragging rights if you think TRFC of 20 or lower is great for some benchmark, and you can get it there, and others cant... but if your at all like me... you gotta live with these settings everyday all day.
theres more to overclocking, then just benchmark speeds, and if your just increasing your boards frequency of freezes, cause you think TRFC of 66 or whatever is too high for you to cope with...
then you might as well get use to random freezes and crashes.

otherwise, for the rest of us, that dont really want to deal with the psychology of numbers, and bragging, and just want stability...

its just best to leave TRFC alone on these boards.

for me, any TRFC setting I set manually, just results in occassional freezing.... when it be 3 hours from now, or 3 minutes, wont matter, cause it always happens.

but leave the TRFC at Auto... and you can run all day, with little to no chances of freezes... assuming of course, that your actually running a stable overclock.

[Kunaak]

Last Observation...

450+ isnt impossible or even hard with the 680i and Quad Core.
I've done it, and done it stable on air...
but that was a few weeks ago when I first got my quad core.
I plan to put that CPU back in this board, and do it again... to show it definatly can be done.

so this spots reserved for alittle Quad Core stuff.

Update:

After Much trial and error... and some advice from a friend Yoxxy, testing some voltages and 1000 reboots...

I found the trick to 450+
but its not a fun thing to do, cause it has a huge drawback...

first off, high FSB's seem to respond to the 1.2 VHT voltage, to above 1.5 volts, and the NB voltage to above 1.7 volts.
you need these voltages high to get it bootable most of the time...

but still it will not wanna boot sometimes, and theres a simple trick to fix that.

power down, and power up.
no reset, just power down and power up from a cold boot.
this seems to work far better then the soft reset from bios.

Lastly,

Lower Multipliers are far easier to boot, then high mulipliers.

6x = Hard.
7x = Harder...
8x = Very Hard...

any higher then this, I dont know.
my CPU only goes to 8X max.

the advantage of running 6 x 450 = 2.7 ghz

versus, 8 x 333 = 2.66 ghz...

besides the obvious, is that once you pass about 430 on the FSB, the latency of the chipset is greatly loosend, so then you get the chance to run 1200 and 1300 Mhz on the ram.

so it's not entirely useless, but not easy still...

but it is possible, just not what I was hoping for.

[vanovich]

nice findings.im still testing ,but sofar ,500fsb and unlinked ram at 400 3 3 3 8 timmings 1T is running stabil . how far up kan you go on 1T ? i feel the system is much quicker with tighter timmings and not so high mhz on the ram.im running with 701 biose which came with the mb ,and it seems that voltages are working fine,besides a bit of drop on the vcore .

[Johnny Bravo]

Fantastic work Kunaak, I shall see how these finding weigh up on the Abit IN9, I've found this chipset to be quite quirky but always thought 1:1 was a safe bet in terms of dividers, judging by your findings I'll have to be a little more careful!

As you say it very much is a case of throw the rule book out the window!

[MTP04]

Worth reading, good info. Thanks Kunaak

[espionix]

Thanks, this helped

[Skrips]

Good times!

[virtualrain]

Last Observation...

450+ isnt impossible or even hard with the 680i and Quad Core.
I've done it, and done it stable on air...
but that was a few weeks ago when I first got my quad core.
I plan to put that CPU back in this board, and do it again... to show it definatly can be done.

so this spots reserved for alittle Quad Core stuff.

Awesome info Kunaak!

I for one, can't wait for more Quad Core results and tweaking guidance!

[OldGuy]

Good stuff Kunaak...I'll have an IN9 tomorrow and maybe some of your tips will apply. I had an evga for less than a week and it "seemed" to have a lot of dead spots and maybe those dead spots were really from the memory dividers. At least I'll have some ideas of what to look for and not lock my poor little brain into a path of conventional wisdom

[msgclb]

About a week ago my 680i crashed. I was using a Vcore of 1.60V to get my E6400 running stable at fsb 400. My memory was at ddr 1000 at 2.2V. When I ran Memtest86+ v1.70 I got all kind of errors. The system was stable with another set of memory.

A few days later I'm working on the text for a rma when I decided that I needed more information. I reinstalled my Crucial Ballistix DDR2-1000 memory and to my suprise I got no errors. At first I'm wondering if they healed themself but I soon realized that I forgot to change the memory voltage from 1.90V to 2.2V.

