Sign in with
Sign up | Sign in
Your question

Overclock of i7 920 @ 3.6 GHz = 95 degrees?!

Last response: in Overclocking
Share
May 29, 2009 4:16:28 PM

I've been overclocking my i7 920 CPU (C0) to 3.6 GHz using a CPU Ratio of 16 and BCLK of 200. When running LinX 20 passes and looking at Real Temp 3.00, my CPU's highest temp reached 95 degrees! Obviously this is too high, but why? If I overclock to 3.4 the highest is 86 degrees and stock 2.6 is 75.

I'm running with a P6T Deluxe V2 with 12GB of Corsair Dominator C8 RAM, Noctua SE1366 U12P (using a single Scythe SFF21G fan), Scythe G fan at the back and front, with an ATI 4870X2 installed. My voltages are: CPU Voltage 1.275, QPI/DRAM Voltage 1.275 and DRAM Voltage is 1.6.

I'm quite sure I didn't apply too much Noctua thermal compound, I was careful to put on a thin layer, and the fans are running full speed. When at 3.6 GHz I touched the Noctua heatsink, it was running warm to hot. Just wondered why I can't reach 3.6? Yet if I Google people can reach that but at around 70 degrees.

Thanks.
May 29, 2009 4:22:01 PM

I'm running a i7 920 (D0) at 4.0ghz, using real temp my CPU's highest temp is 72c after running prime 95 for 30 hours.

The only difference I see is i'm using the original P6T Deluxe and only 6GB of Ram, with the thermal light ultra 120 extreme 1366rt, and artic 5 silver thermal paste. I did lap my CPU & Heatsink.

Are all 4 of your cores close to the same temp? Did you make sure the thermal paste covered all 4 cores?

The only thing I can think of is your using the (C0) stepping instead of spending the extra for (D0).
May 29, 2009 4:41:20 PM

I never bought the D0 CPU as it when I bought my CPU is wasn't available, I bought my C0 months and months ago. Actually just to confirm, my S-Spec is SLBCH and Batch number is 3839A604, is that a C0? You're doing well at 4.0 GHz with only 72 degrees, incredible. My CPU cores at different speeds have these temperatures:

3.4 GHz (17 x 200) = 86 82 82 76
3.6 GHz (18 x 200) = 95 90 90 83

I've no idea why it's running that hot. That is under a LinX load 20 passes. I'm sure I never applied too much compound to the CPU, I have photos of it so I checked. I've no idea, the only thing I've got to try is re-applying it, then again the board is installed, I would hope I could remove the HSF & CPU without taking the board out.

Thanks.
Related resources
May 29, 2009 4:47:08 PM

I have the same fan in push-pull fan setup... my I7 at 3.86 max temp after prime 95 30 hours is 72c. I would check if my heatsink is welle seating, don't put to much thermal paste and used the one coming with the noctua package it's really good.
May 29, 2009 5:02:18 PM

C0 is (SLBCH) and the D0 is (SLBEJ)2. Depending what you do, you can always disable HT to get your temps down, but I leave HT on because I need it for what I do for work.
a b à CPUs
a b K Overclocking
May 29, 2009 7:36:44 PM

aln688,

The problem is that you're testing with LinX, which can be useful for stability testing, but is inappropriate for thermal testing. Since LinX is a cyclic workload which loads the processor registers with all one's during certain test segments, it produces the highest possible temperatures, that can equal 115% workload, and is thoroughly unrealistic.

Prime95 Small FFT's is the standard for thermal testing, because it is a steady-state 100% workload, and allows apples to apples comparisons. With this in mind, remember that even the most grueling games or application will typically never exceed a sustained 70 to 85% worload.

Re-test your rig with Prime95 Small FFT's, then report your temperatures. Make sure that you test with case covers removed, and all fans at 100% RPM, in order to eliminate unforseen cooling variables. A 10 minute test is adequate, since 97% thermal saturation is typically reached within 7 to 8 minutes.

Intel's i7 Thermal Specification from their Processor Spec Finder - http://processorfinder.intel.com/details.aspx?sSpec=SLB... - is 68c, and is CPU temperature, NOT Core temperature, which is a common misconception among many users. There's a 5c gradient between the CPU temperature sensor and the Core temperature sensors, which is shown in the following Intel document - http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf

Although Tjunction Max is 100c, which is for Throttle and Shutdown protection, and is far too hot for sane operation, the corresponding CPU temperature would be 95c. As such, any CPU temperature that exceeds 68c and any Core temperatues that exceed 73c should be regarded as an overtemp condition.

