Noctua upgrade, i7 960 ocing

[wand3r3r]

I just picked up a Noctua NH-D14 in the hopes of a bit cooler system, an upgrade from the Noctua NH-C12P.

My system consists of an Asus P6T SE, Core i7 960, 6GB OCZ gold 1600MHz, GTX 560 ti Hawk and now the Noctua NH-D14 sitting in an Antec Sonata II. I have a Noctua fan blowing out of the case in the rear, replacing the Antec fan, as well as an additional noctua fan inside the case next to the bottom near the HDD's blowing towards the GPU and the rear of the case.

I had previously had a Core i7 920 D0 which I had overclocked to 3.50 GHz (3.66 w/ turbo) by only adjusting:
vCore 1.2v, BCLK 166 (iirc), Ram timings and the rest auto.

The Core i7 960 however seems to need 1.23v vCore to even reach 150MHz BCLK, with QPI/DRAM 1.3v (3.6GHz / 3.75 with Turbo). If I lower the vCore or the QPI/DRAM it gets a BSOD pretty quickly.

I have a couple thoughts, firstly the 920 almost offhand appears to have had more OC potential, at least it required less vCore and the temperatures were under 80. Although, I am not sure if having the QPI/DRAM on auto had an effect and the manual 1.3 is limiting my current OC, or if this 960 is just that much hotter.

At load with NH-D14 I am getting about the same temperatures as I was the the C12P, which makes me wonder if the C12P is that good, or if there is a flaw with my D14 installation (too little paste etc.).

Core temp reports a max of 84 C in 2 cores and 79 in the other 2. That is about the same as I got with the C12P.

I have a few questions regarding this setup and cooling. Does this seem like I have something wrong, or was is just a waste to get a NH-D14? I put paste on the CPU pretty evenly and not too thick, probably a pea sized amount or so. Should I manually adjust a setting in the CPU overclock, could something on auto be at higher voltages then needed and making the 960 warmer then it could be? Do you have any suggestions on OC settings, if you have a similar setup or are more familiar with OCing any input would be nice.

 

[RussianSensation]

84*C at only 1.3V and 3.7ghz?

Put it this way, my setup runs at 65-67*C at 100% load on all 8 threads in LinX/Prime95 with a Megahalems and 1 x 1600 rpm fan (a far inferior setup to the NH-D14). My CPU is at 3.9ghz @ 1.312V, QPI 1.21V.

I would say the main reasons behind your problems are:

1) OCZ Gold ram is poor. It is no wonder OCZ existed the RAM market after DDR3. Their DDR3 ram couldn't run stock speeds stably. You shouldn't even have to touch the QPI voltage at all for DDR3-1600 stock speeds @ 1.65V. I was constantly getting BSOD even at stock settings with that RAM and returned it for G.Skill. I was able to run QPI at only 1.21V at 3.9ghz overclock.

2) Antec Sonata 2 has horrible case air flow. This alone for sure adds 10*C to your temperatures. The PSU is all in the wrong place, there is no large intake fans, there is no top exhaust fan.

I would say go out and get an Antec 300 case for $50.

Your cheapest alternative is to try to reseat the heatsink and see if you have a good thermal paste imprint/spread when you take the heatsink off.

 

[cubeless]

those temps seem a bit high with a 14... the 14 is sometimes tough to get seated properly due to it's hugeness... and you need t have enough airflow out of the case to be sure you are using its entire capacity...

i have antec 300; i7 930; dh14 aiming up into top case fan;1.264 w/droop @ 4.1ghz; ambient is 77f...

my hottest core running occt linpack is 72c...

 

[Idontcare]

84*C at only 1.3V and 3.7ghz?

Put it this way, my setup runs at 65-67*C at 100% load on all 8 threads in LinX/Prime95 with a Megahalems and 1 x 1600 rpm fan (a far inferior setup to the NH-D14). My CPU is at 3.9ghz @ 1.312V, QPI 1.21V.

I would say the main reasons behind your problems are:

1) OCZ Gold ram is poor. It is no wonder OCZ existed the RAM market after DDR3. Their DDR3 ram couldn't run stock speeds stably. You shouldn't even have to touch the QPI voltage at all for DDR3-1600 stock speeds @ 1.65V. I was constantly getting BSOD even at stock settings with that RAM and returned it for G.Skill. I was able to run QPI at only 1.21V at 3.9ghz overclock.

