It's still going, 10 hours and counting. It's stable enough.
I7 4Ghz OC Stability...
- Thread starter Gaidin109
- Start date
It's still going, 10 hours and counting. It's stable enough.
LLC controls vdroop to try to eliminate dangerous spikes in voltages.
Did you look at the link? LLC forces a steady voltage regardless of load, attempting to emulate a perfect voltage source. However it does not have a perfect voltage source available, so when the load changes, the voltage stays 'constant' but rings around the mean value.
The voltage set in the bios is never meant to be a 'target' voltage, it's the voltage intended when completely idle. When drawing more current, the voltage across the processor decreases much like any other circuit component. Intel expect and design for this. By enabling llc you are forcing the voltage to run out of intels spec for the comfort of being able to type in a lower number, the penalty is that the ringing drives far higher voltages through the chip when transitioning from idle to load. I believe you can even watch the voltage doing this using everest, the resolution is sufficient to guess at how high the voltage spikes.
If you were really keen, I believe fourier analysis would allow you to model this. If you'd rather put your fingers in your ears and sing loudly, happy in the knowledge that you've set a lower number in the bios, then do so.
As a more intuitive way of looking at it, in case the graphs are not immediately obvious by themselves. Heat and low voltage combined kill processors, either by itself doesn't matter much. So with llc disabled, you get a higher voltage when idle. However with it enabled or disabled, you need the same voltage under load to maintain stabilty.
So all llc permits is a lower idle voltage, which is when the chip is cool and the voltage doesn't matter anyway. The penalty is driving large transient voltages through the chip, and presumably working the motherboard power circuitry a lot harder. The former is dangerous, the latter surely hurts stabilty rather than helps.
The voltage set in the bios is never meant to be a 'target' voltage, it's the voltage intended when completely idle. When drawing more current, the voltage across the processor decreases much like any other circuit component. Intel expect and design for this. By enabling llc you are forcing the voltage to run out of intels spec for the comfort of being able to type in a lower number, the penalty is that the ringing drives far higher voltages through the chip when transitioning from idle to load. I believe you can even watch the voltage doing this using everest, the resolution is sufficient to guess at how high the voltage spikes.
If you were really keen, I believe fourier analysis would allow you to model this. If you'd rather put your fingers in your ears and sing loudly, happy in the knowledge that you've set a lower number in the bios, then do so.
As a more intuitive way of looking at it, in case the graphs are not immediately obvious by themselves. Heat and low voltage combined kill processors, either by itself doesn't matter much. So with llc disabled, you get a higher voltage when idle. However with it enabled or disabled, you need the same voltage under load to maintain stabilty.
So all llc permits is a lower idle voltage, which is when the chip is cool and the voltage doesn't matter anyway. The penalty is driving large transient voltages through the chip, and presumably working the motherboard power circuitry a lot harder. The former is dangerous, the latter surely hurts stabilty rather than helps.
Soldato
I booted into Windows with 19x multi and 210Bclk earlier, but it was far from stable, as soon as LinX loaded up my cores, I got hit with a BSOD!!
Stop Code 0x00000101, seems to be the same stop code everytime it throws a BSOD...
But with 191 x 20 my system is stable with 1.248v in CPUZ, and 1.300v QPI/Vtt, IOH core is around 1.160v, Vdimm @ 1.66v and all other main voltages have been raised up a couple of notches as well...
Temps are crazy IMO.. it idles @ 51c, and loaded hits 88c!!!
Absolutely gutted guys.. I really don't want this friggin processor.. I asked to have my previous one sent back out, but they couldn't find it...
Not a happy bunny at all.. With my last chip I could boot into Windows @ 4.4GHz, but when I tried similar settings with this chip it BSOD'ed straight after the Vista Loading bar...
My old chip was perfectly stable @ 4GHz (191 x 21), so this is the main reason I am so annoyed, because I have been given an extremely poor chip in place of a not so bad one...
Last chip was 3851A.. seems that these 3910A's aren't very good at all..
I hope that something can be done about this processor, because I don't want it.. I just want my old one back!!!!
Stop Code 0x00000101, seems to be the same stop code everytime it throws a BSOD...
But with 191 x 20 my system is stable with 1.248v in CPUZ, and 1.300v QPI/Vtt, IOH core is around 1.160v, Vdimm @ 1.66v and all other main voltages have been raised up a couple of notches as well...
Temps are crazy IMO.. it idles @ 51c, and loaded hits 88c!!!
Absolutely gutted guys.. I really don't want this friggin processor.. I asked to have my previous one sent back out, but they couldn't find it...
