Separate names with a comma.
Discussion in 'Overclocking & Cooling' started by FrgMstr, Apr 7, 2009.
edit: removed by author. I won't respond to trolling.
Your methodology used in continuing this line of conversation is inappropriate.
Ivybat, knock it off before someone important with a big stick come along.
Intel themselves have documented that the DTS has an inherent error on the order of +/- 1 deg C and the output suffers from skew over the operation range as well. I am not going to go find the exact docs but it was at the IDC last year and an older Intel Tech Journal article .
Speaking only as myself:
Including Coretemp values is not mutually exclusive of using better testing methods when warranted. Both have value to the readers and it is at the editors discretion to use one method or the other or both as appropriate. Think cheese.
You clearly have a total lack of understanding about the issues you're talking about, but since I have some free time, I'll do you the favour of explaining exactly why everything you say is wrong.
Except it doesn't. Core Temp reads the temperature off of a sensor within the CPU, which is designed specifically and only to register a correct temperature value at the Tjmax temperature. For the CPU used in this review, the target Tjmax is 95C, meaning that the farther away from that number you get, the less accurate the numbers are. Intel's DTSes have a variance of about 10C in either direction down to about 50C, and below that they become saturated, meaning that any readings below 50C should not be considered as actually representing the real temperature. So when dealing with temperatures that aren't above 80C or so (and that might even be a generous allowance), a properly calibrated thermal probe will be vastly more accurate than the readings from a DTS that is not even designed to read accurately at those temperatures.
The presence of the thermal probe is constant across the entire testing procedure, so any error that might occur from it (which would be insignificant if there is any at all) would be the same for all the results. So relative to each other, all of the numbers are accurate.
Except that does not represent real-world heat loads. Frostytech does this, but as a result, many of their results actually are inconsistent with real-world testing (for example, on their charts the original Thermalright Ultra-120 scores better than the Ultra-120 Extreme, which is certainly not the case in reality). So that approach is much more flawed.
vengence covered this.
FBDs and Hooke's law have about fuck all to do with any of this.
That is true, but only if the amount of paste initially applied is insufficient to cover the entire surface. If the amount is sufficient, then it doesn't matter how much is applied, since any excess will be displaced by the clamping pressure of the mounting system. As long as enough paste is applied, it doesn't matter what the precise amount is.
Most of that is incoherent, but if I'm interpreting you correctly, you're concerned about excess paste affecting the thickness of the layer. However, I addressed that above. Any excess paste will be squeezed out the sides due to the mounting pressure.
The proof exists in Intel's documentation. I haven't the inclination to go and search it out for you at the moment, but feel free to do so yourself and be educated. All the information about the inaccuracy of the DTSes as I explained above comes straight from them.
That should be sufficient to end this discussion.
I actually have the link handy to what you are talking about.
Thermal and Mechanical Design Guidelines (INTEL)
You might should check Appendix D, Page 77 "Case(CPU casing) temperature reference metrology". I'll save you some time. It is where [H]ard got it's test methodology, none other than Intel it's self. Feel free to go fight with them.
I use both. I use non-silicone for long-term mountings. I use the silicone based for short term test bench stuff. It is easier to work with overall.
Jesus dude. I just have this to say to you. This article was not written to be up to your expectations and I would suggest you discount any and all work done at HardOCP and not visit again if you find issues with our testing methods.
The bottom line is this, we are not going to spend more resources on projects such as this. We are happy with the result and feel that they are solid. You may not, and we note your issues. Now until you PROVE us wrong, we are sticking with them.
I'm still confused as to why people are taking this all so personally, but let's check some things off:
1) The probe's presence. It's relative testing, good. End of discussion. My mistake here for confusing testing for real-world application. It's true, I was talking about how testing was relative in other places too, shouldn't have noted this as a problem.
2) CoreTemp. If Intel's documentation knocks it, good. End of discussion. The chart still doesn't belong as reasoning for this decision.
There's still thickness.
