We Made the Perfect CPU Cooler | Intel vs. AMD Curvature & Coldplate Engineering

Today we are using this $60,000 laser scanner to create the

perfect

CPU

cooler

cold plate with microscopic accuracy and precision. We had these CPU

cooler

cold plates custom

made

to match the

curvature

of the heatsink when under the clamping pressure of the Intel ilm and amds next. We use this chemically reactive pressure paper to create some scans of how the IHS makes contact with the cold plate and then we give it to Mike to do pseudocolor images and that all leads us to the thermal tests where we can finally evaluate a weak point long lasting on cooler views R which is the IHS style now the reason to test on a single IHS style is because they get a proper deep and accurate cooler review just for one CPU it can take 40 hours so which we have to choose carefully, but We want to evaluate this because, in theory, if you choose any of these four coolers that we have on the table, they may perform drastically different depending on which CPU you put it on, because despite them all being smoke 3, we have We modified them a lot, so that's what we're seeing today with science.

We brought you this video featuring our new Cyber ​​Skeleton V2 limited aluminum t-shirt in store. gam access.net, but you can also get one of our PC Building mod mats or highly heat resistant silica corn solders and project bats or other items in the store. These videos are part of our ongoing in-depth lab testing efforts we have been conducting. Working for over 10 years, but thanks to all of you who support us directly, we have been able to really improve the quality of testing lately. All new 3D animations you saw in our crossflow fan video and this one and in our Ultra Precision 3D laser scanner. along with our Hemiano camera are our way of reinvesting in the channel and publicly available

engineering

education for computer hardware.

If you would like to support our efforts to educate and inform, please head to the store. Players go to.net and grab one of our items. Skeleton Shirts will only be available for a limited time and were introduced as a way to support our EK research. We also recently introduced our copper plated Mule mugs with a suitable thermal conductivity laser formula. attached to the hammered surface thank you for your support of our vision for more scientific content and let's get back to that. We added Intel tests to our CB cooler reviews a couple of months ago.

You'll see it in some of the more recent tests. It's because I've been working on this, it's a lot longer, but we think it's worth it and by the way, many of you see this. Thank you, we can support that effort now, so here's some interesting data and we'll get For more information throughout this video, these are four colder plates that we use our 3D laser scanner to scan and observe how cold plates form and the way this came about is we were talking to the founder of Scythe and one of their Kagaoan engineers at Compex last year and the concept came up.

I asked if it is possible to make a gold sample cooler. Can you make the

perfect

cooler and fit it specifically to a platform or IHS? And he said, "Let's try it." So we asked them to make some different refrigerators, they didn't sponsor this. This is purely a scientific curiosity on our part from a methodological standpoint and theirs because they were basically trying to determine if it makes sense to produce something that is actually what we'll call the golden sample, they're magnified 100 times so it really shows. distorted to show the differences. This is one of the others.

This is a retail sample we purchased on Amazon. Much less of that gradient. It's much flatter, but. it's not as flat as these two, we

made

them custom because we used an am4 bench for our primary data set type and we just added

intel

again to try to accommodate these other IHS shapes, so having a custom flat cold plate made in In theory, they can play the system on an am4 platform, but it's also a valid way to try to cool it better, so this is a concept that came up before the cold plate or, for that matter, cooler manufacturers have. to choose if they want fine tuning for basically an IHS or they want to make something that works pretty well on everything that gets most of the vilification between the cold plate and the CPUs, it's actually going to be the CPUs, here's a modern one. int CPU with the film clamped and at 100x magnification you can see that at the points where the clamp applies force, the middle edges on the grass side, the CPU is clearly pushed down and deformed centrally.

You can see that in our animation. We did when we reviewed contact frames that are known to flatten a CPU. We've already shown that it's a known quantity, but what we didn't have were these fantastic 3D scans to understand exactly how it's deforming and that indentation. It manifests itself in the form of contact with a carbon plate, ideally a carbon plate would fit perfectly and connect, and the interesting thing is that an unplugged Intel CPU lying on a flat surface in the scanner is almost completely flat , even if it was mounted before this is also magnified to 100x and we still only see the usual microscopic imperfections but very little bending, this CPU has been used a lot so if it were permanently bent we would definitely see that clamp by now is warping it so the carbon plate can be adjusted with a scythe and there is custom test sample manufacturing today here is a look at an am4 CPU even at 100x and holds the CPU is almost perfectly flat, the old PGA socket won't nothing changes flat, so we're just showing this and there is a slight ripple on the surface, but that's a mix of magnification and the imperfect leveling process of the device being scanned.

