OCUK Official IC Diamond/ Perihelion Test Results

[IC Diamond]

Any comparisons against the Zalman "paint on" stuff? I'm always up for something cooler, but I'm not sure if its worth sacrificing the sheer ease of application of the Zalman stuff for a degree or two?

Couple of comparisons on a giveaway done in December in Germany in post #3 page 1

 

[IC Diamond]

Reading the application method on the IC website suggests that the pea-sized blob method is the best method and that 5-5.5mm or half a cm of paste should be applied in the middle. I agree with that part.
However will this amount of thermal paste really fill the whole IHS of the cpu seeing as they come in different socket sizes and 1366 is bigger than 775? So 5mm may well fill the whole 775 cpu IHS but may not do the same thing for 1366 cpu.

Afaik the general consensus is that centre of the cpu is all that matters where most heat is concentrated. However IC are saying that even with blob method in the middle, it should cover the whole IHS otherwise air bubbles will form around the edges. How much do air bubbles on the edge affect the performance? I understand air bubbles would have mattered the most if they formed in the centre of cpu IHS. But air bubbles around the edge?

But air bubbles around the edge?


Left to right 1,2 4 started with small air bubbles - this halo of air bubbles at the edge expand under heat and pressure and accelerate compound failure

The example we use is for 30mm X 30mm just ratio it up or down for the area. We recommend to the high side to try and cover the variables encountered in end user contact with IHS and sink - concave IHS you need more compound, lapped IHS and sink you could use less so we try to cover both scenarios with the suggested amount.

Excess just squeezes out, I have tested this by applying excess compound between glass slides on a simulated IHS with a couple of binder clips and @ room temps it resolved down to about 95% of the average particle size in about 2 hours and 99% overnight of the final bond line thickness (BLT) measured with a micrometer.

Better to use to much than too little to avoid having to redo your mount, you have 10-12 applications in the samples provided don't be stingy you can refine amount as you get a feel for it you have enough to experiment with so have at it.

There are always exceptions or nuances, this week I had a guy who due his mounting set up did not allow for a blob in the middle as the mounting bracket design had to be locked in on one side first and as you pressed down to lock the other side the compound was pushed ahead of the closing joint leaving about a third of the IHS uncovered.

With the glass slides I tried X corner to corner and a + or a cross side to side and both ended with air in the joint. I ended up with a V pattern with apex on the initial load side as being the final solution.


Users are a great source for us on problem solving giglamesh @ overclockerstech came up with the the idea for subzero cooling to preheat/cure the compound with a hair dryer to thin the joint. Just going to straight to subzero the paste hardens before it has a chance to resolve down to the ultimate BLT or average particle size.

Another end user used a torq wrench on a screw mount and picked up some thing we had not considered.

When you first mount your sink you are not at the final bond line thickness, so when you power up the system the compound continues to flow - the screws do not move so you end up with a loose BLT. A re-torque after warm up improved his temp a couple of degrees. This is like re-torquing the head gasket on the old cars after the first 1,000 miles. I would not have picked that up as it is not an issue with a constant pressure spring mount which I use in all my testing.

I had one guy at Anandtech use a massage vibrator to spread the compound - worked pretty well.

After initial test I encourage people to experiment - you have plenty compound have at it.

Andrew

 

[IC Diamond]

I still haven't recieved mine

we ship first batch Monday or Tuesday

 

[IC Diamond]

I have Thermalright IFX-14 heatsink and it has two UNC 6-32 Spring loaded screws (which are about 1.5 inch in full length) placed on each end of the mounting bracket to tighten the heatsink on top of the cpu IHS.

So when using IC Diamond, considering torque; should I tighten them all the way down fully or partially and then turn on the computer for heating the thermal paste?

I am using the pea-sized blob method with MX-4 and tightened the screws all the way down to max.

Edit: Btw here is the complete mounting system. The black plate is the backplate. The heatsink base goes under the mounting plate held by the 2 screws which have to be tightened downwards

Run at load for 10 min? hour? let the compound heat up liquefy some and flow then you should get a half turn on the screws.