Not a happy bunny at all.. With my last chip I could boot into Windows @ 4.4GHz, but when I tried similar settings with this chip it BSOD'ed straight after the Vista Loading bar...
My old chip was perfectly stable @ 4GHz (191 x 21), so this is the main reason I am so annoyed, because I have been given an extremely poor chip in place of a not so bad one...
Last chip was 3851A.. seems that these 3910A's aren't very good at all..
I hope that something can be done about this processor, because I don't want it.. I just want my old one back!!!!
Caporegime
I had a brief read through this thread jon, i remember when the p45 boards came out, such as my p5q deluxe, a very experienced clocker on the boards here said, its better to disable LLC and simply set your vcore a notch higher, i had it enabled on my previous 3.8ghz q6600, still use it on my current q9550 at 3.8ghz, is it generally reccomended not to use it?
Sounds like it could just be the temperatures causing it problems, does life improve if you cool it down more? Temporarly point a 12" fan into the case or similar
I'm pretty certain this is the case setter, but it's always tricky to be flying in the face of public opinion. Perhaps the electrical engineer at work will know something about it, I think I'll have to describe the symptoms and he's going to laugh at the idea of actually making motherboards do this.
What I know for certain is this, and I'd be interested in your thoughts on it.
Voltage across components decreases with increase in current drawn, as long as the voltage source is not perfect. This is why a potential divider is a crap way of supplying a fixed voltage, and why the power circuitry on a motherboard is quite intense. Intel know this and presumably design for it.
I also know that vdrop and vdroop are interpreted by the community as motherboards being made badly. The former is a standard calibration issue easily solved with a multimeter, and the latter is the above. The extent of vdroop depends on how good a voltage source is available, but it is definitely meant to happen for any current drawn.
More recently, load line calibration was introduced, because the enthusiast community was bitching about vdroop and vdrop. I'm pretty certain the electrical engineers know this is a daft thing to do, but they may well judge the ringing effect as negligible compared to people thinking their boards are rubbish. I'd certainly introduce a daft feature if it would sell more boards.
Now, I think the above is very plausible. Even down to knowing I would make the same decision in the engineers place. What I don't know is whether or not the ringing matters. Everest picks it up at about 2V, and would read a higher voltage were the resolution better. So briefly a very high voltage is passed through the chip to eliminate the vdroop, and this has to be so by the very nature of abrupt voltage changes. This may well be harmless, but I'm putting my bets on it isn't.
Which leads back to what it actually achieves. Lower voltages while idle. Idle voltages can go to hell, they just don't matter very much. The processor needs a given voltage to hold stable under load, and it gets a clean, stable voltage under load with or without llc. All that changes is the number you pick in the bios, and whether it reaches this voltage gradually or with spikes.
Sadly it is generally recommended to use it. I would be very interested in the opinion of some of the guys who have been doing this a long time, I remember finding out that maximum voltages always referred to cpu-z under load when just starting out (e8400/G35, so hardly been at this a long time). The idea of maximum voltage matching the bios seems a new one. So the best I can offer is the above, and to see what you make of it. I look at it as negligible benefit at the cost of stressing the power circuitry a lot more and putting unecessary transient spikes through the processor.
Cheers
Here's a thread from march 09 about it, pretty much sums up the issues I think
And here's one from ocforums
Tweaktown, no sources but a senior member making a lot of sense
And finally this was going to be my example of the standard rubbish talked about it by people who just haven't bothered to look into it. Sadly (sort of) it degenerated into a few of the more experience members describing llc very well and arguing against using it
So. Its something a lot of people speak up about who actually know **** all. Amazing place the internet. I think I've got this one down though, and will not be using this 'feature' myself. I recommend you try the overclock without it setter, though I'd hope you make up your own mind and let me know your decision.
I cautiously tick it off the list of things I don't yet understand. Currently working on 'turbo' which I suspect to be rubbish, gtl refs, determination of sane limits for chipset voltages, and placement of a second pump in series halfway down a loop vs adjacent vs two loops.
I'm pretty certain this is the case setter, but it's always tricky to be flying in the face of public opinion. Perhaps the electrical engineer at work will know something about it, I think I'll have to describe the symptoms and he's going to laugh at the idea of actually making motherboards do this.
What I know for certain is this, and I'd be interested in your thoughts on it.
Voltage across components decreases with increase in current drawn, as long as the voltage source is not perfect. This is why a potential divider is a crap way of supplying a fixed voltage, and why the power circuitry on a motherboard is quite intense. Intel know this and presumably design for it.