You can consider this as functionally flat, meaning Intel and AMD am4 would have different cold plate

curvature

optimizations that a cooler manufacturer could do if they wanted to adjust it. However, am5 does have a movie and that's where this gets really interesting. The am5 pattern is the opposite of what we just saw with ints that you can. I see that it tilts up in the center at points perpendicular to the clamping direction. This is going to be problematic in terms of maximum refrigerator compatibility. This is actually the worst possible outcome because what happens is only to help people understand and map those 3D images to something. in real life basically on the AMD along this strip centrally and pointing towards the camera at these narrower points where there is no clamp what we have technically is a Bo up this comes outwards centrally and in

intel

ligence , if we come here it's basically the opposite, they are almost opposite each other perfectly where this one dips inward and follows that arc path towards these holding points.

Now most refrigerators don't actually set the cold plate to this level; There are some that haven't been talked about. We've done this in previous years, but the fact that the CPU indentations effectively intersect means that the ideal setup would be very CPU specific. In an ideal world, you would have one for the am5 CPUs, one for the current Intel, and then one for the am4, and you would need to maintain all of them. As a manufacturer and for a lot of boring reasons like supply and demand economies of scale and customer confusion, that's not really feasible, but it's still fun to explore the refrigerators we have today.

We have kind of a gold sample review, this is the one where we ask Scythe, hey, take out the best possible one you can find and try to tune it for that cold plate curvature and we basically ask what would happen if you tuned it for Intel from effectively, and then the next one we were looking at was flat samples, so we called them gold samples B and E, so they're flatter, they're on the screen behind me, they're those two on the left and then the last one is a ret sample we just bought on amazon, blind buy and that's just kind of like seeing, okay but what do you get if you just buy one off the shelf now?

The overall variation is extremely low if you have good testing processes that we do and because of the changes made to the carbon plate in each of these, any change outside of a difference of one degree would be considered outside of error and of course All of this content and in-depth testing is only possible because you all have supported us enough over the years with Patreon and in-store purchases for us to purchase that laser. scanner that allowed us to map only the cold plates here is a scan of a normal Scythe smokes 3 without mods when not zoomed in just from outside of Amazon it just looks a little flat you can see some colors indicating depth changes in the level from Micron, but to really see the curvature, we need to magnify it and this scan is enhanced with 100x magnification, revealing details invisible to the human eye.

You can see that the carbon plate curves towards the laser, if you imagine the laser is actually the CPU, it is convex and curved centrally towards the IHS of the CPU, however as it is magnified at 100x it is so slight that it could have no real impact on contact pressure distribution and plate flexing occurs during an installation so flattening is possible once we actually secure it. To the IHS either way, the retail vuma 3 is very curved compared to our custom am4 samples where we request a flat surface. This is a gold sample and as you can see, it is incredibly flat in contrast.

It's actually impressive how flat they got this board. Should work fine for am4. This is sample B, which is also very flat overall. They basically made two because they weren't sure how good they could get it and finally here's the actual gold sample unit where this one should be awesome for Intel because the notch lines up with the concavity in the IHS when pressure is applied. the movie, this will be the one to pay attention to in Intel's testing. The scientific process involves revealing limitations, so we have some for this test in the first place that we can't control. things like the heat pipes or the fin stack, the density of the fin stack, uh, the actual design, it's all the same from the unit, it's the same product, but technically speaking, the heat pipes can have a kind of performance in which they can be better or worse, but Scythe has a control.

The process for selecting your heat pipes is actually another company that has told us in the past, they do a lot of selecting and sampling for the heat pipes, so we should be as controlled as possible, the fans are another variable, but We chose to remove them the way we did, we used the same set of two fans, we put them in the same order, so the front fan is always the front center fan, always the same RPM of the middle and all that is validated to control and care for it. pretty good, now we need a hypothesis first, the laser scans, this is how everything should work, this gold sample cooler should work best on both AMD am4 and Intel LGA 1700.

