I7 load temps = 75° on 1.275v - too high?

[krazynutz]

Hey everyone. I know core i7 temps are a hot topic (cheese) and I have to be another guy who's concerned. I've heard i7's tend to run a little toastier than Core 2's. Lemme know your thoughts on my setup.

Antec Nine Hundred Two
Xigmatek HDT-S1283 w/1366 Crossbow
OCZ Freeze compound
Core i7 920 @ 3.8Ghz on 1.275v

Monitoring with Real Temp

Idle temps: 40-45°
Load temps: 70-75°

With my setup, are those temps a bit high? Should I re-seat my cooler? Maybe take some compound off? I think I may have put too much on.

Thanks

 

[CompuTronix]

HT on or off?

 

[Vythiel]

Based on this, you should be staying below 68 degrees Celsius under maximum load. You can download OCCT and run the stress test using the "small data set" configuration for an hour to see what your current maximum load temperature is.

 

[CompuTronix]

Intel's spec in their Processor Spec Finder - http://processorfinder.intel.com/details.aspx?sSpec=SLBCH - for Core i7 Tcase Max is 68c (CPU temperature), which corresponds to 73c Tjunction (Core temperature), since Tj is 5c higher than Tc.

Also, if you rercall from the Core i7 and Core 2 Temperature Guide - http://www.tomshardware.com/forum/221745-29-core-core-temperature-guide - and from our friends at Xbit Labs - http://www.xbitlabs.com/articles/cpu/display/core-i7-920-overclocking_11.html#sect0 - Prime95 Small FFT's is the standard for CPU stability and CPU thermal testing due to it's steady-state 100% workload. The following quote is from the Xbit article:

"we tested our system stability using 64-bit Prime95 25.7 utility in Small FFTs and Blend modes. This program proved to be the best tool for detecting over-overclocking and reported errors when most other popular stability check tools (including OCCT, LinX and IntelBurnTest) showed a pass."

Also, remember that thermal saturatiuon is reached within 7 to 8 minutes with Prime95 Small FFT's, so a 10 minute test is adequate.

Comp

 

[krazynutz]

HT is on. Sorry for any confusion. My core temps are 70-75° at load which would put my CPU temp at 65-70. Right? So my temps are a bit high for my current voltage then. Yes? I'll redo the compound and re-seat and see if my temps go down.

 

[Vythiel]

@CompuTronix
Let me start by saying that I laughed a little at that "10 minutes is enough" comment. An hour means nothing for stability (let alone ten minutes) as the minimum running stress test should be 8 hours or more (24hours for a final stress test). The only exception to this is testing in between incremental increases, for which one hour is enough to warrant moving up.

I've been overclocking for years and have used Prime95 until OCCT came out and became stable. The rate at which errors show up is generally random. I've had the same overclock settings show errors after minutes, or run for hours before the error appeared. This is especially true as you get closer to a stable setup. Furthermore, OCCT will test the CPU, GPU, RAM, power supply, and motherboard stability (separately or simultaneously) while showing the temperatures and voltages. It creates line graphs (temperature, workload, voltage) after a test fails or completes which helps in pinpointing instability problems. You can use whatever test you like, but they all need to be run a minimum of 8 hours before you can claim that your overclock is stable (I've seen errors in prime after the 22nd hour of testing in Prime95).

Furthermore, while you are correct about Tcase vs Tjunction, it is still not a good idea to take the processor all the way to it maximum core temperature. a slight increase in ambient temperature will throw your processor into unsafe operating ranges. These are things that must always be considered when overclocking for long term use.

@krazynutz
I wouldn't recommend running that processor hotter than 70c (occt/core temp) since your computer rooms ambient temperature will effect your CPU temperature as well. Staying at 70c or below will give you a safe buffer zone for hot days/temperature fluctuations.

 

[RJR]

@CompuTronix
Let me start by saying that I laughed a little at that "10 minutes is enough" comment. An hour means nothing for stability (let alone ten minutes) as the minimum running stress test should be 8 hours or more (24hours for a final stress test). The only exception to this is testing in between incremental increases, for which one hour is enough to warrant moving up.

He just stated that for thermal purposes ONLY, not for stability testing. Second the 24 hour stability testing.

