Posted by Kent Burgess | Aug 9, 2023 | Cases and Cooling | 3
Over the last few generations of CPUs and GPUs, there has been a disconcerting trend towards high power draw, and therefore die temperatures that push the boundaries of what many users are comfortable. This trend has led to a greater popularity of All in One liquid CPU coolers, and GPUs with monstrously oversized cooling solutions. We have also seen custom liquid cooling become more and more popular among people that previously would have been happy with over the counter solutions. 1 Inch Thick Copper Plate
More recently, as the latest AMD Ryzen processors are running at 95 C and the newest Intel 12th and 13th gen CPU are hitting 100 degrees (or more), we’ve started seeing more people looking into CPU lapping and CPU delidding. Not long ago these two (warranty voiding) procedures were viewed as something only the most extreme, some might say reckless, PC enthusiasts would even consider. For a few years, we have had some smaller manufacturers offering products for lapping, delidding, and direct die cooling.
The market has changed and now, in 2023, there are some of the major players in PC Cooling, like EK Waterblocks and Thermal Grizzly offering products designed to help more folks try to tame those temperatures. Of course when big names start jumping on a trend, there are also some inexpensive options (and sometimes knock offs) that can be found from sources like AliExpress. The goal for this article will be to take a scientific look at some of the various solutions that have been offered to lower temps on Intel 12th and 13th gen CPUs, and see what kind of improvement they truly offer.
Upon the release of 12th gen Intel, one of the first controversies to pop up was the fact that the stock CPU mounting bracket was exerting uneven force on the PCB. This issue was shown to cause poor contact with the cooling solution and, in some situations, the PCB would actually bend. A few DIY solutions were suggested here and there, then Der8auer and Thermal Grizzly introduced their 12th gen CPU Contact frame. This precisely machined, anodized aluminum frame replaced the entire Intel mounting mechanism, providing even mounting pressure around the entire IHS of the CPU, and many found much better temperatures utilizing this solution.
With Der8auer’s success, a staggering number of knockoffs quickly followed to the market at much lower prices. These were available at a much lower price from no name manufacturers, and even reputable cooling companies like Thermalright. While in most circumstances the performance of these copies essentially matched the Thermal Grizzly product, some people did report that their frame was not holding the CPU level, and they were still getting uneven contact from their cooling solution. The sample tested in this article is the Thermalright model.
I would like to state that, while this is the simplest of the aftermarket solutions tested for this article, and does not void the CPU warranty, the contact frame is not without its own risks. If the retaining Torx head screws are over tightened, the user can experience instability, the loss of communication with one or both RAM channels, and even motherboard damage is possible (though I’ve never heard of the latter ever happening.)
The recommended torque specification for the attachment of this, or any other CPU retention system for LGA 1700 is 0.03-0.06 Nm. There are some YouTube guides by both der8auer and Gamers Nexus with some excellent guidance for getting close to the correct torque spec by hand, but for this test I made use of a precision torque screwdriver set to 0.5 cN-m (0.05 Nm).
If you really want to void your CPU warranty, but you’ve not fully dedicated yourself to the risk associated with delidding, then lapping your IHS is the perfect way to start. Der8auer and Thermal Grizzly even have the product to help you lap your IHS correctly, so you come away with a flat surface. This should help your CPU make better contact with the cold plate on whatever cooling solution you are using.
This kit includes a CPU holder, and three levels of uniform grit sandpaper. You do need a contact frame to mount your CPU in the tool, and a flat surface for the sandpaper. For this test, I purchased Dead Flat Float Glass to insure the flattest possible work surface, but I do not think most users would need to go to such extremes. The Thermal Grizzly lapping tool is currently available for $21.99 for Intel 12th and 13th gen, and $23.99 for Ryzen 7000 series processors.
As the final intent for this article was to test a delidded i9-13900KS, it was necessary that I actually delidded the CPU. Previously I have used a couple of delid tools from RockitCool, and had positive experiences, but as EKWB had recently released their own 12th and 13th gen delid tool (designed in conjunction with Der8auer) I wanted to give their offering a try.
Overall, I was generally pleased. The design of the EK tool requires the CPU and top plate of the tool to be rotated several times, so it was more work than other tools I’ve used, but I do think this design leaves much less room for error. This method is a bit more tedious and takes longer than other solutions, but delidding is not a process that should be rushed. It is much better to take your time, do it right, and be certain you are not going to destroy your CPU. At $85.99 the EKWB solution is a bit more costly than most of the other options, but the extra cost is worth it for the peace of mind and quality of the final product.
