Gadgetory

we poured blood sweat and movie references into the creation of our 99 80 X II kingpin Edition CPU which used the expert overclockers sanding stones to grind down the CPU with near-perfect flatness this is useful for improving contact to ellentube pots under extreme cooling conditions but can also flatten the cooler and remove unnecessary material in a way that improves ambient cooling capabilities today's testing looks at that aspect of things trying to narrow down what kind of improvement a nearly perfectly laughte CPU can post before that this video is brought to you by be quiet and it's straight power 11 series power supplies the straight power 11 PSU is shipped from 450 Watts up to a thousand Watts accommodating most of the gaming PC build requirements you'd encounter and focuses on delivering a higher quality power supply that doesn't sacrifice on efficiency or stability noise is also a heavy point for the straight power 11 using a 135 millimeters silent wins three fan that can spin as low as 200 rpm for quieter low load operation learn more at the link in the description below this process we'll be discussing two things with the lapping of the IHS there's obviously an expectation that you potentially improve thermals and it's from two different aspects of the process one of them is you're reducing the actual material between the die and the ultimate cooling solution and ideally you want as few thermal interfaces as possible between the die and what's cooling it so in a perfect world what you do is direct eye contact or even more perfect you might just solder a cooling solution directly to the die but that's not how reality works so in reality what we end up with is you could do a direct die you would deal it it and then you get a really fine finely made direct eye kit that has tight tolerances so that you don't accidentally crack the die and then you put the cooling solution on top of that at that point you don't really need the liquid metal anymore because you're so close to the source of heat which is the silicon that has a high thermal conductivity about 200 watts per meter Kelvin that you can use a thin layer of pace and you'll be in good shape now the other thing here though is that you might expect an improvement in thermals from lapping a CPU yhs because has now potentially levelled out the thickness of the IHS or at least the concavity of the IHS so in our scenario what we were dealing with was an IHS that appeared from working with kingpin on sanding it to be very uneven in fact there were three really tall points and we have footage of that where the IHS just took significantly longer to sand down the rest of it and I just realized I haven't mentioned all this on the table coolermaster played a joke on us we'll talk about this later but anyway so that's one of the bigger things is that if you have high core - Kordell toes they're kind of two main theories as to why that might be and one of them is the IHS is imperfect to a point that the cooling material on one side is much thicker than on the other side of the IHS that create thermal gaps or Delta's that are large and then the the other theory is that internally under the IHS there's a thicker layer of solder than is truly necessary or something to that effect the real reason that people like kingpin laugh the IHS though is not to try and gain a couple of degrees or reduce temperature by a couple of degrees when you're cooling with ambience when you're doing liquid cooling or something like that it's just it's it's not what they're after so what kingpin is after is improving the contact between the ellentube pot and the IHS by flattening both of those surfaces so they contact each other pretty directly and then he might even scuff it up a bit to improve some contact area give some some extra grip for the thermal paste to grab on to and then the reason for that is to reduce the chance of the thermal paste cracking when it's under the liquid nitrogen where you're going to be in the negatives you might be minus 100 degrees Celsius or lower so that's what he's going for which is different from what a lot of people are trying to achieve when they lap the CPU for normal use and what we're testing is what's the improvement in normal use it's not going to be a ton we've done this before but now we're doing it with a more perfect approach by using the sanding discs that can been had in reality what we've done is prepare the CPU for better use with liquid nitrogen because we've flattened the surface so the contact between the the cooler the Ellen teapot and my chess is less likely to a gap in a way that causes the pace to crack and require reheating it with a heat gun or just removing the pot and putting new Allen to orden you paste on it for the next Allen to run so that's really what's happened but let's go through the test data and see if it's improved in any meaningful way for air cooling by which we mean liquid cooling but using ambient air so we're not using any extreme cooling solutions here just a crack in X 60 to 280 millimeter cooler and then we'll have additional test bench details in the article linked in the description below in our four point four get your Hertz 1.15 volt tests a light overclock that can be reasonably cooled on the X 62 we found our unmodified Intel I 999 ATX II CPU to perform at 86 degrees Celsius for package temperature with liquid temperature at 39 degrees and average core load temperature at 75 degrees Celsius now if that ambient is logged for all tests ambient was controlled to three tenths of a degree for these tests as a refresher deal it in a 79 80 X E and adding liquid metal got it down to 67 degrees Celsius core and so D 4 degrees Celsius package this is something we previously covered but the advantage to the 79 80 XE is twofold it's easier to deal it so cooling that is not sub ambience is improved with a reduced gap between the die my HS and it's also using a thinner interface of liquid metal in this instance there are downsides to but as shown in some of our b-roll of the deal editing process the 79 80 X II CPUs are the same as the 99 80 X YZ except that the IHS is secured with paste and silicone rather than indium solder so it's just silicone adhesive and thermal paste on that 7000 series once the silicone adhesive is removed the die to IHS gap is reduced and that's further improved when using two thin layers of liquid metal as the contact media the 99 ATX EE meanwhile has potentially more distance between the die and IHS as a result of the soldering process this is why we see such large improvements on the 79 80 XE we're reducing the distance and improving the interface and that's really the benefit of it back to the chart the 99 ATX e and the 79 ad XE really showed their gap when set to 4.6 gigahertz and 1.2 4 volts core which is an awful terrible configuration but one which we use to artificially stress the CPU under AVX loads that thermal Delta is noticeable here going to our kingpin tests we see a reduction from the 1990 DX e is eighty six point two degree baseline package temperature to eighty three point one degrees in the final like-for-like test we did some additional tests with newer thermal paste batches