2) Antec Sonata 2 has horrible case air flow. This alone for sure adds 10*C to your temperatures. The PSU is all in the wrong place, there is no large intake fans, there is no top exhaust fan.

I would say go out and get an Antec 300 case for $50.

Your cheapest alternative is to try to reseat the heatsink and see if you have a good thermal paste imprint/spread when you take the heatsink off.

To verify the magnitude of the impact of your case on cooling, take the side off the case and setup a room fan to blow good air towards the case.

(just for a test, its easier than removing your entire setup from the case just to test this)

That said, the Antec case Russian linked, $50 after rebate and free shipping is a killer of deal! I'd buy it if I had an excuse to buy it.

 

[Spikesoldier]

I just picked up a Noctua NH-D14 in the hopes of a bit cooler system, an upgrade from the Noctua NH-C12P.

My system consists of an Asus P6T SE, Core i7 960, 6GB OCZ gold 1600MHz, GTX 560 ti Hawk and now the Noctua NH-D14 sitting in an Antec Sonata II. I have a Noctua fan blowing out of the case in the rear, replacing the Antec fan, as well as an additional noctua fan inside the case next to the bottom near the HDD's blowing towards the GPU and the rear of the case.

I had previously had a Core i7 920 D0 which I had overclocked to 3.50 GHz (3.66 w/ turbo) by only adjusting:
vCore 1.2v, BCLK 166 (iirc), Ram timings and the rest auto.

The Core i7 960 however seems to need 1.23v vCore to even reach 150MHz BCLK, with QPI/DRAM 1.3v (3.6GHz / 3.75 with Turbo). If I lower the vCore or the QPI/DRAM it gets a BSOD pretty quickly.

I have a couple thoughts, firstly the 920 almost offhand appears to have had more OC potential, at least it required less vCore and the temperatures were under 80. Although, I am not sure if having the QPI/DRAM on auto had an effect and the manual 1.3 is limiting my current OC, or if this 960 is just that much hotter.

At load with NH-D14 I am getting about the same temperatures as I was the the C12P, which makes me wonder if the C12P is that good, or if there is a flaw with my D14 installation (too little paste etc.).

Core temp reports a max of 84 C in 2 cores and 79 in the other 2. That is about the same as I got with the C12P.

I have a few questions regarding this setup and cooling. Does this seem like I have something wrong, or was is just a waste to get a NH-D14? I put paste on the CPU pretty evenly and not too thick, probably a pea sized amount or so. Should I manually adjust a setting in the CPU overclock, could something on auto be at higher voltages then needed and making the 960 warmer then it could be? Do you have any suggestions on OC settings, if you have a similar setup or are more familiar with OCing any input would be nice.

the 920 you had comes in both C0 and D0 steppings. you stated yours was a D0.

the 960 was one of the first i7's to launch and thus was only available in the C0 stepping.

while it has a higher model number, the 920 will be the better OC'er because of the refined D0 stepping. i would attempt to use the 920 if you still possess it.

im also questioning the move from the 12P to the D14. while the D14 is a formidable cooler, i dont think it provides earth-shattering performance over the 12P.

 

[Suspicious-Teach8788]

uh some RAM chips and CPUs require higher QPI. A lot of people say 1.35, but you can see people on Corsair, GSkill, etc testifying that you may need to push beyond 1.55 even to 1.6 to run stably. Granted thats for DDR-2000 with the Elpida chips or whatever I think, but I know that even 1600 requires a little push from my end.

 

[wand3r3r]

Thanks for the thoughts.

84*C at only 1.3V and 3.7ghz?

Put it this way, my setup runs at 65-67*C at 100% load on all 8 threads in LinX/Prime95 with a Megahalems and 1 x 1600 rpm fan (a far inferior setup to the NH-D14). My CPU is at 3.9ghz @ 1.312V, QPI 1.21V.

I would say the main reasons behind your problems are:

1) OCZ Gold ram is poor. It is no wonder OCZ existed the RAM market after DDR3. Their DDR3 ram couldn't run stock speeds stably. You shouldn't even have to touch the QPI voltage at all for DDR3-1600 stock speeds @ 1.65V. I was constantly getting BSOD even at stock settings with that RAM and returned it for G.Skill. I was able to run QPI at only 1.21V at 3.9ghz overclock.

