Be wary of putting your trust in anything, even the test bench itself, when testing coolers. It is possible to make a mistake at any stage of the process. For the past three years, we've been testing our liquid cooler bench and our new air cooler bench.
We're ready to start testing now! Because we've gathered so much data from so many people, we can now figure out almost all of the flaws in our testing and can give you an accurate picture of how well CPU coolers work.
Some people aren't happy about this, but the good thing about it is that we can finally start collecting data in a way that makes sense. Here, we'll lay out a framework for our CPU cooler tests and help set a standard for quality and data accuracy.
Our promise to write a lot about CPU air coolers is coming true in the next month or two. Our first stop is the A500. After that, we'll look at coolers again.
In this article, we go over all the steps and processes for testing a CPU cooler in great detail. Using this document as a guide for years to come will ensure that our testing procedures stay the same and are correct. Trying to keep as many defects as possible to a minimum while still being real-world is the goal.
PC technology doesn't make it possible to completely eliminate variables and traps. To state "My Spire is at 70 degrees" as though it meant anything to anyone else is ridiculous. Unless you compare the cooling capabilities of several coolers in a controlled scenario, you aren't contributing anything meaningful to the conversation because there is no 3DMark temperature score.
During six months of testing, this content will show several months of finding flaws in testing, common and unusual errors, and incorrect data that could invalidate most evaluations without the reviewer ever seeing them. This content will also show how to fix these flaws. So why did this content take so long to show up? We spent months looking for them.
Many of these issues would be present in different reviewer setups, but the trick is to find out about bad testing by using the right methods. From the most basic to the most sophisticated, each of these topics will be thoroughly examined.
It turned out that there was so much bad information out there in our hurry to get to the basics that we should just write a long dissertation about everything.
Many years ago, we found all the places where bad data could happen in our millions of cells of data for CPU coolers, and now we know how to avoid them all. In the next few days, we'll put together a small list of coolers. One cooler at a time, we'll be putting them in over the coming weeks. Look for a lot of new and interesting things from us soon. Most of it will start after this week's factory tour series in Taiwan.
Almost all of our real-world and dummy heater test benches are now almost fully automated. A future blog will go into more detail about our dummy heater, but today we're going to talk about real-world testing hazards and how we've been able to avoid them, so stay tuned.
Even though increasing the pressure deviation and making other mistakes by the technicians could cause a lot of test errors, they aren't nearly as common as the first two. As a result, we've made software and spreadsheets that can tell us if any number we're looking at isn't in line with expected results or the standard deviation (or even a predetermined value).
As part of our automated testing process, we are going to do this. Even if the results are correct, they need to be checked out more before they can be used to make content. To make sure our data is correct, we collect hundreds of thousands of data points for each cooler and look at 45 final results (averages) for each one that we test. There are five averages sets of averages at the end of each of the four charts we're going to show. This means that these values are based on data from thousands of rows of steady-state data.
Dummy heaters are used for internal testing and are usually used to make sure that the data structure is correct. In the grand scheme of things, the numbers won't be the same, but the scale and orientation will be the same. People can quickly test new ideas, like scaling up power, without having to set a "stable" CPU level first.
To meet the needs of GN, we have an LGA115X-style heater and an AM4-style heater. There aren't many heaters like AM4. Resistors show that this one has an IO die and two core chips. This is similar to AMD's AM4 DT CPUs, which have three chiplets, but this one has two.
For example, we can use power from the second core chiplet to make it look like a 3800X. In this case, we only need to power up the IO die and the first core chiplet. People who support us on Patreon and buy our games are to blame for this. A $10,000 investment in benchtop power supplies, monitoring equipment, and special dummy heaters makes this possible.
With just advertising money, we wouldn't be able to keep putting a lot of money into our company and testing operations. Heaters can be set to a certain amount of heat load so that they can be used to mimic CPUs and remove all of the variables that come with a motherboard or computer's motherboard. This can happen.
Even though we still need to do real-world testing, dummy heaters remove any doubt about the accuracy and absence of a mistake by a technician or a test. For now, we'll keep the rest of the specifics under wraps.