For the past few days I've been trying to find what voltage will cause my DDR2-1000 memory to fail and now these observations by Kunaak.

Up to this point I haven't been able to get my E6400 stable past a fsb 400 on this 680i. In the next few days I'll see if this E6400 can go any higher using these observations.

[Kunaak]

one of the most important lessons that we all tend to forget, is sometimes, voltage only makes things worse, not better.

my Quad core is the same way...
I was pulling my hair out trying to get 3.6 ghz stable, and just kept upping the voltage from 1.5 to 1.55 and 1.6 volts and so on...
then one day I just started going backwards... lower the voltage, and then prime.

I found that it didnt take the initial 1.5 volts I thought, but 1.38 volts to get the 3.6 ghz stable.
see, at 1.5 volts, I am simply overloading CPU and cooling becomes ineffective when messing with 4 cores, thats alot of heat to try and cool.
1.4 or less turned out to be the sweet spot.

even my old Corsair 3500LL, that ram was the strangest DDR I ever seen.
like always, I just kept trying to get higher voltages to solve the problems...
then I just started trying to go lower and lower.

by the time I was done, I was using 2.25 volts, and getting cas 2.5-3-2-1 and 1T stable, at 254 HTT speeds.

you can see all my results from that ram here, in which theres DDR1 doing DDR2 voltages...

it was truely the strangest ram I ever seen... even to this day.

http://www.xtremesystems.org/forums/...hlight=corsair

voltage, doesn't always solve the problem.
so be willing to try less, that can yield some interesting surprises sometimes.

[OldGuy]

Yeah, that is an old habit...reach for voltage first, ask questions later

[msgclb]

oh yeah, about PI Fast.
theres 2 major versions out, version 4.1 which is the version supported by Hexus currently. Version 4.3 is out, but its alot faster then version 4.1 by about 5 seconds.
if you ever see a PiFast score thats unusually high, chances are they are using version 4.3 to get the upper hand.

The download page for PiFast only lists version 4.3 so unless you have an alternate site it looks like I'll have faster times. Actually using Google I found this link that is for version 4.1...

http://files.aoaforums.com/I1652-hexus_pifast.zip.html

Also your PiFast screenshot doesn't show your input choices. I noticed in the zip file there's a file hexus.txt that I'm guessing are your input choices. Am I correct?

[Liudavid]

Wonderful article. Thanks for the effort.

[linflas]

Thanks so much Kunaak, you verified something I have suspected about this board and memory voltage, I have the PC6400C5 from corsair, the promo stuff that corsair told me to run @ 2.1V on the board, but I always had stability issues at that voltage, dropped it down to 1.9 and was able to get them very stable with cas5 settings 5-5-5-15 @ 950mhz, ran that for 3 days. Company of heroes for many hours, no problems, 12 hours blend prime stable.

My crucial pc8000 I am currently running @960 4-4-4-8 @ 2.0V, not the rated 2.2V, also 12 hours blend prime stable. And 3d stable as well.

All of these are at 400fsb x9 multi on an E6600 @ 1.4V bios, so really about 1.35

I have never been able to get 1T on either set of dimms to boot from the bios. I am by no means knowledgable about memory overclocking, I tend to like lower mhz with tighter timings, and lower voltage anyway.



http://valid.x86-secret.com/show_oc?id=163595

Thanks for the great post, I look forward to abit releasing a new bios for my board so I can get through the massive hole in the fsb for me. I cannot boot anything between 401-430, abit has issued a statement through an unstable bios changelog that never made it to release that they are aware of the holes, and are working to resolve them.

Thanks again, great read.

[Naughtyboy]

Very good writeup Kunaak...useful stuff, same as your last one.

Kunaak A question thou...did you change you NB/SB cooling to acheive theese results....or are you still on stock cooling..?


And a question to anyone using E6600 or higher cpus..have anyone been having problems running 390-410mhz FSB on multi. 9x..or anyother multi. for that matter(except from the post above)
Because I can´t for the love of my life get it working at 400 FSB...tried both multi. 9 and 8...at multi 8 it posts but then freezes.

[Leeghoofd]

3600 no worries, FSB for mine is 452fsb now (bootable from bios), but will need to replace the northbridge cooling, waterkit being installed soon. anyone got

Look at the voltages mate on my screenie maybe you got some wrong

[Kunaak]

[G.Foyle]

That's some really interesting info, Kunaak, thanks for help!