2) Antec Sonata 2 has horrible case air flow. This alone for sure adds 10*C to your temperatures. The PSU is all in the wrong place, there is no large intake fans, there is no top exhaust fan.

I would say go out and get an Antec 300 case for $50.

Your cheapest alternative is to try to reseat the heatsink and see if you have a good thermal paste imprint/spread when you take the heatsink off.

Your setup's considerably cooler. Have you set any other settings manually (which can have an effect on the temperatures), I'm wondering if I have to set additional settings to limit the auto increases.I guess that's also why OCZ quit making ram. I'll have to pick up a different set when I upgrade. I'll have to keep the case upgrade in mind but it'll be a while before I can do any further updating.

I'll take the heatsink off and see what it looks like, it's good to have an example.

those temps seem a bit high with a 14... the 14 is sometimes tough to get seated properly due to it's hugeness... and you need t have enough airflow out of the case to be sure you are using its entire capacity...

i have antec 300; i7 930; dh14 aiming up into top case fan;1.264 w/droop @ 4.1ghz; ambient is 77f...

my hottest core running occt linpack is 72c...

The D14 is blowing from the inside of the case towards the outside and has a case fan an inch or two from it blowing the air outside.

To verify the magnitude of the impact of your case on cooling, take the side off the case and setup a room fan to blow good air towards the case.

(just for a test, its easier than removing your entire setup from the case just to test this)

That said, the Antec case Russian linked, $50 after rebate and free shipping is a killer of deal! I'd buy it if I had an excuse to buy it.

That's an interesting idea, with the case open the temperatures are about (max) 77 deg. C. That dropped it from ~81C before I opened it. With a fan directly on it, it varies from 74-77 C.

the 920 you had comes in both C0 and D0 steppings. you stated yours was a D0.

the 960 was one of the first i7's to launch and thus was only available in the C0 stepping.

while it has a higher model number, the 920 will be the better OC'er because of the refined D0 stepping. i would attempt to use the 920 if you still possess it.

im also questioning the move from the 12P to the D14. while the D14 is a formidable cooler, i dont think it provides earth-shattering performance over the 12P.

That would have been good to know beforehand, I just got rid of the 920 D0 this past weekend. I assumed that a recently purchased 960 (summer '11) would certainly be a D0. [link]http://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors[/link] all-knowing and never incorrect wikipedia says the 960 is D0 or at least the SLBEU which this one is.
Batch# 3050A366


Processor>
http://imageshack.us/photo/my-images/683/img0929k.jpg/

Case/setup>
http://imageshack.us/photo/my-images/838/img0994tm.jpg/

 

[wand3r3r]

Now I've just been dropping the QPI/DRAM while running windows via the TurboV program, I've got to 1.256v and it's been running Prime95 for almost 2 hours stable. The last time I tried to drop the QPI/DRAM I may have also dropped the vCore slightly (thus the BSOD?). I will probably drop the QPI/DRAM a bit more until it starts BSODing and then put it a bit higher into the BIOS itself and test again. I've read how unreliable the software OCing is but I just wanted to try get in the neighborhood (dropping it on the fly) and then test it in the BIOS with a bit higher then the next BSOD. Having the case open has kept the temperature consistently at 77C on the hottest core. It was hovering around 81 and fluctuating from 79-82 before opening it up.

 

[Spikesoldier]

Thanks for the thoughts.



Your setup's considerably cooler. Have you set any other settings manually (which can have an effect on the temperatures), I'm wondering if I have to set additional settings to limit the auto increases.I guess that's also why OCZ quit making ram. I'll have to pick up a different set when I upgrade. I'll have to keep the case upgrade in mind but it'll be a while before I can do any further updating.

I'll take the heatsink off and see what it looks like, it's good to have an example.




The D14 is blowing from the inside of the case towards the outside and has a case fan an inch or two from it blowing the air outside.





That's an interesting idea, with the case open the temperatures are about (max) 77 deg. C. That dropped it from ~81C before I opened it. With a fan directly on it, it varies from 74-77 C.



