Q9450 @ 3.7ghz

[Dopekitten]

Well I've gotten my Q9450 to 3.6ghz with a measly 1.275V (and LLC), and trying to get it to 3.7ghz within reasonable voltages (for me) is a bit more difficult.

I wouldn't like to go over 1.3 or 1.32ish because with LLC enabled that means my overall voltage is higher than if i didn't have LLC enabled, and thus i do not believe that the intel spec sheets took this into account. Also, here is an article on the "bad" side of LLC:
http://www.anandtech.com/cpuch...howdoc.aspx?i=3184&p=5

It's a bit confusing, but the gist of it says that the vdroop serves to make sure that the voltage on the CPU never goes above the set BIOS voltage ever. When you turn LLC on, the voltage spikes (resulting from changing CPU load) will go above the set bios level, and thus if i had 1.36 V in bios, i could be potentially getting 1.38V spikes (above intel's spec obviously)

So, in lieu of that, lets take a look at my CPU. I am able to have it prime95 (small ffts) stable for about an hour before it crashes with these settings.

Computer Specs:
Q9450 @ 3.6ghz
Asus P5Q-E
Cosair Dominator 1066 @ 927 w/ 5-5-5-15 (probably could loosen, but im gonna keep it this way until i get my overclock sorted out).
Xigmatek S1283 (with an old Dell fan which i solded the wires from into the intel stock fan's 4-pin connecter so i could get fan control) (max rpm is 4k, and 150cfm)
Vistontek HD4870 @ 840/1000
Seagate Barracuda 7200.11 500gb 32mb cache
P&PC Silencer Quad Red 750W
Antec 1200 Case

Then, Bios settings

CPU multi: 8
FSB: 463
PCI-E clock: 100
FSB strap to NB = Auto (changing it did nothing/decreased stability)
RAM clock: 927 @5-5-5-15

CPU voltage:1.3000
GTL 0/2 : .635
GTL 1/3: .670
PLL voltage: 1.54
VTT voltage: 1.28
DRAM voltage: 2.1 (it can run at 1.8, but im playing it safe once again, spec'd for 2.1 also)
NB voltage: 1.28
NB GTL voltage: Auto
SB voltage: Auto
PCI-E SATA voltage: Auto

LLC: Enabled

Evreything else on auto, i included the last three before LLC, becaus i believe i could possibly do something with thme (espically nb gtl and sb volt) to improve stability.

Anyway, any tips or suggestions on voltages/settings to change to improve stability would be greatly appreciated.

*Note* I know that increasing the voltage a bit might solve the problem, but i would rather stay away from that in lieu of that article i linked above (as i have LLC enabled)

 

[KingstonU]

I have a very similar build on the way and I would love to reach the clocks you are at. Seems the parts just can't get here fast enough.

 

[BonzaiDuck]

With Load Line Calibration disabled in BIOS, setting a CPU Voltage VID of 1.38750 resulted in a no-load voltage of about 1.34V and a full-load value of 1.28V. Enabling this feature and lowering the VID to 1.35000V produced a constant CPU supply voltage, regardless of load (or so it seemed), of 1.33V. Setting a lower VID resulted in a blue screen during Windows boot. Idle voltage was relatively unchanged at about 1.33-1.34V but the full-load voltage required increased by 50mV with no benefit. As you might guess, we recommend you leave this option disabled.

Why have you enabled LLC when the Anandtech article on over-clocking recommended disabling it?

I am also curious how you set the Gunning Transceiver Logic Reference Voltage. There is an article -- linked through Graysky's Guide (sticky) -- which deals with this in much more detail. The adjustments are, in turn, linked to your choice of CPU VTT voltage.

I can't speak with authority on the expectational limits of over-clocking your Q9450. In general, a Penryn quad-core will not OC to the same level as a a Penryn dual-core, just as a Conroe C2Q will not OC to the same level as a C2D.

But given the stock setting for the Q9450 of 2.67Ghz, I'm guessing that your expectations of a 3.6Ghz over-clock are a tad optimistic.

There is another article about two months earlier to the one you cite on over-clocking the QX9650, which deals with exponentially rising voltage requirements with increasing FSB.

 

[Gillbot]

I have a thread with my settings for 3.7GHz on my Q9400

 

[Dopekitten]

@ Bonzai Duck --> I know 3.7/3.8ghz are optimistic, but i was able to reach 3.6ghz stable with only 1.275 volts and LLC, so i think i can get 3.7ghz.

I have LLC enabled because although it spikes, i still get lower voltages overall, and its easier to set voltages.

