How do you measure 'load' temps on i7

[magnosis]

I'm running 3.8GHz (BCLK=190 x20, vCore 1.32) on the i7 920 on Asus P6T. No sign of failure after 6h of 8xORTHOS.

But I have a hard time figuring out if my temperatures are OK. I have probably read about different forum threads this w-e, very hard to get a definitive answer.

How do people typically measure 'load' temperature ? I see in 75% of the posts to use Prime95.

If I load 1 instance of Prime95, it only gets to anywhere between 53% and 67% cpu utilisation, with temps as follow:
(using lapped TRUE120 with 2x 44CFM fans in push-pull)


Mobo: 45c, CPU: 47c, CPU0: 58c, CPU1: 58c, CPU2: 56c, CPU3: 56c


That seems to be in line with what the majority is getting (30-40c idle, 50-60c load).

But if I load 8 instances of ORTHOS, the story is very different. Then I get truly 100% utilization (on all 8 cores, i.e. 4 cores w/ HT)
Then the temps go much much higher:


Mobo: 46c, CPU: 78c, CPU0: 88c, CPU1: 88c, CPU2: 85c, CPU3: 85c


This is measures with Everest, CoreTemp and RealTemp. They are all within 1c of each other.
SpeedTemp reports much lower temps (same load w 8xORTHOS):


Mobo: 46c, CPU: 78c, CPU0: 72c, CPU1: 69c, CPU2: 68c, CPU3: 60c


1. What does the majority consider 'load' : 1 x Prime95 (53-67%cpu), or 8 x ORTHOS (100%cpu) ?

2. Am I correct to trust Everest+CoreTemp+RealTemp since they yield [pretty much] the same numbers ?

3. The normal/nominal operating temp for i7 is 67.5c -- is that for the CPU, or for each core (CPU0,1,2,3) ?

4. Is it a bad thing to get 88c on a full 100% load ? Such load in real life is very unlikely, but is it bad ?

 

[nonxcarbonx]

I don't know what kind of threads you're running in prime95 to only get 60% utilization, but if that is it, don't even use those. 88 degrees is too hot, back down and aim for the low 80s. I always use hardware monitor, but if your temps are consistent, trust them.

 

[magnosis]

After talking to some local folks I realized I'm prolly using an obsolete version of Prime95, since it *should* exercise *all* cores at 100%.

I just got 2x Scythe SFLex 1600rpm for push/pull on the TRUE 120, will report new measurements tonight with up-to-date Prime95 load.

 

[magnosis]

Do most people measure load temps with HyperThreading ON or OFF ? From what I read turning HT off can reduce up to 10c ..

 

[nonxcarbonx]

On because hyperthreading is a nice application to have because it can increase performance and multitasking by 15%, so unless you can get a 15% or greater overclock by disabling it, it is nice to keep it on. (It's also nice to have for folding@home or boinc apps)

 

[cjl]

Test with hyperthreading in the same configuration as you will be using it. If during use, hyperthreading will be off, test it that way. If during use, hyperthreading will be on, test it that way.

As for what to test with? I use Prime95 blend, and the latest version of Prime95 should support the use of all 8 threads.

 

[magnosis]

After an evening well spent, I am pleased to report the satisfying results

The short version:
I lowered the temps by 22.5c !
- 7.0c with new fans push/pull & better thermal paste application
- 10.75c with HyperThreading OFF
- 3.4c with lower vCore

The long version:

0. Initial specs and cooling:
Stock i7 920 on P6T with default bios settings
Coolermaster Storm Sniper:
-Front: 200mm 1000 RPM 142 CFM
-Bottom: -
-Side: 200mm 1000 RPM 142 CFM
-Top: 200mm 1000 RPM 142 CFM
-Rear: 120mm 1200 RPM 30 CFM
TRUE 120:
-Push: oem 1600RPM 63.7CFM
-Pull: -

Temps Idle: Mobo=51 CPU=39 CPU0=43 CPU1=40 CPU2=44 CPU3=38


1. Add bottom fan + pull on TRUE120 : avg -3.6c idle
-Bottom: Scythe Slipstream 800RPM 40CFM
-Pull: Scythe Slipstream 800RPM 40CFM

Temps Idle: Mobo=47 CPU=39 CPU0=40 CPU1=40 CPU2=37 CPU3=37
Temps Load: Mobo=47 CPU=60 CPU0=58 CPU1=58 CPU2=56 CPU3=56


2. Overclock @ 3.80: avg +4.0c idle +29.5c load
-BCLK: 190
-vCore: 1.30
-DRAM: 1.65

Temps Idle: Mobo=46 CPU=35 CPU0=44 CPU1=44 CPU2=41 CPU3=41
Temps Load: Mobo=46 CPU=78 CPU0=88 CPU1=88 CPU2=85 CPU3=85


3. New fans + reapply thermal paste : avg -2.5c idle -7.0c load
-Rear: TRUE120 oem fan 1600RPM 63.7CFM
-Push: S-Flex 1600RPM 63.7CFM
-Pull: S-Flex 1600RPM 63.7CFM
->Could the TRUE120 actually be using the S-Flex 1600 as its oem fan ?
-Re-apply thermal paste with single pea-size dot instead of previous (oversized) + shape

Temps Idle: Mobo=48 CPU=36 CPU0=41 CPU1=41 CPU2=39 CPU3=39
Temps Load: Mobo=48 CPU=75 CPU0=82 CPU1=80 CPU2=78 CPU3=77


4. Turn off Hyper Threading: avg -0.5c idle -10.75c load

Temps Idle: Mobo=47 CPU=38 CPU0=41 CPU1=39 CPU2=42 CPU3=36
Temps Load: Mobo=47 CPU=65 CPU0=71 CPU1=69 CPU2=68 CPU3=65


5. Lower vCore from 1.30 to 1.275 : avg -1.25c idle -3.4c load

Temps Idle: Mobo=43 CPU=38 CPU0=40 CPU1=38 CPU2=40 CPU3=35
Temps Load: Mobo=45 CPU=64 CPU0=67 CPU1=64 CPU2=64 CPU3=61


Again, in brief:
- 7.0c with new fans push/pull & better thermal paste application
- 10.75c with HyperThreading OFF
- 3.4c with lower vCore
- 22.5c total

 

[Conumdrum]

Nice to see a brain at work! Good job, your happy we are happy.

 

[cjl]

It's true that turning HT off significantly drops temps. However, depending on what kind of applications you'll be running, it might be better to drop your OC just a hair and leave HT on. If you'll be doing rendering, video encoding, or similar highly threaded, processor limited applications, you'd be best off leaving HT on.

 

[magnosis]

I decided to leave HT off after reading this : http://ixbtlabs.com/articles3/cpu/ci7-turbo-ht-p1.html

Seems like it doesn't provide that much benefits for what I do (software development, photoshop, games).