That would have been good to know beforehand, I just got rid of the 920 D0 this past weekend. I assumed that a recently purchased 960 (summer '11) would certainly be a D0. [link]http://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors[/link] all-knowing and never incorrect wikipedia says the 960 is D0 or at least the SLBEU which this one is.
Batch# 3050A366


Processor>
http://imageshack.us/photo/my-images/683/img0929k.jpg/

Case/setup>
http://imageshack.us/photo/my-images/838/img0994tm.jpg/

sounds like you have a D0 and I am mistaken. i wouldnt be afraid to raise vtt to 1.35 and vdimm to 1.65.

 

[RussianSensation]

Now I've just been dropping the QPI/DRAM while running windows via the TurboV program, I've got to 1.256v and it's been running Prime95 for almost 2 hours stable. The last time I tried to drop the QPI/DRAM I may have also dropped the vCore slightly (thus the BSOD?). I will probably drop the QPI/DRAM a bit more until it starts BSODing and then put it a bit higher into the BIOS itself and test again. I've read how unreliable the software OCing is but I just wanted to try get in the neighborhood (dropping it on the fly) and then test it in the BIOS with a bit higher then the next BSOD. Having the case open has kept the temperature consistently at 77C on the hottest core. It was hovering around 81 and fluctuating from 79-82 before opening it up.

:thumbsup:

At QPI of 1.256V (if stable), try dropping the DRAM voltage to 1.60V if you can and relax the timings to CL9-9-9. CL timings on the RAM shouldn't affect performance of i7 too much and lower DRAM voltage is going to put less stress on the CPU's memory controller.

BTW, I think there is a way for you to access 1 more bin (i.e., 25x multiplier). Xbitlabs was able to run Core i7 950 at 24x multiplier (i.e., 1 higher than stock) by leaving Turbo on. They needed CPU Vcore 1.2 / QPI voltage of 1.2 to reach 3.8ghz. So I am thinking your CPU voltage of 1.23V is pretty reasonable for a 3.7ghz overclock.

 

[wand3r3r]

sounds like you have a D0 and I am mistaken. i wouldnt be afraid to raise vtt to 1.35 and vdimm to 1.65.

Well, who knows if wikipedia is correct. I would assume that if they do produce D0's that I would have one due to the fact I just bought it this summer.

:thumbsup:

At QPI of 1.256V (if stable), try dropping the DRAM voltage to 1.60V if you can and relax the timings to CL9-9-9. CL timings on the RAM shouldn't affect performance of i7 too much and lower DRAM voltage is going to put less stress on the CPU's memory controller.

BTW, I think there is a way for you to access 1 more bin (i.e., 25x multiplier). Xbitlabs was able to run Core i7 950 at 24x multiplier (i.e., 1 higher than stock) by leaving Turbo on. They needed CPU Vcore 1.2 / QPI voltage of 1.2 to reach 3.8ghz. So I am thinking your CPU voltage of 1.23V is pretty reasonable for a 3.7ghz overclock.

Thanks, I'll drop the dram voltage and timings and see what it does for temperatures. I believe the 1.23v vCore is the minimum because at least when I was adjusting it down in the slightest it caused BSODs within minutes.

I actually left turbo on and it does run at a 150 x 25 => 3.75GHz. Otherwise it's 3.6 GHz without turbo but turbo is always active when busy.

 

[cubeless]

or get a 300 with a 430w psu no rebate for $60 and sell the psu...

http://www.newegg.com/Special/ShellS...8-_-08102011_2

 

[wand3r3r]

or get a 300 with a 430w psu no rebate for $60 and sell the psu...

http://www.newegg.com/Special/ShellS...8-_-08102011_2

That's a good deal, although I just bought the 650 TX last week.

I found the culprit (I presume), take a look!!!!
cooler>
http://imageshack.us/photo/my-images/837/img0995uo.jpg/
cpu paste>
http://imageshack.us/photo/my-images/577/img0999v.jpg/

What's the best idea to solve the lack of contact? It's like an hourglass or tornado shape which is making contact but 1/2 of the CPU isn't even contacting the cooler!

Perhaps that helps explain why at times 2 cores were 77 and the other 2 could be as low as 70 C. A 5-7 deg C difference seems like a huge difference now.

Apparently I should sand it? What grit is best, or just as high as it gets? Is the technique presumably to sand the parts with contact until the whole cpu is getting even contact?