That was this guys experience

http://www.evga.com/forums/tm.aspx?m=681429&high=ic+diamond

 

[IC Diamond]

I'm prepared to believe the average difference achieved but I wonder how scientific your tests really are? Have the testers done multiple mounts and taken an average? Have they promised to do nothing else other than dismount, clean and then apply IC-whatever? Absolutely while they were there, not remove a single piece of dust?

1. Even collecting multiple readings the cluster size is = 1. An individual can collect all the data readings off one system, and will still have almost no statistical power (]in statistics this is known as "Intra-Sample Cluster Correlation" because the test set up is dominated by methodology. I can measure 20 different points simultaneously to within a 1/100 of a degree, watts within in 1/10, Airflow CFM within 5%, I can do multiple mounts and average the results. The problem is it is anecdotal and particular to my set up and puts me in the sample group of "1".

I would classify most PC hobbyists as a group as bright and pretty attentive to their mounts and you will get that percentage that break out the thermometers, torque wrenches, micrometers and go to town on it and you will have some percentage that slap it on, lick their finger and put put it to the wind and guess ambient to within 3 C.

So good mount, bad mount - too much compound, too little compound - good ambient reading, bad ambient reading - faster fan, slower fan -personal bias in favor for, personal bias against. So it all averages out The Central Limit Theorem & and the Error law pretty well encapsulates the issues you brought up just have to delve deeper. Statistics have been pretty well vetted for the last 600 years by the world greatest mathematicians, Fermat, Pascal, Bernoulli, Gauss, Laplace with probably Einstein legitimizing it for broad integration with Physics in his 1905 paper on Brownian Motion. Probably most of the products and parts in your house have had an engineer do a statistical analysis for acceptable reject rates study, your PC for example. Take an aspirin lately? Vetted through public trials analyzed with statistics - Have car insurance ? what you pay is determined by statistics Like the color on your shirt? Some marketing guy probably chose it as an acceptable color based on a statistical survey.

2. Absolutely - There are no absolutes. Take the classic problem of a group of astronomers, they all have the best equipment, well educated, competent in their field all measuring the same celestial event they will all have different numbers, their equipment may have slight variations, one may have an extra glass of wine at dinner and been modestly impaired, one took his reading at the wrong time etc. Like Bernoulli's example of shooting arrows at a target you can infer the location of the bullseye from the locations of the arrows. say with 50 arrows the inferred location has a confidence level of 95% sample a thousand and probability might be 2%


For many testers, because it's a test, they take extra care to ensure a good spread and a good mount, and this is likely to have a greater impact than the TIM.

Older tests, like this particularly famous one showed how little difference TIM makes with a proper mounting and two good surfaces to bridge.

Maybe I'm just an old cynic but if TIM can make a 7C difference, then I for one believe it was far more down to the user than the paste.

YMMV, of course.

Compound is a mechanical function, one of pressure. contact % , surface area, amount, of watts applied

In order to qualify a test result you have to quantify it as all thermal compounds have their own unique pressure thermal resistance curves some compounds do better at higher pressure and some do better at low pressures with lower viscosity @ even lower pressures difference tend to become homogenized so they are usually quite close in performance. Was the reviewer at 70 lbs or 30lbs pressure? 90% contact or 50% and how does that correlate to the manufacturers pressure/resistance curve? How does that compare to the Vegemite curve? does Vegemite crap out @ 50PSI? while the other compounds keep going?

The advantage to reviews that are formal or just some peoples experience on the forum, end users can and most do instinctively do what we are doing here and that's mentally averaging the results for comparison

This is a pretty simple survey/promo and has it's pros and con's but people get to try something new and they can either like it or hate it, but at least they get to try it and decide for themselves

 

[IC Diamond]

TO: IC Diamond.

I'm not having a go. I said that I believe your average. I cannot believe differences of 7C and suchlike can only be down to the TIM. Either the original TIM was no longer working properly, it was badly applied, the mount was poor or other environmental factors changed.