The reason is because it is a clever combination of Scythe

engineering

in the adjustment process. This cold plate when we ask them to curve it convexly to mate with the Intel CPU, we both realize that this curve could be shifted slightly to tilt it towards the hottest part of the Intel CPU and still sit right between the majority of the times. of the silicon in the AMD CPU like the IOD, that means that while it fits perfectly with the LJ 1700, even though it is not flat, we can apply the most pressure to the chiplet area of ​​the flat am4 CPU, in theory, this does the best possible for each one. but it would be suboptimal for M 5 due to the perpendicularly intersecting curve.

The coolers in sample e and sample B are flatter, so they should theoretically run better on am4 than on Intel. Sample e is the better of these two. The flatness means that Intel if we exaggerate its curvature will end up with reduced pressure in the center where the die is, while AMD will end up with pressure evenly distributed throughout the IHS, but less than the gold sample at the chiplet location. Here are Intel's thermal results where we theorized the gold sample would perform better. Intel and it does by a long shot, this is really interesting data because everything about the way the cold plate was formed for this one says it should almost perfectly match that indentation at Intel, the end result improves almost 3° over one of the flattest. boards, so it's no surprise that Scythe asked us last year if we thought they should sell AMD and Intel coolers separately and it's no surprise that Noctua has done so sparingly in the past. 2 degreesadditions in the world of air coolers is the difference between the best. and in the middle of the graph, which is huge, the core also stayed similar, so we are still limited there by the IHS design, the silicone adhesive and solder sample B has a very slight curvature and as a result , outperforms sample e here.

Intel, so this is super exciting, let's look at the flattest CPU we have AMD4, the range from top to bottom is 1.8 of, which is relatively tight but definitely outside the typical AR grade of 1 grade. Objective for review in this case with three specially adjusted samples that ends a retail unit. In the case above, the gold sample we ordered from coputex had a performance of 58.2 De, which is approximately a 1 degree difference from our typical acceptable margin of error when measured against the retail unit and part That's a run-to-run variant, but there's a real difference. here this shows that the gold sample we made still works optimally on am4 as a result of the highest pressure point being the chiplet area, while on Intel it is the coupling that benefits the design sample and it is functionally the same because of how flat its performance is it is tied for the best performer on am4 of these samples, which is a big change from Intel's result, where e actually had the worst performance by 3°, so the theory is proving to be real, that's great, sample B was not. so flat that it didn't work so well here, despite being mostly flat it had a slight unbiased curvature that didn't center on the chiplet which worked against it on am4 but helped it slightly on intel.

We also previously ran some quick tests of the am5 using a 7950 Flatter was better than a curve that crossed in the wrong direction, sample B was also up there, followed by the gold sample. unsurprisingly the gold sample unit lost rank with am5 otherwise they only do it for all platforms unfortunately the retail price is low for all of these its ok since realistically we are talking about a couple of degrees, but it seems like they could be better. way to do this perhaps this is an indication that scy should consider adopting one of these experiment designs.

In fact, here's a quick table to rank them on each test. You can see that the retail unit was the worst in two tests and basically tied for that ranking. on LGA 1700 there is now a downside to the way we did these tests, as we asked for finely tuned coolers we may have also crossed an area of ​​sampling bias; otherwise, for example, maybe because we bought the retail unit 6 months after the custom ones were released. did something change either way, generally speaking the gold sample did better in two places and was only slightly disadvantaged in am5.

This seems to be the best for the future in general. We'll regroup with Scythe later this month at Computex to talk about the findings. and see if they can apply them to your manufacturing, so check back in about 3-4 weeks. The next section is great. Here is a pressure scan of an am4 CPU. The far left scan of the gold sample shows extremely high pressure in the center. in the center of the CPU, if you look at the flatter samples B and E, you can see that we have a wide area of ​​coverage across the entire surface of the CPU, but a reduced pressure in the center, so a higher pressure was most important in our thermal tests for this retailer.

The sample has a mix of higher dispersion along the IHS and higher central pressure, but it has some irregular weak spots and here is the Intel LGA 1700 CPU. In this you can see that the gold sample unit has extremely high pressure centrally and the two flat samples do exactly what we expected: they lost contact along the critical edges of the Central Channel and reduced pressure centrally with only point areas of high pressure. For us, this is actually the most exciting part of the job because we get to test a concept and then see how it works. it works and not all aspects worked as we expected, but in this particular aspect it is exactly what you would expect from a flat plate on a CPU that the super cool film curves inwards.