@KingstonU --> are you changing the VTT/GTL voltages?

Also, Bonzai, i don't really have a set way to adjust GTL/VTT voltages, i have it at .635 instead of .630 becuase it appears to have a measurable stability difference (i.e. my core 1 kept failing, and so increasing the GTL would stop it from doing that, but two much would start to fail core 0). The same thing seems to apply to VTT

Thus, i've concluded the VTT/GTL should be used for correcting stability issues for particular cores. For example. I can boot a 4ghz pretty easily, like 1.32 volts or something, and obivously prime95 fails. But it only fails on core 0/1, i can run it (@4ghz, for over 24hours without failing on cores 2/3) So i've concluded that cores 0/1 are somewhat more defective, and i use GTL/VTT voltages to correct this.

I haven't looked at that sticky yet though, ill check it out.

@Gillbot --> nice, ill check it out
*Edit* Looked at your post, voltages of 1.4???? are you kidding, i don't know if you are running it 24/7 or if it was a 1 time run, but thats pretty far above intel's max spec voltage for 45nm processors, (1.3625)

 

[Idontcare]

Originally posted by: BonzaiDuck

With Load Line Calibration disabled in BIOS, setting a CPU Voltage VID of 1.38750 resulted in a no-load voltage of about 1.34V and a full-load value of 1.28V. Enabling this feature and lowering the VID to 1.35000V produced a constant CPU supply voltage, regardless of load (or so it seemed), of 1.33V. Setting a lower VID resulted in a blue screen during Windows boot. Idle voltage was relatively unchanged at about 1.33-1.34V but the full-load voltage required increased by 50mV with no benefit. As you might guess, we recommend you leave this option disabled.

Why have you enabled LLC when the Anandtech article on over-clocking recommended disabling it?

LLC is one of those multi-faceted tools, which can be optimized for providing benefits that outweigh the risks or if left unoptimized can provide so little benefits (or needlessly elevated risks) that it is decidedly undesirable.

I personally use LLC and found it to be of significant value in the manner I use it.

But I agree with Anandtech in guiding folks away from it in general as if you are prone to doing nothing more than enabling it in the BIOS and then walk away from it then yes you have very likely enabled an option in your BIOS that is going to be more detrimental to your CPU than had you left it disabled and proceeded to beat the shit out of your CPU with higher Vcc at idle to buffer against the inevitable Vdroop under load.

Kinda like how water-cooling, vapor-phase cooling, dry ice cooling, and LN2 cooling (in order of increasing difficulty) are generally not recommended wholesale for the masses as they can do more harm than good in the hands of a lot of naive individuals.

No different for pin mods, tape mods, etc. No one is going to be so foolish as to recommend them for everyone.

Also no different from lapping. Done correctly and there is substantial benefits to be had. Done incorrectly or poorly and the result can range from no benefits to even worse results had the lapping never been attempted.

Its of no surprise to me that Anandtech elected to formally recommend people not use LLC.

 

[Gillbot]

Originally posted by: Dopekitten
@Gillbot --> nice, ill check it out
*Edit* Looked at your post, voltages of 1.4???? are you kidding, i don't know if you are running it 24/7 or if it was a 1 time run, but thats pretty far above intel's max spec voltage for 45nm processors, (1.3625)

I run it there 24/7, it'll be fine.

 

[Dopekitten]

Heh, ok, whatever you say

 

[KingstonU]

No kidding I thought these Quads could get 3.6-3.7Ghz with less than 1.3 volts.

 

[Gillbot]

Originally posted by: Dopekitten
Heh, ok, whatever you say

Don't believe everything you read on the internet, unless it's on google. Google rules the world.

 

[BonzaiDuck]

Originally posted by: Idontcare

Originally posted by: BonzaiDuck

With Load Line Calibration disabled in BIOS, setting a CPU Voltage VID of 1.38750 resulted in a no-load voltage of about 1.34V and a full-load value of 1.28V. Enabling this feature and lowering the VID to 1.35000V produced a constant CPU supply voltage, regardless of load (or so it seemed), of 1.33V. Setting a lower VID resulted in a blue screen during Windows boot. Idle voltage was relatively unchanged at about 1.33-1.34V but the full-load voltage required increased by 50mV with no benefit. As you might guess, we recommend you leave this option disabled.

Why have you enabled LLC when the Anandtech article on over-clocking recommended disabling it?

LLC is one of those multi-faceted tools, which can be optimized for providing benefits that outweigh the risks or if left unoptimized can provide so little benefits (or needlessly elevated risks) that it is decidedly undesirable.