 

[Xenonvector927]

I read on a thread somewhere that turning off HT makes a big difference to i7 temps. But HT improves floating point calculations (like Prime95) by 30% or so. I don't think it makes a huge difference in games, so if you're getting stressed about temps and want to OC higher, turning off HT may be the way to go.

 

[Vythiel]

He just stated that for thermal purposes ONLY, not for stability testing. Second the 24 hour stability testing.

I was a little thrown off by his comment that 7-8 minutes reaches thermal saturation. He then followed with "10 minutes should be enough" so it seemed like he meant 10 minutes is a good amount of time to run a reliable stress test.

However, even for thermal testing it's not enough. 10 minutes is not enough time for a computer case's temperature to stabilize. One hours is the minimum "in between" test for this reason. This is what I've found out during my own overclocking experiences.

 

[Xenonvector927]

I was a little thrown off by his comment that 7-8 minutes reaches thermal saturation. He then followed with "10 minutes should be enough" so it seemed like he meant 10 minutes is a good amount of time to run a reliable stress test.

However, even for thermal testing it's not enough. 10 minutes is not enough time a computer case temperature to stabilize. One hours is the minimum "in between" test for this reason. This is what I've found out during my own overclocking experiences.

I tend to arrive at my highest temps after 10mins. I can leave it there for a couple hours, and the temp will be the same as at the 10min mark. It works for me, anyhow. CompuTronix seems to be the Einstein of the cooling world, I've seen plently of his posts on this kinda stuff, as well as his really thorough guide. I trust him on this. After 10mins, your temp will stay stable. Obviously this does not indicate stability (but he never said that it DID!).

 

[Vythiel]

I tend to arrive at my highest temps after 10mins. I can leave it there for a couple hours, and the temp will be the same as at the 10min mark. It works for me, anyhow. CompuTronix seems to be the Einstein of the cooling world, I've seen plently of his posts on this kinda stuff, as well as his really thorough guide. I trust him on this. After 10mins, your temp will stay stable. Obviously this does not indicate stability (but he never said that it DID!).

That's fine, I'll agree to disagree. I trust my own extensive results because they are first hand. My maximum temperature using prime95 is 63c while OCCT takes it to 70 and provides far more data, so I choose OCCT. I've seen my temperature "appear" to reach maximum after 10 minutes, but continue to rise slowly depending on the airflow through the computer case, so I choose to test for at least an hour. I'm not really stating anything based on what some review site prints, but based on my own lessons learned. I definitely will not try to cram this knowledge down anyone's throat, it's shared to assist those that both need and want it.

 

[Xenonvector927]

That's fine, I'll agree to disagree. I trust my own extensive results because they are first hand. My maximum temperature using prime95 is 63c while OCCT takes it to 70 and provides far more data, so I choose OCCT. I've seen my temperature "appear" to reach maximum after 10 minutes, but continue to rise slowly depending on the airflow through the computer case, so I choose to test for at least an hour. I'm not really stating anything based on what some review site prints, but based on my own lessons learned. I definitely will not try to cram this knowledge down anyone's throat, it's shared to assist those that both need and want it.

Yeah. There are so many variables that would affect results, no two results will be the same. Peace.

 

[CompuTronix]

I was a little thrown off by his comment that 7-8 minutes reaches thermal saturation. He then followed with "10 minutes should be enough" so it seemed like he meant 10 minutes is a good amount of time to run a reliable stress test.

However, even for thermal testing it's not enough. 10 minutes is not enough time for a computer case's temperature to stabilize. One hours is the minimum "in between" test for this reason. This is what I've found out during my own overclocking experiences.

vythiel,

I beg to differ. As you appear to be an advanced enthusiast, then you will understand this material. Since temperatures and overclocking are all about specifications, then I'll be very specific.

There is a standardized method for thermal testing and calibration, which among other settings, is performed with case covers removed, and all fans at 100% RPM. This eliminates cooling variables among cases, processors and CPU cooler variants, and DOES indeed result in thermal saturation being reached within 7 to 8 minutes. This has been tested and verified hundreds of times on a wide variety of hardware configurations, so under these conditions, 10 minutes is definately adequate.