The retail unit is more similar to the gold sample. with its less pronounced curve resulting in slightly reduced contact along the top edge Finally, just for some physical pictures of the actual product, we took a bunch of pictures of the coolers with a high precision engineer's square and a flashlight, just We'll show some of that now you can see in some situations where the light actually cuts below the square, which illustrates the amount of curvature in the board, but as we've learned now, curvature can be a powerful tool to improve performance in some CPU by focusing pressure on key areas. or when matching the IHS, there were others that we tried with the engineer's square that blocked most of the light and those were usually the custom plans that we had made, so as an example, that gold unit that we had made was perfect. ended up being the best performing one which was expected, what we didn't quite expect was that the flat units, despite having the most uniform pressure distribution at least in one case, didn't end up being the best performing ones for am4 now in another .

What did work was that we expected them to be worse for Intel for LGA 1700, because of that downward curvature in the center and the pressure scans, we saw exactly that, which is super interesting and exciting, but also discovery or something like that . perhaps emphasis on something that was perhaps already assumed but not demonstrated in our case, the fact that higher pressure contact across the chiplet served as a more valuable approach to cold plate design than having that uniform distribution all over the IHS because uh without trying it, either one seems like it might be a better approach to making a cold dish and you need that data to really understand how it works in reality and now we have cool stuff from a product perspective, basically the way they combine.

What it sounds like is that if you wanted to tune the cold plate for LGA 1700 and am4, you could as long as you skew that curvature as a manufacturer or designer towards where the chiplets are in the am4 and otherwise it will just be mostly. Nest inside the Intel indent, assuming you're using a film, whereas another productive way to do this would be a sort of AR approach where they say no, there's only one way to install this cooler, it's with a contact frame period, that's the only way there are no movies. available, so it would be another way to force an unknown contact patch on the CPU across multiple variations to really maximize this with am5.

Furthermore, you would basically enter an area where to have the best perfect contact patch without any compromise. Whatever it is, you're going to run at least a couple of different biases and that's a problem because it becomes harder to scale, requires more validation, creates more confusion for the customer, and could affect economies of scale in terms of pricing. because now there are different cold dishes. that are being manufactured, so if the company does not own its own factories and that can also start to become an issue in terms of production lot size. For us, we originally ran all of our testing for coolers on am4, we had to choose a platform.

When we picked it, it was the most popular, at least for our audience at the time, and each cooler takes about 40 hours to test on a single platform on two or three CPUs to get numbers that anyone can start to trust, in our opinions. Now we've detailed that process before, but it requires multiple mounts and passes per mount and each cooler requires about a week of test work for Mike and even through all that, due to some optimizations we've done internally, that's why we added Intel. Testing a couple of months ago so you've seen that on freezer 36 review liquid freezer 3 uh, Deep cool Mystique had that test and a couple others that we haven't posted yet that's why we added it and we like to do them in the past because it's just a public and transparent way of saying hey, here's a ton of research on how we should test things and we use that to drive our decision, which is obviously ultimately governed by time and cost and things.

Well, so we are trying to make adjustments to get the best reviews possible, even though these findings show that there may be a difference with the cold plate design, ultimately what we see is that there are a lot of nuances there and , to really know, I would have to perform tests. on every platform that's relevant, and of course there's some kind of realistic limitation on anyone's testing, not just ours, we already go a little overboard, but anyone else's, or you have to make do with something. Define and configure it in a way that is reliable. The most important thing is to add extras after that to try to account for different CPU types, that's a little extra if there's extra time to spare in the process and extra validation that can be spent on it, but we're trying to do it.

We try our best to cover it all and this is part of why there is such exciting and interesting stuff, it definitely shows why if time permits it makes sense to try both platforms but there will be a sacrifice somewhere so an example It would be possible that you end up with only high power tests or only medium power tests, but covering two IHSS, so there is that test matrix and you have to reduce it at some point, otherwise it will end up being easily 80+ hours to try a cooler, eh, but I think We found a pretty good balance here and we're excited to continue implementing more things with these findings.

Thanks for watching, as always it's fun to do these things. Subscribe to get more information. You can support us directly in the store. Gamers access.net or patreon.com Gamers access and see you next time