Table of Contents

• AutoTOC

Intro

Intel is finally trying to unbend its CPUs, despite having to be on a bender to buy a $630 285K right now. Today, we’re using our laser scanner to look at the deflection in the CPU heat spreader from the different loading mechanisms, including these scans of the 285K (read our review) and 245K with different coolers installed. Today’s testing also includes specialized pressure scanning to produce pseudocolor images of pressure distribution across the IHS surface, very brief thermal testing to look at the differences with Noctua’s LBC (Low Base Convexity) flat coldplate, and we’ll look at the mechanical aspects.

Editor's note: This was originally published on November 4, 2024 as a video. This content has been adapted to written format for this article and is unchanged from the original publication.

Credits

Test Lead, Host, Writing

Steve Burke

Testing, Host

Mike Gaglione

Camera, Video Editing

Vitalii Makhnovets

3D animation, Camera

Andrew Coleman

Writing, Web Editing

Jimmy Thang

Unfortunately, Intel’s new and better ILM is optional. It didn’t force motherboard manufacturers to use it, so they can still cut corners if they want to save a few pennies. The new ILM is called the RL-ILM, or Reduced Load ILM, with the old one being referred to as the “Standard” ILM (indicating an assumption of it being the default). Our Z890 Hero ships with the RL-ILM, as do most high-end boards, so we used it as a test platform to swap to other official LGA-1851 ILMs for comparison.

Let’s get into it.

Differences

We’ll get some basic education in and go over the differences:

CPU sockets are one part mechanical and one part electrical. Intel uses what is called an Independent Loading Mechanism for its socket. Some people include the ILM when referring to the socket. On a technicality, the literal socket is the Land Grid Array with the carrier that actually holds the CPU. The loading mechanism is the mechanical part of the socket.

Intel is shipping 2 types of ILM, RL-ILM and Standard, and it is using at least 3 different suppliers that we’re aware of to manufacture these. Our Z890 Hero came with an RL-ILM by Lotes, which is a long-time supplier of ILMs. We also have the same-brand ILM of the Standard variant, plus the two other suppliers you’ll find on boards.

RL-ILM vs. Standard ILM

Here’s a CAD render of the socket. The standard ILM has an angle that increases the force application along the edges of the CPU. That’s the real difference here. This is what causes the depression we’ve seen in previous 3D laser scans we performed. These scans are from our past content: You can see how the ILM causes significant bending and forms a central concavity with the heat spreader, leading flatter cooler coldplates to be worse on Intel despite being better on AMD. You can learn more about that in our previous coverage here and here. 

Back to the CAD model, the RL-ILM is basically just flat. This is the biggest change, as the force should be reduced. This is also why Intel requires a higher force heatsink to be installed in order to ensure contact.

The RL-ILM also has one other difference: There’s an additional adhesive spacer on the underside, which can be thought of as similar to the washer mod that Noctua now ships with its NH-D15 G2 coolers as an option. The additional spacer goes underneath the existing black spacer, meaning that the ILM "leg" component probably was taken from existing Standard ILM stock, then retrofitted with effectively a sticker.

3D Animation

In our original Thermal Grizzly contact frame benchmark, we showed how the ILM clamp appeared to apply very slightly higher pressure to one side of the socket. This was exaggerated by the fact that the ILM has some play in it, where it can shift side-to-side and be repositioned and we saw that still happens on the RL-ILM. 

Here’s our 3D render of the standard ILM: With the CPU dropped into the socket, the standard ILM uses a hook that’s attached to the lever to centrally press down on the ILM lid that clamps directly to the CPU IHS. With the lever fully down and secured, the ILM is now secured at 3 points: 2 on the bottom of the ILM and 1 at the top. All of this is the same on the new RL-ILM. 

As for the CPU, the ILM has two wings that press down on the IHS at the borders, and with that curvature we showed in the CAD model, the force application at these points is high enough that a highly precise gauge can show how light is able to shine through despite the CPU being relatively flat when unclamped and looking flat to the eye.

We’ll refer you to our Thermal Grizzly Contact Frame benchmark from 2022 to learn more about this older style of ILM.

For the new version, clamping the CPU in the socket functions mechanically identically for the end user, with the lever pulling down to hook under a securing latch and clamp the ILM at 3 points, with 2 main contact points at the wings of the IHS. However, the lack of a bend in the ILM reduces the load. Intel still has to keep the force high enough that the CPU’s pads make contact with the socket pins, but has to be careful that it’s the right amount.

