Our Lapped CPU Heatsink Test Platform

Cooling | Reference|Recommended
Viewing page 2 of 4 pages. Previous 1 2 3 4 Next


After this issue with the Dark Pro 2 was put to rest, some questions about our testing platform began nagging me:

  1. Is the degree of concavity in our i7-965E normal?
  2. Was it there before we started using it as a heatsink test plaform? Did it become more concave as a result of the pressure it was subject to with umpteen heatsinks that have slightly convex bases and high pressure mounting hardware?
  3. Do other CPUs exhibit a similar degree of concavity?

Checking the flatness of the test CPU was relatively easy. We lined up its surface with several aluminum or steel straight edges in turn, and compared visually. As suspected, the i7-965 Extreme exhibited a small degree of concavity. We have no instruments capable of measuring such minute gaps, but suffice it to say there was a visible gap. Without a straight edge to compare closely, however, this concavity is hard to see with the naked eye. It's also difficult to get a photo of this gap or concavity without a really good macro lens, although we did managed to get photos of various heatsink base flatness in a review of a couple of Scythe heatsinks back in August 2010. Our point with those photos was that the best performing heatsinks have bases that are flat or slightly convex, never concave.

Now, we've known that most CPUs equipped with an IHS — all desktop models for a decade — are either flat or have slight concavity. We've discussed this in the forums and in many of our heatsink reviews (like the one mentioned above).

Question #1 isn't easy to answer; we'd need a dozen or so unused and used 1366 processors to compare. This is out of the question, as we have just two 1366 models in total, and it is impossible for us to accumulate the large number of new samples needed. Question #2 is also impossible to answer, as we did not record the degree of concavity (or flatness) on our CPU before it was put into service. Question #3 we could tackle somewhat, at least with the stable of CPUs on hand in the SPCR lab. So... after examining some 30 CPUs, most of them used with a heatsink for at least a few hours and a few that have never been used, we can say that a small degree of concavity seems fairly normal in most CPUs, but then again, a few of the never used CPUs seem to be just about perfectly flat.


I mentioned these issues to a contact for another heatsink brand. His response electrified us:

"Thermalright’s use of convex bases is well documented. Our product managers have seen coolers tested following a Thermalright that lost up to 2~3 degrees Celsius of performance. [sic] Buying a brand new CPU to redo the test will reveal the same pattern: A flat base CPU cooler tested before and after a Thermalright cooler will show slight loss of performance. As a result of this, we always make sure to test Thermalright coolers and others that are also convex last.

"However, there are exceptions such as CPUs with smaller die/heat spreader, which aren’t as easily deformed due to smaller surface area. If you look around the web, reviewers with older platforms that have bigger CPUs such as the LGA1366 you guys used that have gone through a lot of tests would likely all have concave CPUs. Consequently Thermalright coolers (and those that are convex) tend to dominate in these roundups."

Thermalright is named explicitly by this contact, but many other heatsinks have slightly convex bases as well. They include models from Prolimatech, Phanteks, Scythe and a few from Noctua. Some of these heatsinks feature mounting mechanisms capable of very high pressure. All, not surprisingly, have had models that performed very well on our 1366 test platform. Our LGA1366 CPU has taken them all on; the implication is that it must be more concave than it was when new.

I decided to query a few reviewers and heatsink manufacturers about this issue. My email to the select few:

"A sample supplier recently claimed that heatsinks that have a convex base will reshape the CPU surface slightly after mounting them. The implication is that it makes the IHS concave (or more concave) & makes cooling worse with flat base heatsinks. Have you ever experienced or heard of this? Have you considered it? What are your thoughts? I trust you use a stock non-lapped CPU for testing? If the claim is true, then it could be argued that all HS tested after tight mounting of a convex base heatsink must be biased from that point onward for convex bases... and HS with non-convex bases will perform worse than they would with a non-reshaped IHS CPU. It gives me a headache to think about how to resolve this if the claim is true."

Some failed to reply, and not all the answers were useful. Matt, a fellow hardware reviewer who's been at it for a long time, replied generously:

"Interesting topic. I’ve always enjoyed SPCR cooling articles, by the way.

"My first impression is that it seems like a pedestrian claim. Like someone just heard they are convex, and is making a derogatory claim off the cuff without any quantitative measurement or further supporting information. I’ve heard the convex claims before in the forum, though.