So, in this situation, FBDs and Hooke's Law have something to do with this. That is to say, the situation is described by basic physics.
For a larger than necessary amount of thermal paste, the spring pushes down with a diminishing force as its extension increases (approaching zero at rest, this is Hooke's law). This movement occurs, pushing excess paste out, until an equilibrium is reached. This equilibrium occurs when viscosity and the force of the spring are at peace with one another.
We remove all of the excess paste and view the scenario again: Now the paste is at a height "h" at equilibrium. We see that this h is greater than zero because, when we remove a cooler from the CPU, there is always leftover grease, not just grease in microscopic crevices.
Let's assume that the spring's force is non-zero for any position of the cooler such that the base's height is >= that of the IHS. In other words, the spring is never at rest with the cooler installed.
Now, if we remove any amount of thermal paste from the equilibrium, h decreases and the spring expands. The paste still covers the CPU because h>0. So, for any amount of paste for which 0 < height <= h, we note that the spring will still push the base to meet with the paste, but the height of the thermal paste will vary.
Summary: Equilibria occur somewhere in the middle. According to physics, the height of the thermal paste can vary between 0 and the equilibrium height.
I'm not the best at physics, so show me I'm wrong
I don't know where you got that idea. Nobody is taking this personally; we're just pointing out that you're wrong. It seems like you're actually the only one who's bringing personal feelings into this.
Wrong. FBDs are a way of visualizing rigid-body systems. Thermal paste is not a rigid material, so FBDs do not apply. Hooke's law is a way of calculating the force stored by the extension or compression of an elastic material, but thermal paste is not an elastic material so Hooke's law does not apply. What does apply to the paste is fluid dynamics.
Wrong. Spring-based mounting mechanisms use compression springs, so the amount of force applied by them increases the more they are screwed down. The system is kept in equilibrium by the reaction force applied by the backplate which acts through the screws.
Sort of. It is true that an equilibrium is reached though, and that depends on the viscosity of the paste and on the clamping force. However, the amount of paste initially applied does not affect anything, since that does not change the viscosity.
That makes no sense. The spring is obviously at rest, because it is not moving. Like I explained above, it is held in equilibrium by the backplate. And the force applied by the spring is nonzero for any position it is in such that its length is compressed by any amount.
The flaw there is that if you apply less than the critical amount of thermal paste to cover the whole heatspreader, when the cooler is clamped on, the paste will not spread out to cover the entire surface. It will only spread out as far as its viscosity will allow it. So it is not the height of the layer of paste that changes, it is the surface area that it covers.
edit: Zero82z's post is more detailed. See his.
I use Shin-Etsu X23 on my rig. It is not all that pricey if you get lucky and find good deals on Ebay for it.
I hear gel toothpaste is better than cheese.
Seriously though, interesting review. I may have to try some Shin-Etsu myself.
What I find offensive is that you're trying too hard to find a fault in a system that is pretty much, for all intents and purposes, about as accurate as it's going to get based on the methods used. I've been lurking on this site for the better part of 8 years, probably longer, and I've rarely had reason to question their methods... certainly haven't had a reason to question, without substance, like you are doing now.
It's like you don't even care to consider the work already done and want to spend your time splitting hairs instead of actually making an effort to do your own research and experiments.
In other words: "Not good enough. I could do better. But I won't."
Not that any of the guys here need someone else to say this for them... It comes down to this: What you're doing is pretty much a slap in the face and you should piss off. Such a slap in the face to them that I, and I'm sure others, felt it as well.
Thanks [H] for the excellent article, especially the inclusion of the compound-application comparison.
I doubt the application of any physics or mechanical engineering degree will provide you greater then 2 C performance improvement at load.
I mean, I don't like what he is saying either, but I'll defend his right to say it. [H] isn't some infallible lab and his tone and manner seem appropriate. Maybe a bit edgy, but nothing worth some of the words I've seen. I've been on [H] for some time, too, and I think that their testing policy is quite worthy to be scrutinized. And, with that scrutiny, I think it has stood. This "discussion" is actually kind of constructive as I've learned plenty of things.