Too much or too little force can cause boot issues and high clock memory stability.

And that’s really it for these ILMs.

Pressure Scans

Noctua just got done spending literal years developing its new NH-D15 G2 and shipped it with 3 different coldplates, which makes it a unique candidate for pressure testing. 

For pressure testing, we take the different ILMs and apply a special pressure paper between the CPU and the coldplate. We then take that and scan it in with a specialized pressure scanner to create pseudocolor images of the pressure distribution.

Pressure Scan Noctua Results - HBC on RL-ILM vs. Standard

Here are the results for the two ILM types with the HBC cooler.

The new Reduced Load ILM with the high base convexity Noctua coldplate yielded low pressure at the outer edges, but especially toward the top of the board near the VRM and EPS12V cables. The pressure centrally remained high; however, because the CPU should be flatter with this ILM, the Noctua cooler ends up with less evenly distributed pressure because it’s designed for a different scenario.

The standard ILM with HBC cooler scans reinforce this: The HBC cooler ends up with more evenly distributed pressure at the top and bottom edges of the CPU IHS.

Pressure Scan Noctua Results - HBC, LBC, Standard

And here’s only the RL-ILM with the 3 Noctua cooler cold plates.

The RL-ILM pressure distribution was the most evenly distributed with the standard and LBC solutions. The two are mostly indistinguishable for distribution, although the precise pressure centrally will influence the results in thermal testing.

The LBC cooler had slight gaps at the left and right edges, but consistently square distribution at the top and bottom corners, with good pressure across the entire center. The standard cooler had less consistent pressure at the top and bottom edges and similar gaps to LBC at the edges. Ultimately though, these two basically look the same for contact.

Laser Scan: Noctua Coldplates

Our Noctua NH-D15 G2 review went into depth with laser scans of the cooler’s coldplates, and that hasn’t changed. We’re showing the LBC, Standard, and HBC scans again briefly here just to help recap the impact because when we’re looking at pressure and how it is affected by the ILM, the cold plate is a part of that equation. 

And now we’ll scan the new Intel CPUs to see how their shape matches the pressure scans we saw earlier. 

285K & 245K Unsocketed Laser Scan

Here’s a look at the plain Intel 285K when it’s just flat in the laser scanner. The CPU isn’t in a socket at all here, so this is as simple as it gets. Even in our 2D screenshots of the 3D scan, we can see the letters -- the CPU IHS is so flat that the very slight indentation for the text is visible.

The IHS itself has a few higher points, one just off-center, one along the right edge when oriented in a legible orientation, and one just off the left edge of the CPU.

Magnifying it 100x, that coloration grows to form just a few high points. Overall, it’s flat, but at high magnification, some small deviations appear. One thing that is clear though is that there is no substrate curvature, which makes sense since it hasn’t been socketed.

Let’s create a grid with the 285K and add the 245K to it. The 245K (read our review) follows a similar pattern: Centrally, it’s a little higher, then just off-center right it’s also slightly higher at 1x. Adding 100x to the grid, there’s a similar pattern as with the 285K.

Finally, we added our unsocketed 12900KS from the golden sample coldplate story. It’s still flat when unsocketed, but the difference in IHS design is also slightly showing through.

Socketed Testing

And now, we’re going to throw this Z890 Hero with the new ILM into the scanner and socket the CPUs in it. We obtained this package of ILMs to test. The ASUS board uses the Lotes RL-ILM, so we’ll start with that one.

Standard vs. Low Pressure Socket - 3D

Here’s a one-to-one 3D visualization in Blender taking STL files from our laser scanner, showing the 285K with the standard ILM first. As usual, 1x magnification doesn’t show much, but bringing that to 100x quickly shows a deep concavity centrally, just like we saw with LGA 1700 CPUs.

Switching over to the new reduced load socket, we can see that the 1x to 100x magnification shows less of a pronounced curvature of the IHS itself. It’s still curved, but much less, with the CPU maintaining a more consistent height instead.

Socket Comparison - Grid (285K)

Here’s a grid comparison of the different ILMs on the same motherboard, tested with the same CPU -- starting with the 285K.