You'll find that the Author of Real Temp, unclewebb (Kevin Glynn), with whom I collaberate, also agrees on these points, as well as with the exclusive use of Prime95 Small FFT's for thermal testing, again, due to it's steady-state 100% workload characteristics. We are well aware of the capabilities, merrits and drawbacks of OCCT, Burn Test, and many other shortcut utilities for testing overclock stability, but only Prime95 Small FFT's is perfectly suited for thermal testing.

The complete "Test Setup" for temperature calibration is described in detail in the Core i7 and Core 2 temperature Guide - http://www.tomshardware.com/forum/221745-29-core-core-temperature-guide - which is a Sticky at the top of this Forum.

As you're new here at Tom's, allow me to appologize for the Stickies, since they're poorly marked as such, so many newcomers fail to notice them, or read them. Below, you may find Part 2, Item D of interest, as it addresses your point concerning case temperatures. I invite you to study the entire Guide, and I look forward to your comments.

Section 9: Calibrations

Default temperatures are rarely accurate. The following two part procedure is designed to achieve two objectives:

■ Provide minimum Ambient to Tcase Delta for accurate Tcase Idle Calibration.

■ Provide maximum Tcase to Tjunction Gradient for accurate Tjunction Load Calibration.

Note: It is preferred, but not required, that Calibrations be conducted as close to 22c Standard Ambient as possible, which provides a normal temperature ceiling, and maintains environmental consistency for comparing Idle and Load temperatures among processor variants and system platforms.

Prerequisites:

(A) CPU cooler correctly installed.

(B) Print this Section to use for BIOS settings and Calibrations.

(C) Record or photo or Save Profile all BIOS settings for quick restore when Calibrations are complete.

(D) Follow the Test Setup: (Standardized configuration for maximum cooling at Auto Vcore, Frequency and Multiplier).

Computer Case Covers = Removed
Computer Case Fans = Manual 100% RPM
CPU Fan = Manual 100% RPM
CPU Frequency = Auto (See Note 1: below) **
CPU Internal Thermal Control = Enabled
Enhanced C1 Control (C1E) = Enabled
Internet = Disconnected
Memory Frequency = Auto
PECI (If Equipped) = Enabled
Speedstep (EIST) = Enabled
Vcore = Auto (See Note 1: below) **
Vdimm = Auto
Windows Programs = Closed


Part 1: Calibration - Tcase Idle (Uses maximum cooling at minimum Vcore, Frequency and Multiplier)

** Note 1: If BIOS does not respond properly to Auto Vcore, Frequency and Multiplier settings, then use an appropriate combination of manual settings to provide the following in CPU-Z:

Core Voltage = 1.100 V
Core Speed = 1600 Mhz (Core i7 processors)
Core Speed = 1600 Mhz (Core 2 - 65 nm processors)
Core Speed = 2000 Mhz (Core 2 - 45 nm processors)

(A) Measure Ambient near the computer case air intake, clear of warm exhaust. A trusted indoor analog or digital thermometer will suffice. The accuracy of this device and measurement will determine the overall accuracy of the Calibrations.
(B) Boot into Windows. Close all programs, background processes, Screen Savers, SETI, Folding and Tray software. Press Ctrl-Alt-Delete, click on Task Manager, then click on the Performance tab to confirm CPU Usage is less than 1%. Use the Applications and Processes tabs to close programs if necessary.

(C) Open CPU-Z and SpeedFan. Observe CPU-Z for Intel's Speedstep to decrease Core Voltage, Core Speed and Multiplier to minimum values. Observe SpeedFan, allow 10 minute at Idle to ensure that temperatures decrease to minimums, then record Tcase Idle.

■ Tcase Idle = Ambient + Z.

"Z" compensates for Idle power dissipation and CPU cooler efficiency. Use the Scales in Section 6 to find "X" Idle Power, use the links in Section 14 to find "Y" Cooler Efficiency, use the Table and Formula below to find "Z", then add Ambient to find Tcase Idle.

■Idle Power and Cooler Efficiency Table:

X = 2 . . . Idle Power: 8W . . . . Y = 2 . . . . Cooler Efficiency: High-end
X = 3 . . . Idle Power: 12W . . . Y = 3 . . . . Cooler Efficiency: High mid-range
X = 4 . . . Idle Power: 16W . . . Y = 4 . . . . Cooler Efficiency: Mid-range
. . . . . . . . . . . . . . . . . . . . . . . . . Y = 5 . . . . Cooler Efficiency: Low mid-range
X = 6 . . . Idle Power: 24W . . . Y = 6 . . . . Cooler Efficiency: Low-end / Stock Intel

■Tcase Idle Formula:

(X + Y) / 2 = Z + Ambient = Tcase Idle.