@ Naughtyboy: I tried several times to boot at 400x9 or 400x8 and it always froze, until yesterday when I read Kunaak's first post (he was writing the rest at the very moment) and some strange thought struck me that I should OC now, and I tried, and lo! it booted fine but the sound was messed up so i only took a cpu-z screenie and spi and turned it off.
If you have problems try what Kunaak said before: cold boot - the board doesn't want to boot at first time you set high fsb.

[ooztuncer]

now heres the monkey wrench...

1200 was stable at 500...

but try 400 FSB - 1200 mhz ram... and it wont boot.

again, theres some welll just touchy dividers there that just dont seem to like various combinations of FSB and ram.

but heres where it gets truely strange.

OK, this was one of the theoretical questions in my mind when I moved to c2d platform.

What is the role of mem controller or something in that line? I even opened a thread about it without much interest on the topic.

In my mind here is the culprit (rather than a divider problem):

at fsb=400 with tight nb latencies (1066 strap), somehow cpu can't hold fast ram speeds at the same time.

at fsb=500 (due to relaxed nb latencies), cpu is able to run the ram at 1200mhz.

It reminds me of socket 939 platform (it is almost impossible to merge maximum ram speed with maximum cpu speed).

Any insights on this idea on c2d's or was it already a known issue?

[Naughtyboy]

Yeah I got it working now....but in a strange way. I read this post and said to my self -"I´m gonna get this baby working"
So I went at it...but nothing seemed to work..no matter what I did it simply just either frooze at post or didn´t boot at all....until someone or something told me to try the 5:6 divider Kunaak was talking about. I set 400FSB unliked and memory at 480...saved and voliá it booted and went straight to windows..ran SPI32M without a glitch.
And now I can boot at 400fsb regardless of what other settings I use....this s***t beats me for sure.....have no clue what was wrong.

[Kunaak]

Yeah I got it working now....but in a strange way. I read this post and said to my self -"I´m gonna get this baby working"
So I went at it...but nothing seemed to work..no matter what I did it simply just either frooze at post or didn´t boot at all....until someone or something told me to try the 5:6 divider Kunaak was talking about. I set 400FSB unliked and memory at 480...saved and voliá it booted and went straight to windows..ran SPI32M without a glitch.
And now I can boot at 400fsb regardless of what other settings I use....this s***t beats me for sure.....have no clue what was wrong.

dont feel bad...

I think I've spent well over over 150 hours with this board, in bios alone... I mean doing nothing, not a single thing... other then tweaking one setting, and seeing if it booted.
try another, and see what happend.
then seeing if I could recreate the failures... and try to figure out why they happend....

and even after all this time... I really can't say I figured this board inside and out... its definatly a finicky board, but theres a whole lot of potential in the board.... it just takes alot of trial and error figuring it out...

if anything I do helps you, thats cool, thats what I like to hear.

I like that the 5:6 divider worked so well for you, that gives me abit more confidence that I am onto something with the "some dividers are stronger then others" idea of mine...
which just makes me wanna try it more.

[Kunaak]

In my mind here is the culprit (rather than a divider problem):

at fsb=400 with tight nb latencies (1066 strap), somehow cpu can't hold fast ram speeds at the same time.

at fsb=500 (due to relaxed nb latencies), cpu is able to run the ram at 1200mhz.

Any insights on this idea on c2d's or was it already a known issue?

you may very well be onto something...
I really didnt test the chipset latencie this time around, since I really just was focusing on 500+ on the FSB.

but now 400 and high ram speeds is definatly something I need to try now.

if 1200mhz+ ram is as easy at 400 or even 300... as it is with 500 FSB's. then that might just show that the latencie of the chipset isnt whats determining the OC ability of the ram...
or vice versa, so its definatly something to look into.

I will give that a shot too...

but the thing I wonder is...
see, I know the 5:6 divider is very strong...
but when trying to run say 8 x 400 or even 8 x 300... and 1200 mhz ram... what dividers will I get? will the dividers be strong ones... or weak?

cause if the chipset latency was the only thing making 1200 mhz possible...
then all dividers regaurdless would be "strong".... but thats not happening.
alot of ram/FSB combinations past 500 just simply wont work...

man... now I feel like I am trying to figure out the whole "what came first, the chicken or the egg" theory....