You can see that the Standard ILM at 100x magnification shows a huge deflection centrally, as we’ve seen before, with higher pressure on the far ends of the mechanism. While we can sort of see the slight ridgeline down the middle of the CPU, the bigger issue is how deeply it indents.

The reduced load socket is significantly flatter, with less of a central deflection. The ridgeline in the CPU IHS becomes more pronounced in the graphic because it is more consistently the highest thing in the image. Remember that this is at 100x magnification, so the differences are exaggerated intentionally.

Socket Comparison - Grid (245K)

Unveiling the 245K results in the grid, we see the same patterns: The standard ILM deeply indents the CPU centrally, deflecting and deforming it in a way that coldplates with matching convexity will cool it the best. The reduced load socket is flatter and more consistent, though is still slightly deflected centrally.

Laser Scan: ASUS Cooler

We need a laser scan of the cooler coldplate before moving to the pressure maps, as the cooler and IHS alike contribute to the pressure distribution. 

This laser scan shows the coldplate of the ASUS Ryujin liquid cooler, which is what Intel sent with its CPUs to reviewers. Other coolers would fit, but we wanted to test what Intel officially endorsed.

At a 1x scale, the ASUS Ryujin coldplate looks relatively flat, but still shows a protrusion dead-center, gradually reducing height towards the outer edges. When we did our in-depth testing on Intel performance with varying custom-made coldplates from Scythe, we found that this pattern often did well for Intel.

Scaling it 100x, we get this almost comical tower protruding from the coldplate. This helps us see the steep slope as ASUS applies massive pressure dead-center with its coldplate design. This is sort of a hamfisted approach and version of what Noctua did more precisely with the D15 G2 for LGA 1700, except Noctua had more nuance in the exact shape of the convexity, which will better align with the concavity in LGA 1700 CPU heat spreaders previously. It’s similar to what we saw in the $60 Thermalright liquid cooler, which managed to brute force its way in performance thanks to similar protrusions.

Pressure Testing Results

Time to look at some pressure scans of the ASUS cooler with the new Intel ILMs.

These images show a new pressure scan of our 14900KF with the ASUS Ryujin cooler and the old (or “standard”) ILM. In this scan, you can see the 14900KF has narrowing pressure on the left and right sides, with most of its pressure centered. That’s where it should be, and most of that is thanks to the comically protruding ASUS coldplate, but fuller coverage is ideal. The older IHS also is a little bit different shape than the new one. The second column represents the pre-installed reduced load ILM using the 285K. Looking at the third column, adding the standard ILM with the 285K, doesn’t look too different. The pressure profile appears to be distributed taller and narrower. There’s still some of that slimming effect going on when we get to the left and right sides but not nearly as pronounced as with the last gen IHS design and ILM. 

Ultimately, what we see is that ASUS’ older brute force approach gets a better pressure distribution on the prior LGA 1700 socket than on the new ARL 285K socket, which is thanks to the massive central protrusion. This is the approach Thermalright took with its $50-$60 liquid coolers previously as well. It’s relatively hamfisted, but works, whereas the more carefully shaped approach of air coolers like the D15 G2 and the Scythe FUMA 3 are technically a better pressure match; now, that said, a 360mm liquid cooler is still “better” (with regard to capability) overall, and it will cool better, but the Ryujin could improve with more purposeful coldplate shaping.

Thermal Test Setup

Thermal testing is up now. Full transparency that we’re keeping this really simple this time, mostly because it doesn’t take much testing to verify if there’s a difference at all.

We’re only running the comparison thermals with one cooler this time. The ASUS cooler is so heavily deflected that we’re not sure the comparison would be that useful, so instead, we approached it with what should be a worst case scenario: The Noctua NH-D15 G2 LBC, or low base convexity, which is the flattest of Noctua’s options. In theory, this should be the worst on the more deflected standard ILM+IHS combination and the best on the flatter IHS from the RL-ILM.

Other coolers could have more or less impact. Coolers with higher force application centrally and with more convexity would continue to compensate for design problems of the standard ILM, but we want to just run a quick evaluation on one of the uncompensated scenarios.

We are also not testing anything below the minimum spec Intel declares for the socket, which is a 35 lb. force from the cooler. Anything high-end that’d be paired with the current CPUs will meet or exceed this requirement anyway.

