|
Page 8 of 14
Testing Methodology
Let's start this section by defining Newton's law of cooling: which asserts that the rate of heat loss of a body is proportional to the difference in temperatures between the body and its surroundings, or environment. Science (and fact) has a strange way of eluding the average enthusiast, so if you didn't understand that definition Benchmark Reviews will summarize Newton's law and apply it to computer processors: the cooling rate of a heatsink is equal to the difference in temperature between that heatsink and the ambient surrounding temperature. Therefore, a heatsink with a 60°C temperature will cool faster (and with better performance) in a 20°C room than a heatsink of the same temperature in a 30°C room.
One of the toughest parts of our benchmarking process was maintaining a temperature-controlled test environment, which is critical to comparing results according to Newton's Law of Cooling. To establish a stable and consistent ambient room temperature, all of the testing hardware was partitioned off and the room was sealed. While we did our best to operate at exactly the same testing temperature for each product, there were very small fluctuations which resulted in fractional differences inside of one degree. As we progressed into the later stages of testing, the ambient temperatures were much tighter and more controlled because testing could be completed in one day instead of weeks or months.
Probably the most critical step in the testing process was the preparation. Thanks to the team at Arctic Silver, we were able to clean and prepare the surface of our test heatsinks and Intel Q9450 Core 2 Quad IHS (Integrated Heat Spreader) to a pristine state with their ArctiClean products. While these products are praised by product analysts like myself, they can be equally as useful for anyone who frequently separates the cooler from the CPU or frequently applies TIM material such as extreme overclockers and hardware enthusiasts.
Test System
-
-
Processor: Intel Q9450 Core 2 Quad 2.66 GHz (overclocked to 3.2 GHz)
-
CPU Cooler: Reference Intel Thermal Management System (comes with retail Q9450)
-
-
-
-
Enclosure: HighSpeed PC Large Tech Station (open environment)
-
-
-
-
Operating System: Windows XP Professional SP-3 (optimized to 16 processes at idle)
Procedure Notes
The Gigabyte GA-EP45T-EXTREME offers a waterblock cooler on the P45 Northbridge, which we connected into a water-cooling radiator through the ZOTAC GeForce 9800 GTX+ Zone Edition video card. This enabled our tests to isolate CPU temperatures, and yield results without residual heat interference from other hardware components. The video card and motherboard Northbridge consistently operated at or near ambient room temperature, which dramatically reduces the temperature a CPU would report under normal circumstances.
Benchmark Reviews would rate the contact pressure between the Intel Q9550 quad-core processor's integrated heat spreader (IHS) and the Intel push-pin thermal management system to have been moderately firm. The surface condition of the processor's nickel-plated copper IHS was smooth with light texture visible, while the CPU cooler featured a polished copper surface. The use of a standard-performance cooler with mounting clip system and unpolished IHS finish was intentional, as improvements would narrow the range of thermal performance. By testing with a mild yet fundamentally sound configuration, recorded temperatures were not as closely staged between products.
Several stock Intel thermal cooling units were availble during our test period, which were discarded whenever a material appeared to permanently contaminate the surface (such as CooLaboratory's liquid metal products). While using the exact same unit throughout all tests would help reduce potential test result differences, Benchmark Reviews stands by our decision to rotate these parts out of the test configuration whenever necessary.
Benchmark Reviews used the following hardware parameters for all of our test equipment:
-
Intel Speed Step Technology was disabled. This allows the processor to operate at full power at all times, and does not reduce power or heat output.
-
All fan speed control mechanisms were disabled. The CPU cooler fan operated at 100% full power, and did not adjust speed to match the temperature or workload.
-
The Gigabyte GA-EP45T-EXTREME was installed into a HighSpeed PC Large Tech Station with an open-air environment. This eliminates heat-rise as an influencing factor in our temperature results.
Cure Time
It is the position of this author that when Benchmark Reviews compares products, they should be tested using identical methods. However, there are a few Thermal Interface Material products which recommend special application or added cure time. Although only a couple of the items we tested suggest or mention special application needs (either on the product or at the manufacturer website), we understand that certain interface products will exhibit improved thermal conductivity if they are allowed several temperate cycles to expand and contract the material.