Note 2: CPU and cooler combinations which are both lapped may subtract 1c from Tcase Idle.

Example 1: Duo

Idle Power: E2160, Stepping M0, Idle 8W, so X = 2.
Cooler Efficiency: Mid-range – Arctic Cooling Freezer 7 Pro, so Y = 4.
(2 + 4) / 2 = 3, so Z = 3.
Ambient = 22c.
3 + 22 = 25, so Tcase Idle = 25c.

Example 2: Quad

Idle Power: i7 920, Stepping C0, Idle 16W, so X = 4.
Cooler Efficiency: High-end – Xigmatek HDT-S1283, so Y = 2.
(6 + 2) / 2 = 3 so Z = 3.
Ambient = 22c.
3 + 22 = 25, so Tcase Idle = 25c.

(D) Configure Offset correction as shown in Section 11. Since the Tcase sensor was designed to be linear from Idle to Load, Tcase Load will also be accurate.


Part 2: Calibration - Tjunction Load (Uses maximum cooling at Stock Vcore, Frequency and Multiplier)

** Note 1: If BIOS does not respond properly to Stock Vcore, Frequency and Multiplier settings, then use an appropriate combination of manual settings to provide the following in CPU-Z:

Core Voltage = 1.250 V
Core Speed = Stock Mhz

(A) Start Prime95 Small FFT's. Observe CPU-Z for Intel's Speedstep to increase Core Voltage, Core Speed and Multiplier to Stock values. Observe SpeedFan. Heat saturation is typically reached within 7 to 8 minutes, so allow 10 minutes at Load to assure that temperatures increase to maximums, then record Tjunction for each Core.

■ Tjunction Load = Tcase Load + 5c.

Example 1: Duo

Tcase Load = 45c
Tjunction Load = 50c

-Tcase/Tjunction-
--45--/--50--50--

Example 2: Quad

Tcase Load = 45c
Tjunction Load = 50c

-Tcase/Tjunction-
--45--/--50--50--50--50--

(B) Configure Offset corrections as shown in Section 11.

(C) Stop Prime95, then allow the system to Idle for 10 minutes. Tjunction Idle should be ~ 4c higher than Tcase Idle.

Note 2: Tjunction sensors were designed to be linear at high temperatures for Throttle and Shutdown protection, so Tjunction Idle could indicate too low or too high. Many 45 nanometer variants have faulty sensors that "stick" and might not Idle below 50c. Sensors can be tested using Real Temp - http://www.techpowerup.com/realtemp/

■ If Tjunction Idle is not ~ 4c higher than Tcase Idle, then use Tcase Idle for accurate Idle temperature.

(D) Repeat Item (A) and allow Prime95 to run past 10 minutes while reinstalling covers. If temperatures increase, then computer case cooling should be improved.

(E) Restore the system to original or custom BIOS settings and hardware / software preferences.

Clear enough?

Comp

 

[Vythiel]

Thank you for the information. I followed the work on real temp and tried the application when it was released on the XtremeSystems forums. I did agree with you on Tcase and Tjunction before so that information was never in question. Having a set of standardized rules for experimentation is absolutely necessary, the same holds true for overclocking under the same conditions that the computer will be run at in the long term. This means that the original poster needs to consider what happens while his case is closed and thus needs to overclock with these conditions in mind.

Now, consider my point of view on OCCT. I've used prime95 for years and it's always taken 8+ hours to see if my overclock was close to stable. Since switching to OCCT, I can tell if my system is close to stable in approximately 1 hour instead. Because of this, the time it takes to reach a completely stable overclock in considerably less. Furthermore, the tools provided by OCCT amplify it's convenience even further. I switched from prime to OCCT because it exposes instability much faster than prime95 in my experience.

I believe what you are trying to say is that keeping the CPU at a "100% steady-state" gives a better indication of the temperature that the CPU will actually reach under 'real-world' conditions. I find this to be untrue as real world applications do not put a "steady-state" of stress on the CPU and will fluctuate quite frequently.