 

[bryanW1995]

84*C at only 1.3V and 3.7ghz?

Put it this way, my setup runs at 65-67*C at 100% load on all 8 threads in LinX/Prime95 with a Megahalems and 1 x 1600 rpm fan (a far inferior setup to the NH-D14). My CPU is at 3.9ghz @ 1.312V, QPI 1.21V.

I would say the main reasons behind your problems are:

1) OCZ Gold ram is ABSOLUTELY HORRIBLE. It is no wonder OCZ existed the RAM market after DDR3. Their DDR3 ram couldn't run stock speeds stably. You shouldn't even have to touch the QPI voltage at all for DDR3-1600 stock speeds @ 1.65V. I was constantly getting BSOD even at stock settings with that RAM and returned it for G.Skill. I was able to run QPI at only 1.21V at 3.9ghz overclock.

ftfy. I bought 3x2gb ocz gold on BF 2009 when I built my i7 rig, I jacked around with it for a year before finally breaking down and getting a set that actually works. I won't even look at other OCZ products these days b/c their ram was so horribad.

To verify the magnitude of the impact of your case on cooling, take the side off the case and setup a room fan to blow good air towards the case.

(just for a test, its easier than removing your entire setup from the case just to test this)

That said, the Antec case Russian linked, $50 after rebate and free shipping is a killer of deal! I'd buy it if I had an excuse to buy it.

I agree, my antec 300 has an oc'd 1055t + a gtx 260 running seti 24/7 in my office. Airflow is excellent, and the design is suitable for a work environment as well.

That's a good deal, although I just bought the 650 TX last week.

I found the culprit (I presume), take a look!!!!
cooler>
http://imageshack.us/photo/my-images/837/img0995uo.jpg/
cpu paste>
http://imageshack.us/photo/my-images/577/img0999v.jpg/

What's the best idea to solve the lack of contact? It's like an hourglass or tornado shape which is making contact but 1/2 of the CPU isn't even contacting the cooler!

Perhaps that helps explain why at times 2 cores were 77 and the other 2 could be as low as 70 C. A 5-7 deg C difference seems like a huge difference now.

Apparently I should sand it? What grit is best, or just as high as it gets? Is the technique presumably to sand the parts with contact until the whole cpu is getting even contact?

Ah, good times. I sanded down a few cpu coolers until it started to become too much like work instead of just a fun hobby for me. Yes, you should try to sand the contact patch on the hsf down to point that it is making good contact across the cpu. When I did it I started off with coarser paper like 200 grit for the rough work, then went all the way to 1000 grit for the finish.

 

[RussianSensation]

What's the best idea to solve the lack of contact? It's like an hourglass or tornado shape which is making contact but 1/2 of the CPU isn't even contacting the cooler!

Nice! Before you sand it, there 2 things you should try:

1) Clean off the CPU and heatsink and reapply thermal paste. Instead of the pea dot method, which clearly didn't work, put a vertical line across the heatsink spreader. You can try spreading the paste with a card (but that gets tricky). So perhaps use a ziplock bag to spread it with your finger.

2) Reposition the heatsink by turning the cooler by 90 degrees this time. Xbitlabs had this similar issue and achieved a better contact and produced a better imprint by simply rotating the heatsink. This may help.

 

[Spikesoldier]

Well, who knows if wikipedia is correct. I would assume that if they do produce D0's that I would have one due to the fact I just bought it this summer.

yeah and running cpuz will give you a definitive answer. the stepping you listed corresponds to the D0 though.

i think when you turned down the vcore and got the bsod is where i got my assumption of lackluster oc came from.

 

[wand3r3r]

Nice! Before you sand it, there 2 things you should try:

1) Clean off the CPU and heatsink and reapply thermal paste. Instead of the pea dot method, which clearly didn't work, put a vertical line across the heatsink spreader. You can try spreading the paste with a card (but that gets tricky). So perhaps use a ziplock bag to spread it with your finger.

2) Reposition the heatsink by turning the cooler by 90 degrees this time. Xbitlabs had this similar issue and achieved a better contact and produced a better imprint by simply rotating the heatsink. This may help.

Thanks for the tips.