I also know that vdrop and vdroop are interpreted by the community as motherboards being made badly. The former is a standard calibration issue easily solved with a multimeter, and the latter is the above. The extent of vdroop depends on how good a voltage source is available, but it is definitely meant to happen for any current drawn.
More recently, load line calibration was introduced, because the enthusiast community was bitching about vdroop and vdrop. I'm pretty certain the electrical engineers know this is a daft thing to do, but they may well judge the ringing effect as negligible compared to people thinking their boards are rubbish. I'd certainly introduce a daft feature if it would sell more boards.
Now, I think the above is very plausible. Even down to knowing I would make the same decision in the engineers place. What I don't know is whether or not the ringing matters. Everest picks it up at about 2V, and would read a higher voltage were the resolution better. So briefly a very high voltage is passed through the chip to eliminate the vdroop, and this has to be so by the very nature of abrupt voltage changes. This may well be harmless, but I'm putting my bets on it isn't.
Which leads back to what it actually achieves. Lower voltages while idle. Idle voltages can go to hell, they just don't matter very much. The processor needs a given voltage to hold stable under load, and it gets a clean, stable voltage under load with or without llc. All that changes is the number you pick in the bios, and whether it reaches this voltage gradually or with spikes.
Sadly it is generally recommended to use it. I would be very interested in the opinion of some of the guys who have been doing this a long time, I remember finding out that maximum voltages always referred to cpu-z under load when just starting out (e8400/G35, so hardly been at this a long time). The idea of maximum voltage matching the bios seems a new one. So the best I can offer is the above, and to see what you make of it. I look at it as negligible benefit at the cost of stressing the power circuitry a lot more and putting unecessary transient spikes through the processor.
Cheers
Here's a thread from march 09 about it, pretty much sums up the issues I think
And here's one from ocforums
Tweaktown, no sources but a senior member making a lot of sense
And finally this was going to be my example of the standard rubbish talked about it by people who just haven't bothered to look into it. Sadly (sort of) it degenerated into a few of the more experience members describing llc very well and arguing against using it
So. Its something a lot of people speak up about who actually know **** all. Amazing place the internet. I think I've got this one down though, and will not be using this 'feature' myself. I recommend you try the overclock without it setter, though I'd hope you make up your own mind and let me know your decision.
I cautiously tick it off the list of things I don't yet understand. Currently working on 'turbo' which I suspect to be rubbish, gtl refs, determination of sane limits for chipset voltages, and placement of a second pump in series halfway down a loop vs adjacent vs two loops.
Last edited:
Caporegime
Jon, tbh mate i know sod all about electronics and voltages (im a simple painter and decorator who loves messing about with pc's)
but i remember reading on xs forums about LLC, ive heard it can be detrimental, particularly with 45nm quads, prior to my current q9550 i ran a q6600 @9x422 fsb, LLC enabled, kept it at that speed for a few months then sold it, got the q9550 at a very nice price to try one out. I remember the vdroop/drop on my old 680i e6600 combo being very extreme, pencil mod was the reccomended solution.
but i remember reading on xs forums about LLC, ive heard it can be detrimental, particularly with 45nm quads, prior to my current q9550 i ran a q6600 @9x422 fsb, LLC enabled, kept it at that speed for a few months then sold it, got the q9550 at a very nice price to try one out. I remember the vdroop/drop on my old 680i e6600 combo being very extreme, pencil mod was the reccomended solution.So what your saying is that when I just disabled the LLC in my bios that my cpu didnt use the same 1.232v idle regardless, and then drop under load to 1.8v and then to 1.2v to make my OC completely unstable.
And when I re-enabled it, funnily enough the cpu still uses 1.232v idle, but drops to 1.216v under load making my OC stable again.
Oh, and I must misunderstood my motherboard manual when it says it controls vdroop to keep the voltage more constant. No spikes in other words.
within the passage quoted is this
" Finally, let's take one last real-world look at the consequences of removing Vdroop. ASUS' implementation of this feature, labeled as Load Line Calibration and included with their latest line of motherboards, is particularly worthy of our attention for a number of reasons. The first is that setting lower voltages with this option enabled actually results in a condition in which the CPU voltage under load is higher than the idle voltage."
LLC as implemented by the Gigabyte UD5 mobo doesnt do this, it allows vdroop, it doesnt eliminate it. It also doesnt raise the cpu voltage under load, in fact the opposite is true and it keeps that voltage at a stable voltage depending upon the load on the cpu, dont use LLC and then leave the vcore on auto or anything silly like that and it will help your OC and will not damage your cpu as you suggest. Although it would seem that ASUS's implementation gives a constant voltage no matter what load is going through the cpu which seems strange.