As an aside, I do recommend using some form of conformal coating on any of the delidded CPU’s surface mounted devices. You don’t want to apply this too thickly as if the coating became taller than the silicon die from the PCB surface, the cold plate would not make correct contact.
Since I used the EKWB delid tool, it should come as no surprise that I also tested their direct die cooling solution. To be clear this is the EKWB direct die upgrade kit, being used to convert the standard LGA 1700 Velocity^2 CPU block to direct die. The kit consists of a new CPU retention frame that is part of a revised EKWB Exact Mount backplate, as well as a new cold plate and internal jet plate. If you already have the LGA 1700 Velocity2 the price of the upgrade kit is $124.99. The complete LGA 1700 Velocity^2 Direct Die block can be bought for $219.99.
Visually the EKWB kit is great. I have always really liked how the plexi top Velocity^2 block looks, and the Exact Mount backplate has the appearance of a really top notch mounting system. Unfortunately in practice I’ve always found the Exact Mount system to be a bigger pain and annoyance than just about any other CPU Block mounting system. It has to be attached to the block from the back of the motherboard, so you have to either lower the motherboard onto the block, or (as I’ve found to be easiest) turn the motherboard sideways with its edge on the work surface as you hold the Block in one hand and the backplate in the other.
The overly complicated mounting design makes it difficult to get mounting pressure even on the first try, even when using the specified EKWB torque specs of 0.6 Nm. I actually had to throw out my first round of tests with the EKWB Direct Die kit because I did not have equal mounting pressure and a couple of cores were over 90 C within seconds of beginning the Cinebench run. The results shown below are from my second mounting attempt, which yielded much better temperatures across the board.
There has been a bit of controversy surrounding the EKWB Direct Die block since it was introduced earlier this year. Several users complained of poor temperatures, and many claimed that contact plate was not flat, and therefore did not make full contact with the CPU die. EKWB addressed both issues, first by stating some of the early samples had exceeded margin for error in the production of the PCB retention bracket (which is a part of the backplate) and they revised their tolerances on later units. Secondly, EKWB’s statement is that the contact surface was machined on a lathe instead of milled, and while it might not look flat (due to the concentric rings visible from the lathe machining) they assured user it was flat, but they were revising production to produce a new plate that visually met what users were expecting to see.
The unit used in testing was ordered two months following EKWB’s statements, and appears to match photos of the revised units which I have seen online. Even with these revisions, there are still many online reports of users complaining of temperatures higher than expected with a direct die solution.
Now we get to the real reason which I originally began this project. With so many experiencing issues with a very expensive EKWB solution, many others have turned towards some less well known manufacturers in an effort to get their CPU temperatures under control. One of these options is the Iceman Cooling Direct Die block, ordered on AliExpress for $76.97 USD and received in 12 days. At the time of writing the price has increased a bit and it looks to list now for $87.04.
The Iceman block is a nickel plated, all copper solution. Iceman uses a nylon bracket which does the same job as a contact frame, replacing the stock CPU hold down. It was a super easy install, and I just used the same 0.5 Nm torque spec to mount it using the stock backplate. The biggest issue is insuring the liquid metal is applied in the correct place on the cold plate. While it may be a little tedious, I would recommend test fitting with standard thermal paste to show you where the die contacts the cold plate.
EKWB Velocity CPU Block and Arctic MX-6 thermal paste used in testing of stock Intel retention bracket, Thermalright Contact Frame, and Contact Frame with lapped IHS. Thermal Grizzly Conductonaut Liquid Metal used for both direct die configurations. Testing conducted with a controlled ambient temperature of 23 C. Load testing conducted by running a full 10-minute Cinebench R23 benchmark.
The first major surprise I noticed during testing was that there was almost no difference in temperatures between the stock retention bracket and the Thermalright contact frame. Many cores were the same temperature between the two, and the contact frame only showed a max improvement of 1o C on the other cores. This was very surprising to me as I have personally seen a benefit to these frames previously, and my esteemed editor Sebastian saw massive improvement with his 13900KS sample.
I was so surprised at how close the temps were in this test that I remounted the CPU and the contact frame and double checked the torque specs. My re-test results were essentially identical to the initial test. It is possible that this particular bracket might be one of the flawed examples mentioned earlier, but as it was the only example I had on hand, I have posted the results of the re-test.