For 1.) I already dabbed the paste around with my finger in a plastic bag to get it (quite) evenly spread. Then I put it (D14) on ensuring that I only went one twist at a time on each side to ensure it went down evenly. That's an excellent way to spread it out pretty evenly and thin. Just thick enough to not see through it is about how thick I put it. The result was basically the same.

As for 2.) The brackets only allow a 180 degree spin, I will need to take out the motherboard to try it (to refit the cpu fan connections), do you think it's still worth testing?
*edit After taking a look at their results I probably should test it. The only problem then would be that the cooler will be blowing into the PSU or alternatively into the wind blocking video card below. They had a similar situation, albeit not as extreme.

See case pic from above>
http://imageshack.us/photo/my-images/838/img0994tm.jpg/

 

[wand3r3r]

yeah and running cpuz will give you a definitive answer. the stepping you listed corresponds to the D0 though.

i think when you turned down the vcore and got the bsod is where i got my assumption of lackluster oc came from.

lol, lackluster to some. I'm more concerned about the heat in the end, well and bang for the buck / performance gains. At some point the real gains diminish. I'd like to keep it under 70C. That was my attempt to reduce heat.

Oh yeah, and CoreTemp reports D0.

 

[bryanW1995]

Low end voltage requirements quite often don't have much correlation with oc headroom on cpus. Your extremely high temps at the relatively low voltage/clock level is of much more concern atm. Will your already poor case airflow you're probably going to need to keep the air exhausting out of the case. Looks like there is going to be some sandpaper in your future!

 

[Idontcare]

When you start to lap the HSF and IHS what you want to do is use a permanent market and draw reference lines on the surface, I did a star pattern here:

Then inspect the removal rate of the marks as you polish, if you have an uneven surface (and you will at first) then the marks will not go away evenly:

Once you have removed all visible signs of the original marks then you remark it and go up one step in grit number (200->400->800->1000->2000):

When you can put the marks on the IHS or HSF surface and they come off uniformly as you polish the surface that is when you'll know you've sanded/lapped/polished all you have to at that grit level.

The above shots are from my tuniq120, but I'm sure I'll be lapping my D14 soon enough

 

[RussianSensation]

The above shots are from my tuniq120, but I'm sure I'll be lapping my D14 soon enough

The base on the TT120 wasn't smooth at all given that it was a premium heatsink, but at least it was pretty flat. The base on the D-14 in contrast should be nearly perfect.

He should use a ruler to see if the problem is his heatsink OR the heat spreader on the CPU. It would be a shame to lap the heatsink only to realize that it was the CPU heatspreader that was responsible haha.

 

[wand3r3r]

It's very difficult to see which has the uneven surface with a clear plastic ruler. I thought the difference would be more visible. I'll probably need to buy something with a perfectly flat surface and thin enough to see light leaking between it and the surface?

Edit** it appears to be the processor heat spreader. Does that mean I need to sand it instead? Is there anything special regarding sanding processors?

I assume it voids the warranty?

 

[Idontcare]

It's very difficult to see which has the uneven surface with a clear plastic ruler. I thought the difference would be more visible. I'll probably need to buy something with a perfectly flat surface and thin enough to see light leaking between it and the surface?

Just cover the surface with a paste, use toothpaste if you don't want to waste your TIM, and scrape it off with one straight-direction swipe of your ruler.

The low lying spots will still be covered with paste.

You need to do this twice (paste, swipe, inspect, repeat), at least, going up-to-down and then rotating 90 degrees and doing it again to ensure you capture the valley.

 

[xXCrossCheckXx]

i am running a core i5-750. with a corsair A70 http://www.newegg.com/Product/Produc...82E16835181011

i have it running @ 4.0ghz. with no voltage adjustments.
it runs at 28-32c on idle when running cod:BO it jumps to 60C mid game. no bosd's
running it in a haf-932 amd edition (debadged-killer deal from the egg) with bottom(helps keep dust out), rear and top as exhaust. and front as intake.

you might fair better with a higher OC with a case with better air flow. the best cpu cooler wont do squat with a case with crappy airflow

 

[Grooveriding]

I would say something is off. I have a D14 and an i7 920 @ 4.2 with 1.3V and it loads with prime at 70C.

I have very high idles though, around 45C. I write this off to having a lot of hot components in my case (3GPUs and 6 mechanical hard drives)