For more information on processor temperatures, please refer to my Sticky at the top of this Forum: Core i7 and Core 2 Temperature Guide- http://www.tomshardware.com/forum/221745-29-sticky-core...

Comp :sol: 
May 29, 2009 7:54:54 PM

CompuTronix,

Thanks for the insight, I wasn't aware of that. Maybe when I see people post temps, they are from something other than LinX? I'm going into the BIOS and will set the voltages I had before, with a 18 ratio and 200 BCLK, this'll give me 3.6 and just over 1600 MHz for the RAM. FYI, anything over 1650 will shrink my RAM from 12GB to 8GB, so I try to keep the RAM per spec. I'll grab Prime95 and try a run with that, then I'll report back with the findings.

Thanks again.
a b à CPUs
a b K Overclocking
May 29, 2009 8:29:28 PM

I've found that when discussing temperatures, the most common problem is that users fail to comprehend the importance of being specific about the information they're providing. Temperatures and overclocking are all about specifications, definitions and standards, so when users fail to be specific, then the entire discussion is reduced to apples and orages thermal fruit salad in a blender.
June 8, 2009 11:07:23 PM

Alright, I managed to find the time to re-seat my CPU HSF, using a small line method of Noctua thermal compound. If I use the blob method, as there's fine lines on the Noctua heatsink, it squashes the compound into a long egg shape.

After aiming for 3.6 GHz using a CPU Ratio of 18 and BCLK of 200, with CPU Voltage and QPI/DRAM Voltage set to 1.25 each, still using 12GB of RAM, running Prime95 Small FFTs, I get the highest temps of:

Core 0 = 82
Core 1 = 79
Core 2 = 80
Core 3 = 77

Is this more reasonable? More in line with "normal" temps of a 3.6 overclock? The highest I get running LinX, which granted thrashes the CPU more, I get a highest temperature of 85.

Thanks.
a b à CPUs
a b K Overclocking
June 8, 2009 11:39:40 PM

Your average or "mean" Core temperature is 79.5c.

Specifications:

Vcore Max 1.375
Tcase Max (CPU temp) 68c
Tjunction (Core temp) 73c
Ambient 22c

What was your ambient?

Did you test as I suggested with covers removed and all fans at 100% RPM?
June 9, 2009 1:02:55 AM

The ambient here is 71 degrees. Yes I tested with the side covers removed and all fans running at full RPM throughout the tests. Maybe I should keep trying to re-apply the compound and HSF until I get a temperature lower than 80 degrees?
a b à CPUs
a b K Overclocking
June 9, 2009 2:30:29 AM

Reducing Vcore will reduce temperatures. If you haven't already done so, you might try investing some time in finding exactly where your stability threshold is for 3.6 Ghz. Fine tuning Vcore often improves temperatures.
June 9, 2009 3:27:47 AM

The CPU Voltage (assuming this is the same as Vcore) is at 1.25, which is the lowest it will go in the BIOS before it jumps into Auto mode. I already altered the voltages the lowest they will go, so far this is the best temperature (if you can call 82 degrees "best") I've had.
a b à CPUs
a b K Overclocking
June 9, 2009 3:58:23 AM

The mean Core temperature is only 6c over spec during the 100% workload of Prime95 Small FFT's, however, your processor will alwys run considerably cooler during games and applications. My best recommendation is that you declare victory and have a beer.

Comp :sol: 
June 9, 2009 4:18:33 AM

LOL! Sounds great.

I have to say, the temperature I'm getting now is better than before, but I'm skeptical of people who reply with supposed low temperatures under a similar setup. It also has to be considered that I have an ATI 4870 X2 under the CPU HSF, throwing out a lot of heat as well.

Under the current temperature I'm getting now, I'm using the Noctua fan, a single fan, not dual. I'm considering replacing it with the Scythe SFF21G 75CFM fan that I was using before, that for a fact I know will drop the temperatures by 1 or 2 degrees.

You think 83 (occasionally 84) degrees for Prime95 is acceptable at 3.6 GHz (18 x 200)? In a way I'm pleased it's better than before, but in another way I'm unsatisfied as other people get less, although I don't know their ambient, super-hot video card underneath, 10lb HSF cooler cooling down their system, etc.
June 9, 2009 4:29:33 AM

Right now the ambient is 77 degrees, I ran Prime95 again and it peaks at 85. Tomorrow I'm going to run it in the morning to get an idea of the highest it gets when the day temperature isn't so high. Then I'm going to plug in that SFF21G fan onto the cooler, see how that reduces things further.