Thermal Results

Here are the results from a simple A/B test. For this testing, we did two full mounts and at least 3 passes to average the numbers. This allowed us to check for variance mount-to-mount. All our other CPU cooler standards and methodologies apply, like manually spreading paste, controlling the fan speeds, and fixing the voltages and frequency. We disabled all power and thermal limits and set a fixed voltage with fixed frequencies. We have a known power draw down the EPS12V and 24-pin ATX12V through the 4 phases that it has (without PCIe slot power). That allows us to get these numbers.

The result was 61.8 degrees delta T over ambient for average P-core temperature with the standard ILM and 59.6 dT with the RL-ILM, or the improved one, meaning about a 2.2-degree reduction when accounting for  ambient. Without the deltas, we were running the 285K in the 80s to low 90s because we disabled all TVB 70-degree throttle controls. Running the CPU hotter allows us to see more of a gap between the results. A CPU consuming less power with a stronger cooler would likely not show as big a gap.

Checking briefly with Der8auer as a peer review, we learned our results are roughly in-line with his own. The differences are aligned with cooler and heat load differences.

We observed a slightly lower core-to-core delta with the new ILM, but it was within error. The AVG all-core temperatures were not significantly different from the P-core temperatures in this one due to the proximity of the P-cores to E-cores in this architecture (combined with our adjustments in BIOS). 

So, as short as possible, the RL-ILM is about 2.2 degrees better at this heat load with this cooler.

Tutorial to Remove and install the ILM

Before we move to the conclusion, in case you buy a motherboard with a standard ILM and want to move from the high pressure standard one to the low-pressure RL-ILM, we’ll walk you through how to do that. 

If you are going to swap the socket, we recommend sticking with the same brand for the replacement if possible. In our case, we used Lotes. 

To begin, we recommend starting with the CPU installed to protect the pins underneath to mitigate the risk of dropping, say, a loose screw down into the socket. 

From there, unscrew the 4 screws. We used a regular T20 Torx screwdriver. 

Removing the screws frees the top and bottom pieces of the socket. It also frees the backplate. When you’re installing the backplate, it’s important to get the orientation right and to ensure that the plastic sticker side is touching the bottom of the motherboard and not the exposed metal side. The backplate also features a notch that aligns with the triangle that’s on the corner of the CPU.   

We found that it’s easier to install the lever arm piece first with its 2 screws. Once that’s in place, it’s time to secure the other side with its 2 screws. You don’t need a lot of torque for the screws. We recommend that you tighten them in a star pattern to evenly distribute pressure.  

Conclusion

We regularly see people online saying that some cooler is 3 degrees lower than some other cooler, so just a reminder here on how all of this works: That 2.2 degrees is specifically at the power load we tested and with the cooler we used, under the conditions we employed. It will be higher or lower based on how these parameters change.

As an example, we did one round of tests with all the Intel throttle controls still enabled and saw less than 1-degree of difference -- but that’s because it was just throttling itself to regulate the temperature.

The new RL-ILM is definitely an objective improvement in both the curvature of the IHS and substrate and of the temperature in our brief testing. The pressure distribution depends on the cooler more than anything and it isn’t always clearly better, but the thermal result tells us that the net result is positive.

Frustratingly, this is optional. Intel is not at a stage where it should be making clear, simple, easy improvements “optional” for motherboard vendors. 

Although we don’t want Intel or AMD to force certain lock-downs, like taking away overclocking features, we do think both companies should enforce a default or baseline configuration that is in the best interests of the consumer, with the option for the consumer to tweak as their motherboard allows once exiting default settings.

In this situation, we do think Intel should just bite the bullet and force the better solution. It may be a situation where board partners had already purchased millions of these older mechanisms. Regardless, Intel has at least improved its mechanism. It is technically slightly more expensive than the original ILM, but since we’re talking pennies, we’d like to see this forced in the next generation as the standard ILM since it is just better. Intel needs to stop taking a soft-handed approach to its partners and taking the small victories when it can get them.

This doesn’t kill the contact frame market, though: That’ll still provide uplift, as the RL-ILM remains a mid-step improvement without going full flat like the prior contact frames we’ve tested.

That’s it for this one. We probably won’t do a ton of Arrow Lake follow-up testing since it doesn’t make any sense to buy right now, but we may explore a few other features.