I'll stay out of this thread from now on.

-7C why not? Not everybody is @ stock speeds. The compound thermal resistance @ 60W is the same @ 120W so for example you have a 2.5-3.5 C delta at the lower value you will have 5-7C at the higher value.

Alternatively often when I am testing I will switch from a 1' inch die to a 1 centimeter die when I want more resolution and the delta temp difference approx is 2X at the same power levels 6C . So IHS to bare die you can easily do it with increased watt density on a smaller surface area.

GPU data which I keep separate from CPU data has been running much higher averages with the increased power, what are the new cards running over-clocked? 300 W+? I don't keep up with this stuff but 5- 10C is common with maybe an average of -6 C improvement. It's all in the watt density.

Frequency chart below results are representative of end users temperature results, if you wanted to take the time to compare watts vs Delta C you will find a pretty good correlation on a number of tests. increase the average users power and the curve moves right.

Granted you have some whose before result may have had a poor sink mount but correspondingly you have some % that had a poor after mount.

Granted you have some paste failures but at the same time there were many reasonably fresh applications.

We did statistically analyze AS5 for compound failure and is about 30% of all our sample comparisons to date for compound failure and found nothing significant, now we do not know if that is because I think you guys change your sinks more often then you change you underwear generally having a fresher application or perhaps reliability was good, AS is silent of the issue so not much more to draw on. some other compounds only lasted a couple of weeks.

I have dropped the outliers from the calculation but out of 652 tests they comprise just a couple percent and change the final average a few hundredths of a degree

I have considered Varying the routine some by perhaps doing a giveaway with 100 tubes of competitive compounds, say a 100 tubes of AS5 and 100 tubes of ICD on a head to head basis with fresh apps all the way around to fine tune comparisons. But I think it would not make much difference(see AS5 chart, a good sample size) considering the above, but I may be wrong

 

[IC Diamond]

14 end users results to date - any errors omissions on my part let me know.

Remember to take ambient temps & before results

 

[IC Diamond]

noted will make change , thanks

 

[IC Diamond]

I always appreciate the extra effort as it is a non trivial task to remount sinks and take systems apart and put them back together so much thanks on this end.

w3bbo - Noticed you high mounting pressure of 70lbs so cure time would be minimal - Max 1/2 hour, so you you were probably close to optimum bond line thickness. Generally from sample testing we have done 80% are over +50 psi so adequate to get a positive result and at around +50 psi for the average user we recommend 2 hours to let settle before benching.

Also noted comments on difficulty in testing - Being able to make a link pressure wise 70lbs +/- to the thermal result vs a test where the user sees maybe only a degree or two diff but unknowingly is @ 40 lbs or some other random number is huge. Thermal compound is a mechanical function of pressure/contact/power applied/ amount etc. Define the dominate mechanical features and it goes a long way to qualifying the thermal result. Excellent job.

I do not know whether you had lapped your set up but @ 70lbs most contact issues get resolved, not all, but high pressure is the next best thing to lapping and sometimes as good as.

Again thanks for the thorough effort and your time to do so.

tommytinkle - A major bonus to changing the interface material in a laptop such as mine - the gpu might live to see its third birthday as they are known to fail in these things. The gpu drop was most


impressive though what I must add here is that I changed the way it was mounted by bending the heatsink more and using the IC diamond instead of a thermal pad.

IC diamond excels in high stress thermal environments and is popular with the notebook and gaming people trending with re-pasting their systems to resolve thermal overheat/failure issues.

Some advantage in notebooks is probably evidenced by a failure in the previous compound. Fresh application to fresh application CPU probably is 5-8C depending on whether it is a stock compound or one of the more competitive TIM's. A number of end users experienced improvements of +30 C possibly/likely indicating severe compound failure and also as the MTBF for notebooks as somebody in the biz related to me recently is around 14 months(do not know whether it true or not but seems reasonable), life is reduced with high thermal stress.

GPU's Delta temps are typically much higher.