With both the stock retention system and the contact frame I noticed a huge variance in core temperatures. Generally the hottest cores were 12-13 C higher than the coolest cores. Also, until delidding, cores #5 and #7 were always the hottest two, and cores #0 and #1 were the coolest. Lapping the IHS provided a measurable difference, although again not quite as much as expected. There was only a maximum benefit of 3 C on the hottest cores, but overall the majority of the cores stayed a few degrees cooler throughout the duration of the test.
Things get interesting with the EKWB Velocity2 Direct Die block. As mentioned before, my first test went wrong immediately and I knew something was wrong. After reapplying the Conductonaut and then remounting the block, the numbers looked very good. The hottest core was a full 13 C lower than stock and 9 C lower on the coolest core. Overall I was pretty satisfied with this result and felt that the annoyance of the EK mounting system may have been worth the hassle. I felt that way right up to testing the Iceman.
There is no need to beat around the bush here, the Iceman Cooling direct die block has ASTOUNDING performance. The package temperature was a full 11 C lower than with the EKWB block. That’s 24 C lower than stock under load! Only one core exceeded 80 C, and oddly enough, that was core #3, not #5. The hottest any other core hit was 76 C.
I did notice a couple of interesting things when checking out the all the data. All three results before delidding pulled over 360 watts maximum during their Cinebench loops. Despite the fact that all the motherboard settings were unchanged from before (I did double check this) the CPU drew less power after delidding. During their R23 loops, I only measured 356.49 watts while using the EKWB block and 350.41 with the Iceman mounted. Also, the Iceman results were the only where the package temp did not match the hottest core, but it is also the only result in which the hottest core was not #5. This tells me that the package temperature sensor is very close to core 5, or core 5 is the sensor used to report package temps.
Despite all that was said in my introduction, I would not generally recommend lapping or delidding to most end users – even those of the enthusiast variety. It may sound like I am just repeating the company line, but these processors (both Intel and AMD) are designed to operate at temperatures that would have sounded downright dangerous only a few years ago.
A few months ago Der8auer had a fantastic, two part video interview with an engineer at Intel. During this they went in depth on temperatures and why we really shouldn’t be concerned if our CPUs are hitting 100 C during Prime 95 or Cinebench. The first reason is that the workloads these benchmarks replicate are not realistic for most users. If you are gaming, you will never see loads and temperatures on your CPU like you will encounter during a run of R23. Second, as I just mentioned above, these chips are designed to run at, and withstand such temperatures. But if you really want to push your overclocks on a desktop or bench system, then delidding is going to give you the best shot at maximizing headroom (with the exception of exotic cooling like LN2).
With all of that out of the way, if this is the path you want to walk, I don’t see enough benefit in lapping to make it worth voiding your warranty. Delidding is a different story, both blocks showed significant improvement over stock, and even over lapping. The IceMan Cooling Direct Die block is simply incredible though. It is on another level from the EKWB Direct Die kit, or any other direct die solution I’ve used previously. Before this testing I would not have believed such a difference was possible between two direct die solutions. All but core #3 showed over a 20 C improvement from the stock configuration.
The original concept for this article was to examine the issues that some were experiencing with the EKWB direct die kit. It grew from there into this, rather extensive mass of numbers and charts. It was never intended to be a review of one particular product, but that changed with the results of the IceMan cooler block. I’ll give it the PC Perspective Editor’s Choice. It was a bargain at $77. It is still a bargain at $87, and will still be a bargain if it goes to $100.
The ICEMAN Cooling LGA 1700 direct die waterblock earns our highest recommendation
Enjoyed the article Kent. It’s always nice to get a look inside the mind of a masochist.
Great article. Your thoroughness is to be commended.
Hi Kent. Thank you for the review. Unfortunately, there is one big mistake. When comparing coolers, you should limit the core voltage and/or the power consumption. As you can see on your screenshots, the core voltage, the power consumption and the clock frequency differ from the Iceman to the EKWB. It is therefore impossible to say exactly which cooler performs better or is cooler. The EKWB with 6.0 GHz or the Iceman with 5.9 GHz? To be able to really compare these two coolers, the values must be fixed. However, the Iceman is a really good cooler at a good price. Best regards.
Your email address will not be published. Required fields are marked *
Copper plate This site uses Akismet to reduce spam. Learn how your comment data is processed.