Caporegime
Ive found with my q9550, (1.200 vid chip) i set 1.21875 in bios, boot into windows, i get 1.200, when i put the system under 100% prime small fft load, my vcore in cpuz goes to 1.216, for 3.8ghz on a 45nm quad this is still pretty low from what ive seen on various forums etc.
Last edited:
Sounds like a very good voltage for a 3.8ghz q9550. Rather suggests turning llc off, since it's not going to struggle to get stable anyway. P5Q; I think I was dropping from 1.35V to 1.2625 under load on my one. If you jsut turn it off, it's going to be unstable.
If you turn it off, up the vcore to somewhere south of 1.3625V and load it, then note down how close to 1.2V it drops you'll be able to work out, probably with a couple of attempts, what you need to set over 1.2V to have it droop to 1.2V under load where it'll be stable and generally a happier board. Shouldn't need to change any other settings.
Of course it might be unecessary, but it's definitely the safer option.
I really, really miss my q9550. Not at all convinced by the 920 I'm on now
If you turn it off, up the vcore to somewhere south of 1.3625V and load it, then note down how close to 1.2V it drops you'll be able to work out, probably with a couple of attempts, what you need to set over 1.2V to have it droop to 1.2V under load where it'll be stable and generally a happier board. Shouldn't need to change any other settings.
Of course it might be unecessary, but it's definitely the safer option.
I really, really miss my q9550. Not at all convinced by the 920 I'm on now
Well, no, they wouldn't be. Cpu-z wont spot it either, nor will a dmm. An oscilloscope would do it beautifully, or everest takes a fair shot at it.
If I've encouraged you to try it, I'm pleased. The links in my last post may be persuasive even if I'm not. Good luck!
If I've encouraged you to try it, I'm pleased. The links in my last post may be persuasive even if I'm not. Good luck!
Caporegime
JonJ678;14682386 So. Its something a lot of people speak up about who actually know **** all. Amazing place the internet. I think I've got this one down though said:Its somehing ill definitely try, im currently running 8.5x450 fsb, vcore in bios of 1.21875, pll, 1.54, nb. 1.26 and vtt at 1.26, theese setings are prime smal fft stable, im gonna try and run blend, but one interesting thing i discovered with the p5q dlx, was cpu margin enhancement, i changed this from optimized to performance and it helped to achieve stability on both small fft and blend at 425 fsb, im gonna push to 471 fsb for 4ghz but i think ill need to start looking into gtl settings, cooling wise im more than covered, but theese 45nm quads are a bit more trickier than their older counterparts, though saying that my current 3.8ghz setup is amazing, encodes video at a nice rate, plays blu ray to my 32 inch telly and games on my 24" monitor.But the clocker in me wants more mhz lol
Last edited:
Caporegime
I might be joining you on a 920 soon, geting some good paid work in england, and thinking of geting an i7 setup with an sli x58 board, current s775 rig might replace my old athlon s939 4000 system.
Yeah, no kidding they're more difficult. The e8400 I had just went faster and faster the more voltage I offered it. The q9550 would throw a wobbly if I increased the northbridge voltage above 1.28, mad.
gtls refs can help reduce vcore, but i dont think they're expected to do anything else. I made a crude attempt at working them out by taking stable at 3.7 (I think), lowing vcore until just unstable then playing with the gtls to bring it closer to stabilty, leaving them set to the optimums. They drift with fsb, but at least then you've got a better starting point, I intended to recalibrate them every 100mhz or so but never got it stable above 3.8 (c1 stepping may not have helped)
Cpu margin enhancement drove me mad. I'd have been far happier with it absent, it's an absolute nightmare trying to find out what it does, or even if compatible/performance/optimized is meant to be better. I don't mind lots of options, but completely undocumented ones which behave fairly randomly annoy the hell out of me.
Whats your boards fsb limit? Mine ran stable up to 475 but I couldn't get it further, even at x6. Saw your thread in the motherboards section, in the same position as you a while back I decided to start watercooling. Better toy, improvement from q9550 to 920 looked too slight to bother with.