Notebooks are a PITA to take apart and nobody likes to do it more than once and if you are a novice you are less inspired to do it, here is where the longevity factor comes into play - longer life, fewer repairs, more stability under stress. Heat is the # 1 enemy of electronics.

Many complain about the compound's high viscosity most likely as it is just not what they are used to but it is an indicator of high reliability & durability (Quality) OEM's Spec this by recommending bulk loading above +90% for the simple reason High viscosity compounds (More solid) are harder to pump out than a low viscosity (more liquid) compound.

So if you are not changing your compound every week a high viscosity compound like IC Diamond or Shin Etsu would be the quality long term solution.

I like getting notebook samples outside the notebook community forums as I maybe get only 1 or 2 per overclock giveaway, exceptional effort for a sample test you got to be hardcore.

Thank you

 

[IC Diamond]

You mentioned you have very few laptop tests so.....
Compaq Presario V4000 ( pentium M 1.7 )

Load test was done using LinX for 30 mins, and idle temp was taken 15 mins after that.

Ambient temps in normal environments obviously fluctuate, so temps have been noted at max ambient


Before Compound:Akasa AK455 Been on about 3 months
Ambient Temp: 21.3 - 20.9
Idle Temp: 45
Load Temp: peaked at 65 with ambient @ 21.3 10 mins into test, but sat the majority @ 63-64

ICD - fresh application
Ambient Temp: 21.0 - 21.5
Idle Temp: 45
Load Temp: peaked at 64 with ambient @ 21.5 25 mins into test, but sat the majority @ 61-62

Will test my main machine during the week.

I only have a couple of other Akasa tests from the last German giveaway, seems to do pretty well although I would like to see 20-30 samples - it may be an underrated or unappreciated performance compound.

And thanks for the Laptop test

 

[IC Diamond]

Another one from me on my main PC i5 760 @ 3.8GHz
Custom watercooled.
CPU block is a D-Tek Fuzion MK1 with a standard mount kit. Tightened down to the stops as suggested by D-Tek. This if I remember rightly produces 40~ lbs of pressure

Idle temp was taken 1 hr after booting the PC
Load test was Prime95 Small ffts for 1 hr


Before Compound: Arctic cooling MX4 applied 10 weeks ago
Ambient Temp @ idle: 20.6
Water temp @ idle: 22.2
Idle Temp: 29-24-30-24

Ambient Temp @ load: 20.8
Water temp @ load: 24.3
Load Temp: 49-45-47-45


ICD - fresh application
Ambient Temp @ idle: 20.7
Water temp @ idle: 22.2
Idle Temp: 29-24-30-24

Ambient Temp @ load: 20.9
Water temp @ load: 24.2
Load Temp: 50-45-49-46

Seems the MX4 has the edge over the ICD in this instance. Maybe it needs time to cure or there's not enough mount pressure.
I'll re run the load test again in a few days or so to see if curing time helps any.

Try a re-torque on the screws

like re-torquing the head gasket on the old cars after the first 1,000 miles.

Compound spreads with heat and pressure so on a screw system if left at the first torque point paste will thin out but distance between sink and IHS remains the same so you would not see the optimal bond line

This guy got a couple more C after re tightening his screws after warm up and I have been thinking it might be a good Idea to make an observational request on this survey if it checks out with you people I would incorporate into the application procedure

i torqued all gpu screws to 180oz wait 5mins then re torqued to 220oz then 5mins then torqued to 415oz done

do to how thick this stuff is i think anyone that use it might want to let sit 10mins and torque them down a 2nd time
after 5mins with each step i got almost 2 full turns more on the screws the hold around gpu

Nice block - .025 C/W has to be close to material limits

If it is 40lbs it would be under our recommended pressure minimums

 

[IC Diamond]

D-Tek Fuzion MK1 - That's a bowed base?

 

[IC Diamond]

Nope.

Been touched up several times to remove marks/scratches etc. over the years. and never took any longer than a couple of mins lapping with 1200 grit to polish it up.

Why do you ask ?