For this article, each and every single product tested received the curing time recommended (see below), or approximately one hour of thermal cycling prior to testing when no cure time was specified.
(0) No Curing Time or Special Application Suggested
At the start of each test, the ambient room temperature was measured to track any fluctuation throughout the testing period. Lavalys EVEREST Ultimate Edition was utilized to create maximum processor loading and measure each individual CPU core temperature. It's important to note that software-based temperature readings reflect the core temperature output as reported by the processor through the motherboards BIOS. For this reason, it is critically important (for testers) to use the exact same software and BIOS versions throughout the entire test cycle, or the results will be incomparable. All of the units compared in our results were tested on the same motherboard using the same BIOS and software, with only the product itself changing in each test. These readings are neither absolute nor calibrated, since every BIOS is programmed differently. Nevertheless, all results are still comparable and relative to each products in our test bed.
One unfortunate problem is that CPU's report temperatures as a whole number and not in fractions. This in turn causes the motherboard BIOS and subsequent software applications such as EVEREST to also receive whole-number reports from the processors thermal diode. Thankfully, EVEREST also does offer averages in the statistics panel, which gives us more precise readings. To further compensate for this, our tests were conducted between five and nine times after complete power down thermal cycles. Conversely, the ambient room temperature levels were all recorded and accurate to one-tenth of a degree Celsius at the time of data collection.
When each cooler is tested, Benchmark Reviews makes certain to keep the hardware settings identical across the test platform. This enables us to clearly compare the performance of each product under identical conditions. While the ambient room temperature did fluctuate between 20~21°C, this would not be enough to cause a noticeable impact on our test results since only the thermal difference is reflected in the charts.
The Accuracy Myth
All modern processors incorporate an internal thermal diode that can be read by the motherboards' BIOS. While this diode and the motherboard are not calibrated and therefore may not display the actual true temperature, the error is constant. This means that if the diode reports 40°C when it is actually 43°C, then it will also report 60°C when it is truly 63°C. Since the design goal of a thermal solution is to keep the CPU core within allowable temperatures, a processor's internal diode is the most valid means of comparison between different heatsinks, or thermal compounds. The diode and motherboard may be incorrect by a small margin in relation to an actual calibrated temperature sensor, but they will be consistent in their margin of error every time.
|
Comments
This is one of the best review ever consider the amount of writing and scrutiny the writer had been through! I cannot give you enough compliment!
Makes me want to know more detail about how to prepare/polish the heatsink surface as it seems really important. Like exactly how you go about it and exactly what materials are used? I imagine you could make it worse rather than better by not doing it well?
Brilliant article, thanks again!
Haven't even seen the result charts yet, but just having the pictures of different ways to apply the paste is hugely helpful.
Thanks again!
Great article. Thanks for all the work.
If you're looking for a follow-up article - investigate why the machine OEMs use such a thick layer of ITM in their boxes when it's blazingly obvious that a thin layer works much better.
I would expect that even the ones that are without a curing recommendation would do better. That would be a great comparison.
Still, great review. It does however tell us that near the top, we are probably chasing diminishing returns here. I suspect that the 0.5 degrees could be within a margin of error.
found out on my test that any paste that require spread method. will give max/best result right away but over time the paste bleeds out of place, due to air bubbles. therefore decreasing performance.
i learned so much, and not just about TIM
Now I'm dying to see how the new ones like MX-3 go in the next test.
Now I just use a bit more TIM. I didn't mind lapping my old E4500 as there wasn't much at stake but I didn't lap my (current) O/Ced QX9560.
Now I just use a bit more TIM. I didn't mind lapping my old E4500 as there wasn't much at stake but I didn't lap my (current) O/Ced QX9650.
I am not talking about "Shin-Etsu MicroSi G751", I am talking about "Shin-Etsu MicroSi G765"
I happened to run out of Arctic Silver one night while assembling a build for a customer and I had a small tube of Radio Shacks white Tim. I used it on a a few builds and was surprised at the performance. The only thing that was worth mentioning stating is that it does dry out after 3-4 years and needs to be replaced IMHO.
Anyway, great review, later........................................
It would be grate if each product's electric conductivity was listed in the charts at the final results.