The only real question I have is why does the "steady-state" of prime95 provide a better stress test when first hand experience proves otherwise?

 

[CompuTronix]

I emphasize "steady-state" as imperative for thermal testing, which is especially critical during calibrations. I'm sure you'll agree that it's next to imposible to calibrate load temperatures when the load is fluctuating, which in turn, causes the temperatures to fluctuate.

However, for stability testing, if an overclock will remain stable throughout OCCT as well as Prime, it must also remain stable throughout benchmarks and applications, in order to be regarded as "stable". I've seen instances where overclocked rigs pass various stability tests, yet crash on certain applications, so we can never be too thorough.

I think most overclocking enthusiasts understand that "real world" workloads rarely spike above 80%, so temperatures rarely spike to test levels.

 

[Vythiel]

I agree about being thorough, but that doesn't really answer my question. How is Prime95's "steady-state" better than OCCT for testing the stability of an overclock? From my experience, OCCT has proven to be a better stess testing utility that reveals instabilities more quickly and causes the CPU to reach a higher temperature than Prime95 (with my current system, Prime95=63c and OCCT=70c). Unless there is convincing evidence provided which shows that Prime95 is better than OCCT for standard overclocking (not finding the Tjunction of a CPU) then I stand by my initial recommendation.

 

[Xenonvector927]

Prime95 seems to be the most widely used, but just because OCCT gets your temps higher doesn't neccesarily mean that it's better as a TEMPERATURE testing program. Temps NEVER go as high as Prime95, so OCCT would just be overkill.

 

[jay2tall]

I test with OCCT, but make sure you have the newer version that supports the i7. I have 3.0.1 and make sure you set it to custom and check "Hyperthreading as Dual Core" otherwise it won't hit it 100% it will HT the test and not give you a complete stress test.

I'm running at 1.3v @ 4Ghz and with my THermalright 120 Extreme I still top out at just over 90C. I sit at about 55-58C idle. Normal loads I hit about 70Cish

 

[Xenonvector927]

I test with OCCT, but make sure you have the newer version that supports the i7. I have 3.0.1 and make sure you set it to custom and check "Hyperthreading as Dual Core" otherwise it won't hit it 100% it will HT the test and not give you a complete stress test.

I'm running at 1.3v @ 4Ghz and with my THermalright 120 Extreme I still top out at just over 90C. I sit at about 55-58C idle. Normal loads I hit about 70Cish

Geez! I hope that the i7 revision will eliminate or diminish these temperatures. It's getting newbies to computer building like OP all freaked out. There's thousands of threads out there with the exact same question. It's getting ridiculous.

 

[krazynutz]

i7 revision? What is this you speak of? You mean the D0 stepping?

 

[krazynutz]

i7 920 @ 4.1 | Vcore Load 1.360
68c CPU | 73c Core @ 22c Ambient
Xigmatek HDT-SD964 | Dual Fans

Comp, you need to share your secrets. At 4.1Ghz @ 1.36V with a HDT-SD964 (92mm HSF combo), how the hell are you topping out at 73° on your cores? Do you even have this in a case?

I have an Antec 900 II, an HDT-S1283 (120mm HSF combo), and my 920 @ 3.8Ghz on 1.28V (at load) and my core temps max out between 70-75° (hence the thread). I have OCZ Freeze as a compound as well. My house thermostat says 72°F (22°C).

What voodoo magic are you conjuring up over there? Do the dual fans really make that big of a difference? Is there an offset I need to plug into Real Temp? Do you have HT on or off? Tell me man!

 

[ezynetwork]

Wish i could get these temps at stock speed I am running idle 41 - 45c and load i am getting 67 - 70c using artic silver 5 and Noctua.

 

[Xenonvector927]

i7 revision? What is this you speak of? You mean the D0 stepping?

I've just read on a couple forums that a revision/stepping of the i7 chips is not too far away. Could just be rumours, though...

 

[jay2tall]

I have a D0 stepping i7, Runs nice and toasty, but I tell you what it OC's well, and it runs ROCK SOLID stable at 4GHZ no sweat.

 

[Xenonvector927]

I have a D0 stepping i7, Runs nice and toasty, but I tell you what it OC's well, and it runs ROCK SOLID stable at 4GHZ no sweat.

I have a D0 stepping P4!