"So yes, some heatsinks have a slightly convex base but I haven’t seen it actually measured, likely because it would be a hard thing to accurately quantify - we’re talking millimeters and less here, and who knows what their manufacturing tolerances are. It wasn’t that long ago when we were looking at heatsink bases to feel how bad ridges were with our thumbnail, but at least smoothness seems to have improved over the years. I’m not sure I’d have much confidence in how convex the bases are from one sample to the next – I’m sure there would be some kind of range. Again though, I’ve never seen a reliable measurement on how convex a base is though. JoeC did some pressure testing, but not specifically looking at what we’re talking about here, and not in a controlled environment to really analyze this:

"Taking this one step further, does a convex HS possess the ability to warp the IHS? Maybe. Doubtful. And if we’re honest, I’d dump this into the same range as other things that are difficult to control or quantify… Like
tinting. Every time you remount a cooler on the IHS, you are reapplying thermal paste – sure we clean it in-between, and it may look visually as good as new, but on the microscopic level there is leftover paste/particulate. It could be argued that over time, the IHS interface improves due to better tinting of the IHS as the result of multiple paste applications over time. Or it could be argued that it gets worse, from impurities working their way into the interface and leftover cleaning material. Then there’s also the way the paste spreads – I’m not sure if you saw the article Joe C. did on this, but a while back he did some evaluations with multiple mountings and there’s a lot of variance. All the factors like this, if I get into them, will make my brain hurt on the amount of variance there could be from poorly observed and difficult to quantify factors involved in the testing we do on heatsinks… Yet our testing is fairly reliably reproducible. I think there are a lot of factors that exist, but just don’t matter much for the resolution at which we’re measuring.

"You may already be familiar, but if you want to test it you can look into pressure-x film from sensorprod.com. They give free samples, and if you pick out the right stuff for sampling (I can help if needed as I figured out which product version we’d need in the past), it will probably be enough to do a dozen or more tests without any cost. That should be the easiest way to quantify the convex nature of a given base and how it affects mounting pressure across multiple samples. You can analyze the results visually with a color scale they send you, or if its worth it to you, you can send the samples in and get some more scientific results ($$$). If you go the free route using samples, the only cost is time and their sales rep following up with you a dozen times or so."

Matt brought up many pertinent, relevant and informed points, but he lacked enough experience or fine measurement tools (like us), to provide real clarity on the particular issue of whether a convex base heatsink can deform a CPU IHS to a more concave shape.

Another heatsink rep confirmed the claim, however:

"Sorry to give you headaches, Mike, but yes, this is a real issue and we have reproduced it with heatsinks from Thermalright and other brands that use convex bases.

"It's true that most Thermalright heatsinks have convex bases, but this is extremely difficult to control in production, so they vary from almost flat to extremely convex. For obvious reasons, the issue is more likely to occur the more convex the base is and the higher the mounting pressure gets. Some coolers (from TR and other vendors) feature mounting mechanisms that apply pressure well beyond sane levels (or at least allow to do so). This way, you're also more likely to run into said issue.

"By our testing, the implication is exactly as you've said, namely that flat coolers will no longer produce good results on the CPU in question. Once it is deformed, a flat base heatsink will struggle to keep up with a convex base one even if it is thermally more efficient."

Both of the heatsink reps cited above are individuals I've known, worked with and respected for many years, for their technical knowhow as well as their personal integrity. I doubt greatly that they would feed me bad or misleading information deliberately.

Sifting through their words and considering my own experience with heatsink testing over the past dozen years, I came to accept that some level of IHS deformation probably occurred during our extensive testing with convex base heatsinks, especially the ones that provide a hefty screw over the center of the base to increase contact pressure. How much deformation is hard to assess, but knowing that the difference between a top performer and a middling performer is just a few degrees, my educated guess was that the drop in performance for a flat base heatsink (as a result of the extra IHS concavity) would be no more than about 2°C. (The second rep's comment, "...no longer produce good results...," seems like an exaggeration, especially against the comment of the first rep, "slight loss of performance.")

Now the issues were:

  1. Our CPU is probably more concave now than the typical CPU used by our audiences. Most people don't subject their CPUs to multiple torture tests with 30+ heatsinks over a 3+ year period. So if it is atypically concave, it might now be providing worse results for all heatsinks that don't have convex bases.
  2. How much worse, for a flat base heatsink?
  3. The question of how much worse the concave IHS CPU is making concave or wavy base heatsink perform doesn't matter; they're never going to perform well without at least a flat base.

Could question #2 be answered accurately? Yes, there was one way that it could: By testing some flat base heatsinks with our CPU as it was... and then again after lapping our CPU to a higher level of flatness, or at least so that it is no longer concave.

Previous 1 2 3 4 Next

Reference|Recommended - Article Index
Help support this site, buy from one of our affiliate retailers!