I personally use LLC and found it to be of significant value in the manner I use it.

But I agree with Anandtech in guiding folks away from it in general as if you are prone to doing nothing more than enabling it in the BIOS and then walk away from it then yes you have very likely enabled an option in your BIOS that is going to be more detrimental to your CPU than had you left it disabled and proceeded to beat the shit out of your CPU with higher Vcc at idle to buffer against the inevitable Vdroop under load.

Kinda like how water-cooling, vapor-phase cooling, dry ice cooling, and LN2 cooling (in order of increasing difficulty) are generally not recommended wholesale for the masses as they can do more harm than good in the hands of a lot of naive individuals.

No different for pin mods, tape mods, etc. No one is going to be so foolish as to recommend them for everyone.

Also no different from lapping. Done correctly and there is substantial benefits to be had. Done incorrectly or poorly and the result can range from no benefits to even worse results had the lapping never been attempted.

Its of no surprise to me that Anandtech elected to formally recommend people not use LLC.

I'll accept what you're saying here. I'm still using nForce boards after our mutual experiences with the Striker 680i. I don't have LLC on my 780i board. But the white-paper explanation in the cited article actually clarifies more about the variation in voltage between idle and load. This would have seemed evident from earlier-gen systems monitored with either SpeedFan or Asus Probe, in that the voltage variations at the two points of interest are evident through real-time graphic representation.

It seems to me that with the LLC, a user attenuates the offset between a set voltage and the equilibritum idle, but the spikes occurring when returning to that equilibrium would exceed either. The remaining question would be whether those spikes are detrimental to the CPU or even the system's stability over some part of the voltage range. Here, I'm speculating that the higher you push both the actual CPU voltage, the closer you come to making the transitional spikes a potential problem. Or -- there may be more pronounced transitional variation the higher the voltage setting.

But it wouldn't seem necessary to have the feature for any better reason than to align set values with actual static idle values. There would still be a voltage droop.

In that regard -- correct me if I'm wrong -- but the OP cites a "safe" threshold at 1.36+V, and I think this corresponds to what we know about both Conroe and Penryn. Even if one were to set the voltage at -- say -- 1.4V (insinuated by Gilbot), with LLC disabled, the idle equilibrium would be very close to that self-imposed limit.

From what I can see, the OP has a lot of room to bump up the voltage to make his 3.7Ghz work, unless his target is in the range where voltage requirement increases dramatically for a few extra megahertz.

 

[Gillbot]

Originally posted by: BonzaiDuck

Originally posted by: Idontcare

Originally posted by: BonzaiDuck

With Load Line Calibration disabled in BIOS, setting a CPU Voltage VID of 1.38750 resulted in a no-load voltage of about 1.34V and a full-load value of 1.28V. Enabling this feature and lowering the VID to 1.35000V produced a constant CPU supply voltage, regardless of load (or so it seemed), of 1.33V. Setting a lower VID resulted in a blue screen during Windows boot. Idle voltage was relatively unchanged at about 1.33-1.34V but the full-load voltage required increased by 50mV with no benefit. As you might guess, we recommend you leave this option disabled.

Why have you enabled LLC when the Anandtech article on over-clocking recommended disabling it?

LLC is one of those multi-faceted tools, which can be optimized for providing benefits that outweigh the risks or if left unoptimized can provide so little benefits (or needlessly elevated risks) that it is decidedly undesirable.

I personally use LLC and found it to be of significant value in the manner I use it.

But I agree with Anandtech in guiding folks away from it in general as if you are prone to doing nothing more than enabling it in the BIOS and then walk away from it then yes you have very likely enabled an option in your BIOS that is going to be more detrimental to your CPU than had you left it disabled and proceeded to beat the shit out of your CPU with higher Vcc at idle to buffer against the inevitable Vdroop under load.

Kinda like how water-cooling, vapor-phase cooling, dry ice cooling, and LN2 cooling (in order of increasing difficulty) are generally not recommended wholesale for the masses as they can do more harm than good in the hands of a lot of naive individuals.

No different for pin mods, tape mods, etc. No one is going to be so foolish as to recommend them for everyone.

Also no different from lapping. Done correctly and there is substantial benefits to be had. Done incorrectly or poorly and the result can range from no benefits to even worse results had the lapping never been attempted.

Its of no surprise to me that Anandtech elected to formally recommend people not use LLC.