Never mind I think I was thinking of the Swifttech

 

[IC Diamond]

Ongoing results - any errors or omissions on my part let me know

 

[IC Diamond]

*** HEADS UP ***

For the people running the test I would recommend using LinX. I have just done a run with Perihelion and noticed the load temps where 5'C better with Noctua NT H1. However after looking at the LinX results you can see that the Noctua NT H1 was infact causing thermal throttling with the time/GFLOP going all over the place. Anything from 31 GFLOP to 48 GFLOPs. Obviously using Primeor .... you would never know and Noctua NT H1 would come out on top.

I am going to let Perihelion bed in some more and re run the test. Interesting, very impressed.

I'm using a pea size as recommended and works very well.

It might be worth adding this to the first post otherwise results could be invalid.

Thanks this is something I would not have picked up on. Posted up on intro.

 

[IC Diamond]

Data Results Reporting Request.

Try to summarize your results into a final number you are comfortable with I always appreciate detail but the conundrum on this end is do I take results from test #5 posting #3 or test #2 posting #7 ? If you are running a cure cycle just report the final number +/- X degrees after X time is sufficient.
State that this is your final result, keep it simple.

 

[IC Diamond]

Is there a reason why ICD has to be so thick ?

The short explanation

OEM's Spec bulk loadings above 90+ % for reliability, so 92% diamond + 2 % carbon black puts us at 94% which is pretty much as high as you can go and still be user friendly.

We had one mix that was 10 W-mk but was not real user friendly on a viscosity basis and provided little thermal advantage so was shelved.

There are different kinds of thermal resistance. One is contact resistance which the retail compounds mostly rely on by being of higher liquid content so they flow real nice into the voids and air gaps and with great or low contact resistance so that out of the gate many users are thrilled with the performance. Instant gratification

The Achilles heel for these type low viscosity liquid thermal pastes is in the nature of their initial advantage. The major weakness of being liquid makes it susceptible to thermal cycling, The system heats up the joint compresses and a little of the liquid is pushed out and after some number of cycles the paste has shrunken in size due to the loss in liquid volume (dried out) leaving voids and causing a contact resistance failure negating their initial advantage, this is called Pump Out. A variant of compound failure Bake Out is due consistently high thermal stress which is more of a thermal degradation of the liquid but is similar enough so that bake out and pump out can pretty much used to describe either failure.

There are literally hundreds of technical papers on the subject and the issue is well understood in the thermal community.

Liquid is a necessary evil as far as thermal pastes are concerned otherwise you would be applying a powder. IC Diamond uses much less liquid so consequently is much less prone to failures as the liquid is wicked or baked away The shrinkage of the compound is almost non existent due to the low liquid volume content to begin with and so maintains contact/ thermal performance for extended periods and by design to set up into a crayon like consistency over time, still pliable and relying on the high bulk Diamond conductivity for performance as you are left with basically diamond held coherently together with the polymeric binders and is easily removed when re-liquefied with a solvent

So in short the reason it is thick is that it is harder to pump a solid than a liquid.

Depends on view point but end users looking for reliability, thickness of compound would be a positive feature to seek out and not one to complain about.

The supplied tubes have enough for 10=12 applications after you have done your initial tests and reported results for those where thickness is an issue experiment a little by applying ICD to the chip then mix in a solvent like acetone or IPA to thin it then use your preferred spreading technique, I believe it to be non optimal application but give it a shot if you want. OEM's add a solvent to ICD when silk-screening or rolling it on and would be more uniform but you could try. NOTE SOLVENT HAS TO COMPLETELY EVAPORATE BEFORE MOUNTING OR YOU WILL JUST BLOW VOIDS INTO THE COMPOUND AS THE SOLVENT HEATS UP UNDER PRESSURE ON MOUNT

From the designer viewpoint the more liquid retail compounds are the magicians trick, distract with one hand to hide what's happening with the other. You can put the processed cheese on the CPU and get a great number but do you want to close up case and forget about it for a year or two? I have seen examples of compounds that start to fail in a couple of hundred hours and completely poop the bed at 400 hrs.