I mentioned this earlier, but the use of LINPACK vs. P95 might offer a greater delta between the TIMs, thus making an easier judgment on these mostly close results. Any thoughts?
I'm going to chime in here despite my better judgement .
While the 'thermal paste testing procedure' and really , the whole 'thermal paste shootout' thread is good reading , it's all moot except for e-points.
Take a look at the test results.
The best interface material on that list only beat the cheese by a couple degree's at most.
I've done commercial/professional electronic installations for years.
1 degree of improvement ain't worth it........It's irrelevant in a commercial setup.If you need 1 degree , your thermal management setup is crap.period.....
I bought my MG silicone thermal paste for 9 bux , and It'll last me till the day I die.
Anyone still using Type 44 compound? Seems to be serving me well.
I agree for the most part. I've worked on application where you need to move 150 watts in under 10C from a small die area. In those cases, quality TIM is important. For computers, it's not that big of a deal.
Nice videos. Should be helpful to some.
That was pretty friggin awesome to say the least.
Those should be put in a new thread and stickied, thumbs up
Your setup lists a 9500GT, but that looks like a 7900GTX to me.
Excellent review, there are such wide opinions/experiences from different people on their preferred Thermal Paste it's nice to get some [H]ard figures on up-to-date hardware.
Thanks! Fixed. Thought I had caught all of those then/than typos. - Kyle
Shin-Etsu G-751 on sale for $2.99 with free shipping. Small 1.0g syringe, but I always lose it after using it anyway.
Going to chime in here on cleaning off old compound.
I have never found anything that works as well as goodl old charcoal lighter fluid, it will disolve everything from the old sticky pink pads that some heatsink manufactuers use, all the way up to AS5 and Shinestu.
Qtips and paper towels used for the wiping.
For $1.79 you get about a 30 year supply
I've always been a fan of the cross method for the compounds that don't spread so well, otherwise it is the dot method for me.
I ordered some X23-7783D, and after the results from the shootout, I am going to use the cross method on that.
Really the only ones I had trouble with is the spread method and methods that have multiple dots or whatever. I got a result similar to the video, wherever the multiple dots meet up the compound is way too thick.
Huh, never heard of or tried that, gonna give it a go when I build my next rig and re-seat my TR XP-90 on the old system.
Not BS, just a typo, I meant Socket 939 (the board was an AsRock 939Dual-SATA2 because I still had an AGP video card at the time). I was thinking AMD in my head and the first thing my fingers typed out was "Socket A". Accidents happen (especially doing two things at once), and I've fixed it in my original post. But the Ceramique does get sticky enough that I wound up yanking it out of the socket with the heatsink.
I am not the only one this has happened to if you Google it. Example:
It performs well enough though.
Just make sure to clean it off with isopropyl alcohol afterward because lighter fluid will leave a residue.
Possibly a flammable residue!
Great article BTW, good 'ol AS5 never done me wrong. I will surely give some of the others a try for applications where I myself will be unable to re-seat the CPU after a year.
Thanks for this article guys
The cheese cracked me up. When I first looked at the picture I was thinking it was one of those Zalmann cushy chicklets or something.
Since every paste fell within 4 degrees Celsius I know that if I am in a pinch and need to pick up something available, that there is a good chance I can find one of these products. My AS5 went MIA on my last build; hopefully I can find it since it is still half full.
Your 1200 hours or so of constant testing is a lot of tedious work for my five minutes of reading but the conclusions resulting from the tests show that we have again only moved up slightly in better tempreature conduction in the past four years.
For my curiosity can someone answer me when this article was started? Last year sometime?
Kyle do you have any information about what is best to use at sub-ambient temps (for phase change, LN2, etc applications)? Thanks for doing real testing.
As far as I know it doesn't but I could be wrong, never had any trouble with it if it does leave something behind.
I say test without any paste and see how it does.
This was some [H]ardcore testing and that is why I have been coming back and reading the articles for years!