I'll accept what you're saying here. I'm still using nForce boards after our mutual experiences with the Striker 680i. I don't have LLC on my 780i board. But the white-paper explanation in the cited article actually clarifies more about the variation in voltage between idle and load. This would have seemed evident from earlier-gen systems monitored with either SpeedFan or Asus Probe, in that the voltage variations at the two points of interest are evident through real-time graphic representation.

It seems to me that with the LLC, a user attenuates the offset between a set voltage and the equilibritum idle, but the spikes occurring when returning to that equilibrium would exceed either. The remaining question would be whether those spikes are detrimental to the CPU or even the system's stability over some part of the voltage range. Here, I'm speculating that the higher you push both the actual CPU voltage, the closer you come to making the transitional spikes a potential problem. Or -- there may be more pronounced transitional variation the higher the voltage setting.

But it wouldn't seem necessary to have the feature for any better reason than to align set values with actual static idle values. There would still be a voltage droop.

In that regard -- correct me if I'm wrong -- but the OP cites a "safe" threshold at 1.36+V, and I think this corresponds to what we know about both Conroe and Penryn. Even if one were to set the voltage at -- say -- 1.4V (insinuated by Gilbot), with LLC disabled, the idle equilibrium would be very close to that self-imposed limit.

From what I can see, the OP has a lot of room to bump up the voltage to make his 3.7Ghz work, unless his target is in the range where voltage requirement increases dramatically for a few extra megahertz.

I never once claimed that 1.4v was a "safe" voltage for the average Joe to OC their chip at. My point mainly insinuates that most OC'ers want a generic set of numbers to "plug and chug" instead of doing the legwork and tweaking like many of us have already done. EVERY setup is different and while asking for advice is a good place to start, you need to read all of the guides and skim through everyone's settings and adjust specifically for your system. Otherwise, you are walking a thin line between a safe OC and disaster no matter what settings you use, unless you leave it STOCK.

 

[Big Lar]

OP, Try adjusting your GTL/S by using Prime Blend. I too started at .630 and.670 and used Coretemp and Prime to nail down my gtl's. It helps alot.

Larry

 

[Dopekitten]

Well, i upped the NB voltage to 1.36 and my stability incrased from 1 hr 12mins to 2 hours 20 mins. Not really sure what that does.
I have notices that the core that always fails is Core 1. Tweaking GTL's/VTT does increase stability but as i said (or maybe not) in my OP increasing one or another to much results in stability failure in Core 0, so there seems to be a fine line there, and because my motherboard only allows voltage tweaks of .02 (for VTT) and .005 for GTL, i can get to a point where setting it at like .640 = stability for Core 1 but instabilty for Core 0, and if its at .635 Core 1 becomes instable and Core 0 is stable.

Sometimes i can use VTT voltage to compromise for this inability have smaller quantized tweaks, however, its kinda finickly and doesn't always work.

Also, the reason i use LLC is becuase without it, the idle voltage is higher than load voltage, whereas with LLC, the idle is below the load, so thus (as i never have 100% load all the time) the voltage most of the time is lower. This is the benefit i see of LLC (directed towards Gillbot and BonzaiDuck). If you see some reason why this is stupid (could be, i only skimmed that vdroop article, it was far to scientific for me to bother reading it), please explain. However as far as i see, using LLC, whether or not it has any benefits doesn't matter as long as you take precautionary measure (as i have done, by limiting my BIOS voltage)

Anyway, i've come up with a question
Should i tweak the SB voltage and/or NB GTL voltages. What does this do for stability, shoudl it be changed? What are the maxes that i can use such that the degradation of CPU/NB/SB doesn't affect them within between 3 years to 5+ (optimistic i know, but just wondering)

Also, can anybody determine whether my stability issues are due to FSB instability, NB overheating, or actual CPU instability?

With 1.36+ volts on the NB, will i need cooling? I mean, i have an antec 1200, so i dunno, would it be a problem. (On a side note, MOSFETs are not a problem, merely warm to the touch under 100% load, because i the Xigmatek spoiler + the massive fan on the Antec seem to cool them pretty well.)

Would flashing the BIOS to a newer version be a smart idea at this point. I.E. i don't need to, no problems with my mobo, but would it improve stability/overclocking ability?

Anyway that's it for now, thanks again.

 

[GundamF91]

What's your temp like at full load? You should also try to use Intel Linpack for fully stressing your system. I would hit 3.7Ghz no problem, but the temp was too high for my liking. Even at 3.6Ghz, I was hitting above 70c, which I'd try to stay under.