Of the compounds I have tested I would say from my perspective Shin Etsu earns my respect for performance & reliability factors.

It's easy to design a compound for performance and it is also easy to design for reliability

It is very hard to do both in one package

 

[IC Diamond]

Good explanation ...thanks
Have you done any longevity tests with any of the other compounds.

http://innovationcooling.com/Reliabilty1.htm

Just an idea, as ICD is so thick, why not have a Page on " how to get the best out of your ICD " especially as very high mount pressure is essential. Not every bodies heat sink or water block is capable of attaining such pressures. Perhaps outline some of the better alternative methods of mounting to help get around the lack of pressure.

We provide a trouble shooting guide on our site with a lot of detail but the detail is stuff most of you end users know anyway we just state it a little differently.

If somebody posts a thermal compound problem on a forum the typical responses run something like "is your sink tight?", "How is your contact?" "maybe you need to lap" "try using more/less compound" etc.

Basically the suggested solutions are just good shop practice for all thermal compounds as the issues are most always mechanical (sometimes measurement calibrations as noted in recent posts) and one of contact and pressure.

So for ICD & ABCDE compounds if you want to get the most out of it just tighten down the sink and lap if necessary.

Intel specs something like +/- 10lbs tolerance from a base of 45 lbs I think so @ 35lbs you would be to light for ICD and probably a little light with other compounds also, @ 55 lbs all compounds will perform better but the more liquid compounds get closer to their optimum bond line thickness (BLT) at lighter loads than ICD so when you increase pressure they perform a little bit better. ICD being thicker and farther from optimum BLT improves more with added pressure than the compound which are near optimum to begin with.

LM have no particles and no real mechanical resistance and are candidates for people who are in the lightly loaded category probably hit optimum BLT @ 20lbs. But for some rough contours products like the metal pads may have a problem filling the gap but would work well with lapped surfaces. Just a thought while I am writing this - Do liquid metals Pump Out? Any thoughts? I think that maybe surface tension holds it in place, never read anything on it. just curious.

In any event from our sampling 80% are over 50 lbs and see an improvement in performance. I have studied the 20% that had no improvement and usually it's a contact and pressure problem. Special guides from users exist on all forums who have experience with particular equipment like shims or brackets can do a more thorough job advancing other end users performance gains than we ever could.

Also it is kind of up to the end user what effort he/they want to apply for better temps.

So what's funny about all this to me is from the first few giveaways I have made a significant effort on the 20%. Expended money on private lab tests, extensively detailed every nuance in an effort to move the curve over a degree, so now pushing 20 giveaway's the avg moved maybe a couple tenth's of a degree with all the educational effort. I can work 1 on 1 with an individual and improve results most times in a focused effort on a problem but as a group it is awkward and has not yielded much.

So maybe the final solution is a 5.5mm bead on center, tighten down the sink as much as you can, lap if your are motivated and have done with it.

 

[IC Diamond]

Strange I would have thought it would get better.

Me too -every paste I have tested drops as it resolves down to it's bond line and stabilize not reverse temp direction

I have had mounts blow on me when I disturb the compound - bump the sink or move it maybe the screw tightening?

 

[IC Diamond]

This is correct.

Guys if you are going to use IBT/Linx for testing the compounds, make sure your GFlops values are more or less same in all the tests otherwise your results will be invalid.

For example

MX-3 or any other thermal compound: 33-35GFlops in IBT

IC Diamond: 40-45GFlops in IBT

What will happen is that your temps will be higher in the case of IC Diamond as you are getting higher GFlops values even if IC Diamond is better than MX-3.
So you need to get GFlops values in a similar range for both compounds. So it should be e.g:

MX-3 or any other thermal compound: 43-45GFlops in IBT

IC Diamond: 43-45GFlops in IBT

Consistent/same GFlops values across all tests

Edit: I would request IC Diamond to place this in opening post aswell so people have idea on how to get accurate results when running Intel Burn Test/Linux .

Done & thanks for the info