I will stick to using the pea method for applying thermal paste, as it is most convenient and quick without sacrificing performance in most cases.
Here's a quote from W1zzard
completely wrong way to measure this. they just looked at how intel suggests coolers should be qualified (qualify == keeps the system stable != scientific measurement)
only one application per thermal paste? or did i miss something?
a $60 multimeter to measure temperature? wtflol. that meter has an accuracy of +-1°C
"Ambient temperature will be kept at 25C for the duration of the tests and measured with a MicroTemp EXP non-contact infrared thermometer." - that meter has something like +-2° accuracy...
total that's +-3°C .. as a result those charts are pretty worthless since all pastes are within 3°C of each other.
great review. pick 6 pastes, roll dice, less work. if you have no idea how to review something, stay away from it.
He's clearly a software guy, not a hardware guy and doesn't even understand how a thermal couple works.... Just another idiot who thinks because he's good at one thing he suddenly has knowledge of everything.
He dosen't review hardware for TPU? How is he wrong? His points are valid, a little harsh but valid.
His points are really not valid at all. How is he wrong? Read below.
Intel's methods are pretty damn scientific out of pure necessity. If they weren't, they'd be fairly useless as a means of doing anything. They make use of thermocouples with good accuracy, a consistent testing protocol, and a standard of measurement that can be carried across their entire range of products while retaining its usefulness. I don't see a reason why this methodology would not be sufficient, and it is certainly much more effective than any other method that's been tried.
That's true, but realistically it doesn't make a whole lot of difference.
+/- 1C compared to the DTSes built into the CPU that have an accuracy of about +/- 10C? I'd say that's pretty damn good. Considering the fact that we're dealing with temperatures in the 35-45C range at load, that's an uncertainty of roughly 2.5%, which would be considered excellent in most cases.
An 8% uncertainty at most. Still quite acceptable.
Two things: first, the mere fact that all pastes have similar results technically means that the entire idea of a comparison such as this is fairly pointless since all of the products have relatively similar performance to begin with. However, the real benefit of this test is not to show which pastes are necessarily better, but rather to show that none of them offer any major advantages over the others.
The second thing is that although there is uncertainty, the testing was performed with a high degree of redundancy, and the averaging out of all the obtained values did greatly reduce the impact of the uncertainty on the final results. Therefore, the numbers that were obtained in the end can be considered to represent the reality with a good amount of accuracy.
This guy should take his own advice. It's best not to preach about things you have no understanding of.
I buy AS5 and this article tells me that I can keep buying AS5, as everything else is the same. I just prefer AS5, cause it's easy to apply for me. Not to mention the big tube I bought has lasted me 4 years.
His points aren't valid. A multi-meter doesn't have an accuracy of +/-1 degree. For a type N thermocouple has a sensitivity of 41 µV/°C. This means if you have a multimeter with sensitivity of at least 4 µV (You can buy a multimeter in that price range with 1 µV/°C) then your measurements are accurate to 0.1C.
The temp gun has an absolute accuracy of +/- 1 degree C, not +/- two degrees. Furthermore it has a resolution of +/1 0.1 degree C. (Yes I understand what resolution and accuracy are.) Such guns have a bias in relation to temperature which widens the absolute accuracy. In other words, at 100C it may tell you a temp thats 0.5 high, at 20C it'll tell you a temp that is 0.4 low, but at 99C it will tell you a temp that's 0.5 high and at 21C it'll still tell you at temp that is 0.4 low. As long as youre measuring something like room temp with it, the absolute accuracy isn't an issue. This is CLEARLY shown in the curing graph. If there was really a +/-1 error in the gun that graph would have looked like random white noise instead of clear trends.
So no, I don't think his points are valid. I also don't think he knows what he's talking about.
anyone know where to buy the Shin Etsu X23-7783D for around the $10-15 price stated?
i've only found like two sites who have them, crazy-pcs being one of them, and with shipping the cost of it goes into the 20s and close 30s