There is a slew of information in this article and it's definitely going into my bookmarks. Great job and many thanks for the extensive effort that was put into this project!
Helped me alot.
Or even a dumb (non-contaminating, non dangerous, cheap and available) poor mans product like toothpaste, or ??
Or a test showing the best product with poor application technique?
Seems like with all the work, a few more comparisons made to show the importance (or lack) of this.
Was repeatability also ensured for accuracy of claimed results? (eg, 30 tests later, do the same test with XYZ and ensure same results obtained)
By the way I had a ghetto thought about this If Money is no problem why not weld the heatsink base and CPU IHS...
Will love if this is done..
Sincerely,
/Antonio
You did a hell of a job
Keep the marvelous work
Funny how a 2 or 3 hour a day PC user could eyeing up the next upgrade path before as5 has cured. Any way I'm off to lap my sink
Try reading the article.
Must say though, that adding the viscosities of each thermal goop was a brilliant move as it's hard to find any real info on them elsewhere.
I would love to see some of the newer TIMs tested and added in a section at the end of this article as an addendum. I have been very interested in the metal TIMs I have seen around. Some people claim they are getting a 10c difference using a metal TIM over a TIM like AS5. I've never tried a metal TIM myself, and although I hear they are a little bit of a pain to clean off I would still like to try one.
As it stands this is still the best article around on TIMs, but it could use a little bit of an update.
Saying they nearly all perform the same may have been true 2-3 years ago, but not so much now. I would sure like to put some of those claims to the test, but since I can't get TIMs like Liquid Ultra locally I can't test them myself.
Frankly, until they are, no one can make any claims regarding them and be taken seriously, and that includes the naysayers. I have seen these claims on ocn. I can't counter them because I haven't tested these new TIMs.
Have you?
Then there are the people this article is aimed at, the enthusiasts who care that the difference between the poorer TIMs are better TIMs can be 5c. When you're overclocking, and aiming for a particular clock rate, then 5c can be pretty important.
Your question is akin to comparing the fuel in a street car to that in a race car where a few octane can make a difference.
AND when I read all of the comments that were written thanking you and praising you for your all of your hard well thought out work, Nice elements table too, Then there were the "when are you going to do a new one" Persons. I have to just shake my head sometimes in wonder at the different densitieses that are exhibiteded in such exchanges.
I have worked in metallurgicaliFielde ild much of my life and pretty well versed in such matters, had to laugh at the suggestion of soldering the processor to the cooler, I wonder do people actually take the chance on refinishing the surface of the processor? I can see flat lapping the cooler but I would never touch the processor, except to install.
Thank you for all of you efforts and I enjoyed all of it and agree with all of your findings.
Bay Blues
P.S I also put the antec 6 on my 5670 card voided the warrantee, but I more than likely extended its life its getting 32c vs 40c plus before.
When you put up 80 products agains each other the possibility of favoritism practically vanishes and the real picture begins to emerge.
I do agree with several here who have expressed wishes this test included results after long term use and curing of the TIM application(s). At the same time I must understand the time and cost required to do such an in depth test with so many products. All in all considered this is a great trade off.
In Short, I didn't think it mattered. It does! Buy it!
One thing I think could have really driven your important point home would have been to add a short table at the end, showing the "Before and After lapping" temp results using an A+ and a D- (perhaps thin and thick examples of each) on the illustrated Ultra-120 eXtreme heat sink, thus demonstrating the benefit obtained from prep, and the much lower importance of the paste used.
Also, as someone who laps metal pieces flat and polishes them to a mirror finish in my daily work, checking with an optical flat/monochromatic light, I'd like to emphasize to the masses that a FLAT surface is not the same as a SHINY one. Obviously, flat AND shiny is what we're after, but it's important not to "unflat" a surface in the pursuit of "ultrashiny", which is easy to do if not performed properly.
Finally, in answer to a question asked upstream to the effect of, "why would anyone not overclocking worry about this stuff?"-- those of us who are in quiet environments, particularly audio, like to keep our stock-volted machines as quiet as possible, which is partially accomplished by improving heat dissipation thus enabling lower, quieter, fan speeds.
Thanks again for the insane number of hours this (and previous articles) must have taken, including the excellent presentation.