Sadly my motherboard stole my cpu (and psu... and ram...) from me, and the i7 I have now is an apology for the rma taking so long. Its hard to convince rma people that your processor is actually dead, and it's not you being foolish
I hadn't noticed you were Irish, but you love the ocuk free shipping. I won't be doing anything exciting with my system until I can source a new top for my ek block, so perhaps we'll end up clocking the same board at the same time. Got to be fun
gtls refs can help reduce vcore, but i dont think they're expected to do anything else. I made a crude attempt at working them out by taking stable at 3.7 (I think), lowing vcore until just unstable then playing with the gtls to bring it closer to stabilty, leaving them set to the optimums. They drift with fsb, but at least then you've got a better starting point, I intended to recalibrate them every 100mhz or so but never got it stable above 3.8 (c1 stepping may not have helped)
Cpu margin enhancement drove me mad. I'd have been far happier with it absent, it's an absolute nightmare trying to find out what it does, or even if compatible/performance/optimized is meant to be better. I don't mind lots of options, but completely undocumented ones which behave fairly randomly annoy the hell out of me.
Whats your boards fsb limit? Mine ran stable up to 475 but I couldn't get it further, even at x6. Saw your thread in the motherboards section, in the same position as you a while back I decided to start watercooling. Better toy, improvement from q9550 to 920 looked too slight to bother with.
Sadly my motherboard stole my cpu (and psu... and ram...) from me, and the i7 I have now is an apology for the rma taking so long. Its hard to convince rma people that your processor is actually dead, and it's not you being foolish
I hadn't noticed you were Irish, but you love the ocuk free shipping. I won't be doing anything exciting with my system until I can source a new top for my ek block, so perhaps we'll end up clocking the same board at the same time. Got to be fun
Well, it seems that I only had to increase the vcore to 1.275 (was 1.25) in the bios to
get the same voltages under load that made my system stable with LLC enabled. I think I'm starting to see what is happening, With LLC enabled the voltage when under load is constant even though the load may not be, whereas with it disabled the voltage matches the load thus eliminating the spike when going from contant load to constant idle voltages and as all this happens in milliseconds cpuz or mobo warnings cant pick up the spike quickly enough.
If you do a prime95 cool down test, I think you can see this happening as the different loads give different voltages as the load decreases, yet with LLC enabled the voltage stays the same until the cpu is completely idle and then increases.
Also an offshot of this is lower temps at load by a few degrees, or so preliminary tests show. Will have to do a 6hr Prime95 blend to see for sure, but I'm confident it will be fine.
Am I correct in my assumptions...?
Caporegime
You gotta love the E8**** duals, my nephew of 14 has one at 4.5ghz on air for day to day running, no gtl tweaking, pity the quads arent the same, ive heard that the c1 stepping quads were a bit more difficult to clock, like an idiot i sold a q6600 that ran comfortably at 3.8ghz on air and got a q9550, as for fsb limit jon, its something ive never tried tbh, i started with the default multi of 8.5, first oc was 400 fsb on stock volts of 1.200, had to go up to 1.21875 for 8.5x425 and up nb and vtt to 1.26, pll at 1.54, theese same settings are small fft stable at 450 fsb.
Last edited:
Looks good Gaidin, though I'm too tired to be able to tell whether it should be cooler or hotter with llc enabled. I think you'll get a hotter processor but cooler mosfets with it turned off, but I'll think on that one
I tend to be thorough (slow...) with overclocking. For example I know my reaper would do 5-5-5-15 at 1120mhz but would only do 4-4-4-12 up to 880mhz or so. Annoyed me that I couldn't get it to cl4 at 471 fsb. I reason that the cpu is the hardest one to clock, and the most important. So I learn as much as I can about the rest first, so that I can better judge when its a different component holding me back. Sound in theory, takes forever. I'm currently undervolting my i7 to learn the new bios, it'll do 2ghz at 0.8V as a dual core, loads at 45 with the stock cooler
If you sell a q6600 for a q9550 at 3.8, then sell that for an i7 920 which then only hits 3.8, you'll be so angry...
I tend to be thorough (slow...) with overclocking. For example I know my reaper would do 5-5-5-15 at 1120mhz but would only do 4-4-4-12 up to 880mhz or so. Annoyed me that I couldn't get it to cl4 at 471 fsb. I reason that the cpu is the hardest one to clock, and the most important. So I learn as much as I can about the rest first, so that I can better judge when its a different component holding me back. Sound in theory, takes forever. I'm currently undervolting my i7 to learn the new bios, it'll do 2ghz at 0.8V as a dual core, loads at 45 with the stock cooler
If you sell a q6600 for a q9550 at 3.8, then sell that for an i7 920 which then only hits 3.8, you'll be so angry...
Caporegime
Well thankfuly i sold the q6600 for £100.00, Id won a few quid on the horses and bought the q9550 of a mate for £140.00, nice low 1.200 vid chip, tbh i got bored with the q6600 and just wanted a new cpu to play with. More money than sense that time