 

[Dopekitten]

My temps at 100% load with ~30% fan sped are about 62 62 60 60

Temps are not a problem for me

 

[Gillbot]

One thing you should be cautious of, if you want 100% accurate voltage readings you should measure them with a volt meter. Software isn't always 100% accurate.

I won't comment on LLC as my motherboard doesn't support it. I will say, After doing forther tweaking of my VTT and a few other voltages, I was able to drop my vcore voltage under 1.4v and maintain stability. Most people aren't aware (in my experience anyway) is that the 45nm cpu's are VERY picky with a myriad of voltage settings. One minor adjustment on one may take major adjustments on another to maintain stability. It's a very big and complex balancing game with voltages.

 

[Big Lar]

Q9400 2x2gb 1402 bios Settings at 8x480 09/19/08

FSB Strap = 333
PCI-E = 101
Dram Freq. = 961
5,5,4,15,3,52,6,3
8,3,5,4,6,4,6
13,5,1,5,5
Static = Disabled
Training = Disabled
OC Charge = Auto
Ai Clock = Medium
Ai Booster = 10
All Enabled
1.33750v
0.670v 0/2
0.675v 1/3
1.56v
1.28v
2.02v
1.34v
0.640
1.20v
1.60v
LLC = Disabled
D,D,A,A

Thats my settings, kinda pick at those if you please.
Larry

 

[Dopekitten]

Ill look and see what i can do with those Big Lar, thanks
I saw that you set your PCI-E freq to 101, why?

 

[Big Lar]

Originally posted by: Dopekitten
Ill look and see what i can do with those Big Lar, thanks
I saw that you set your PCI-E freq to 101, why?

I run not only a High voltage VGA card but Also a SCSI RAID card off the pci-e, not to mention the NIC is off the same volts, etc, and of course, sorta common thing NOT to run any voltages on Auto with these boards/ they tend to Overvolt if on Auto.

Larry

 

[GundamF91]

Originally posted by: Dopekitten
My temps at 100% load with ~30% fan sped are about 62 62 60 60

Temps are not a problem for me

Did you use OCCT or Prime, etc? You should try the Intel Linpack to really stress it. It'll easily make it run 10c hotter. If you haven't tested with Linpack, you haven't fully stressed your system.
http://www.computer-juice.com/...ss-test-program-19010/

That said, personally I don't feel the need to hit 3.7 or 3.8...because it's really not that much of a difference performance wise, but it does heat up your computer quite a bit.

 

[Idontcare]

Originally posted by: Big Lar

Originally posted by: Dopekitten
Ill look and see what i can do with those Big Lar, thanks
I saw that you set your PCI-E freq to 101, why?

I run not only a High voltage VGA card but Also a SCSI RAID card off the pci-e, not to mention the NIC is off the same volts, etc, and of course, sorta common thing NOT to run any voltages on Auto with these boards/ they tend to Overvolt if on Auto.

Larry

I suspect the question was more motivated by "why 101 (not stock) instead of 100 (which would be stock)"?

Most folks around here would recommend you not overclock your PCI-E. Taking it off auto and locking it to 100MHz is the recommend. Not sure why you'd set it to 101.

 

[Gillbot]

Originally posted by: Idontcare

Originally posted by: Big Lar

Originally posted by: Dopekitten
Ill look and see what i can do with those Big Lar, thanks
I saw that you set your PCI-E freq to 101, why?

I run not only a High voltage VGA card but Also a SCSI RAID card off the pci-e, not to mention the NIC is off the same volts, etc, and of course, sorta common thing NOT to run any voltages on Auto with these boards/ they tend to Overvolt if on Auto.

Larry

I suspect the question was more motivated by "why 101 (not stock) instead of 100 (which would be stock)"?

Most folks around here would recommend you not overclock your PCI-E. Taking it off auto and locking it to 100MHz is the recommend. Not sure why you'd set it to 101.

I have found better stability at 101 myself as opposed to 100. I have no idea why but I usually set everything manually as opposed to auto. In my tweaks, I found that 101 was more stable than 100 on a variety of manufacturers products.

 

[Dopekitten]

Hmm ill try and see if that has any effect. Do you have any articles by the way, on that topic (i.e. 101 freq vs 100)

 

[n7]

PCI-e @ 101 is an old trick that was used when some older boards didn't lock it properly @ 100.

It's so common (though usually totally unnecessary) that even i do that...it's just a habit for many i think.

On newer boards, there is no reason there should be any difference between 99 or 100 or 101...

As for the OP...FSB limitations

This is why low multis suck.

His chip obviously has a lot more in it...but it needs a higher multi to get there.