I think from here on I'm not going to re-seat the HSF again, I've already re-seated it seven times this evening, trying different methods for applying the compound, as stated, blob wouldn't work due to the lines in the Noctua heatsink, the line method seems to work fine, resulting in even coverage.
a b à CPUs
a b K Overclocking
June 9, 2009 4:40:36 AM

Rember to take into consideration that under these test conditions, your graphics card is at idle. Also, Intel's factory calibrations of the Core temperature sensors (DTS) aren't perfect. There are always Cores which are higher or lower than their neighbors, so "mean" or " average" Core temperature is the most relevant value.

You might try re-testing with your tower positioned on it's side, but I doubt that your results will improve notivably. If you instead install a second fan on the rear of your cooler for a push-pull configuration, then your load temperatures wll most likely meet spec when ambient is at 22c, which is the standard value for thermal testing.
June 9, 2009 4:50:25 AM

I thought about another fan on the CPU, but I don't have a motherboard header for it. I also routed all my PSU cables so that I didn't need a regular 4-pin molex, but I do have the Scythe SFF21G fan on the rear of the case literally two inches from the CPU cooler.

Yes I noticed the cores have varying temperatures, which seems odd. I had one core at 84 and the fourth core at 77, amazing. I'm wondering if I should try re-seating the CPU cooler once more? Or trying another compound? But I wonder if another compound will really produce much better results, the Noctua NT-H1 is supposed to be good.
a b à CPUs
a b K Overclocking
June 9, 2009 4:59:07 AM

Thermal differences among indivdual Cores is not at all "odd", but is instead both typical and common. Relax and enjoy your rig.
June 10, 2009 1:03:58 PM

One other thing to check is how flat your CPU is. I had a very frustrating afternoon trying to get my heatsink to work correctly (I was getting temps in the low 90s) and finally realized the top of my CPU was slightly concave, giving me contact only on the sides but with enough pressure to blob my thermal compound.
June 10, 2009 5:02:41 PM

The CPU seems quite flat actually, I seem to have gotten lucky there.

Well I tried another run, this time using LinX at 9:30 AM in the morning here, this time I get 77 degrees, still at 3.6 GHz, the ambient is 75F. I'm wondering if I should bother re-applying the compound to get it as "lean" as possible, hoping to see a lower temperature.
June 16, 2009 4:46:04 AM

Just something to try, why dont you raise your multi and lower your bclk? also put your vcore back to 1.20. dont know about 3.6ghz but mine is stable at 3.5ghz with only increasing the bclk (xbit labs found the same on their test bench). To go to 3.8 I need to increase vcore and other voltages and thats when my temps start going crazy. between 1.3vcore and 1.20 i went from 70c to 60c max temps on real temp.
June 16, 2009 3:17:04 PM

Thanks for the suggestion, I actually did that, an update to the situation...

I found why I couldn't reduce the CPU voltage below 1.25. There's three jumpers on the motherboard which allows CPU Voltage, QPI/DRAM and DRAM voltage overvolting. I put all three jumpers back to their default position and now I can set low voltages in the BIOS. These jumpers were in the non-default position from new (Newegg.com).

With the Noctua U12P heatsink, Noctua NT-H1 compound, a 4.0 GHz overclock has a maximum temperature in LinX 100 passes of 71C. My ambient temperature is 80F. I'm happy with this, I think I'll stick with it. Previously I got 71C at 3.6 GHz and 76F ambient. To get 4.0 GHz I had to use 1.23125 CPU Voltage, anything less will crash LinX and Prime95.

I'm going to move my drive cage from the middle three slots to the lower three slots, then install a Panaflo medium 120x120x38 in the middle to blast cold air into the middle of the board. The area around the Northbridge is very hot, I need more air blowing over that area. I'll also put a Panaflo 120x120x38 at the back as well.

Thanks for your help.
June 16, 2009 10:30:29 PM

Dam thats a good OC on that low voltage must be a real good chip, I cant even get to 4ghz on mine...
July 3, 2009 9:59:48 PM

Which cpu core temp program are you using? It may be of help to know that temperature programs can give very different temperature results.

I wasted two days reseating CPU, etc because I used Speedfan. CoreTemp seems to be giving correct readings.

August 23, 2009 9:29:39 AM

Can you guys tell if my temps are correct with this spec.

Just build my i7 rig:
proc: i7 920
mobo: Gigabyte GA-EX58-UD4P (Rev. 1.0)
ram: Corsair 6GB (3x2GB) DDR3 1600MHz XMS3 Memory TR3X6G1600C9
cooling: Noctua NH-U12P Special Edition Socket 1366

My temps in idle are: 36 36 38 36 ambient temp was 22c
Prime95 blend test after 9 hours: 65 65 66 63 ambient temp was 22c

I have used thermal compound provided with cooling fan.

!