than the one we originally used for the pre lab testing finding performance falling within a single degree of variation for each one the package temperature improved by about two to three degrees Celsius depending on test versus the baseline benchmark this is a good improvement from just removing a layer of nickel and flattening the copper interface we emailed experts kingpin and ten about this and received confirmation that a two degree reduction is common for CPU lapping but we also received a lesson on the greater item of importance a flatter interface for liquid nitrogen cooling solutions is really the reason that these CPUs get lapped tin told us the following quote while we hoped for more significant results two degrees Celsius is quite usual for us the problem with non flat CPUs lies in a different route when you get too much grease filling voids between the flat CPU Allen to pot and the curved IHS it's more prone to crack and lose good thermal contact making CPU overclocking on at liquid-nitrogen poor with lapped flat CPU integrated heat spreaders and flat pot bases this almost never happens even when benching 18 cores at 6 gigahertz by the way with little CPUs like 1151 this is a much worse issue due to high power density versus the area we also previously discussed this with Joe Stefan Z who pre lapped all of his Allen teapots and CPUs before starting benchmarking in our initial bench build when we were working with Stefan Z on our Alan two efforts we asked why the lapping was needed at all when working with liquid nitrogen because at that point it seems like a couple of degrees is sort of a pointless reduction the answer was the same this can reduce the chance of paste cracking during testing means that contact loss is less likely when under extreme temperatures with large sub-zero temperature swings the end result is that our lapped 18 core CPU Falls with an expected range of what kingpin and 10 C for the average temperature reduction just from lapping but beyond that the CPU is now more suitable for future liquid-nitrogen overclocking as its service is much flatter and if you want to see our efforts with Joe previously we have a video preparing and recapping for the live stream that we did with liquid nitrogen but for smaller CPUs like the normal case q desktop CPUs the difference might be marginally higher than what we saw as a result of the increased power density in that tiny area but overall the real difference here is that the improvement is in a time reduction versus the manual sanding approach which can take hours and the thermal reduction being about the same versus a perfect manual laughs the big question now is whether we improved the core to core Delta because that's where we saw the worst performance in our original testing for core to core doubts as we previously observed the max versus minimum core difference of about 39 degrees on the 99 ATX II went under a stress test that pegged us to the maximum package temperature it's about 104 degrees Celsius more reasonably under the 4.4 gigahertz and 1.15 volt overclocks we were closer to a 30-degree delta across all of our 99 ATX e tests this is about the same even the 79 t DX II only saw an improvement to 25 degrees core decor when we use the liquid metal with these same tests and that's exactly why we thought lapping would post a bigger improvement because that's the one thing we hadn't done to the 79 t DX II but we didn't laugh the 79 t DX e so we don't get that comparison we left the 99 t DX e which means we're still dealing with an interface between the die and the IHS which we haven't controlled it was made at the factory disappointed with these results overall we are now hypothesizing that this is potentially a matter of how the CPU was soldered or made potentially indicating an issue under the IHS rather than with the IHS Intel did not reengineering ID 9 t DX e IHS or substrate and dies and so the processor is the same as the 79 80 X e aside from some more process maturity and the solder change it's possible that intel has thicker solder than is required because the heat spreader is unchanged for the most part from when it was used for thermal paste so solder would need to be at least as thick as thrown paste or the throne paste to use on the 79 80 XE to ensure contact between the die and the heat spreader there are some minor changes in here but the amount of changes for the 99 to DX e whereas the 70 90 d XE are not as significant as what you would typically see when going from paste to saw like the changes we saw with a 9900 K versus something like this 7700 K so overall it's possible that the solder thickness is the next point of consideration and for that it gets tougher to test paste can be compressed but solder can't reasonably be compressed in one of these CPUs it's possible yes but you'd really have to cook it the next thing we would have to do would be deleting the IHS that we just laughed which would improve things further but that'd be a lot of work for a CPU which is already perfect for the could nitrogen overclocking and is functional for anything else and it's $2,000 and D letting these is dangerous something that Rowand demonstrated clearly when he attempted to deal it his own 99 80 XE that's the deal at King doing that we didn't get the results that we were expecting to get for the quartic or deltas everyone was expecting a much lower court to court Delta because in the sanding process you can clearly see that the die has leveled out and it is level this is not an issue of an imperfect sand I mean we had arguably the world's leading overclocker basically do it for us so it's it's a more or less a perfect lap there's some imperfection toward the very outer edge but that wouldn't cause any of the the lack of improvement that we saw in court Accord Delta so realistically what we're left with is a couple of different options one of them is that potentially under the IHS there's another problem that we can't see and that might be something to do with the way the solder has been applied in theory and this is just my own speculation in theory Intel didn't really change a whole lot for the IHS with the 99 80 X E versus 79 80 DX e so this was not a the the substrates the same the CPU is the same so it wasn't re-engineered for solder maybe that is part of this but we're not really sure we can't speak to that the 79 80 XE also had pretty large Delta's core decor so it gets a little tricky to figure out what's expected because the one place we see this change is if we do direct eye contact and with direct dye we see those Delta's on a 79 80 XE drop to about 10 maybe 12 degrees instead of 25 degrees with liquid metal and an IHS now the 79 80 XE is not the same as our 99 80 XE so we still have variables there where with our 70 90 a d XE we didn't laughs the IHS but we deleted it and put liquid metal on it so we've we've removed the one variable that we're questioning on the 9090 d XE but we have a variable that we've removed on the ninety nine eighty XE which means it's not a perfect comparison so really if we wanted to go crazy with this the thing to do would be to next lap the 79 80 XE IHS put the liquid metal back on it with no sealant no no glue to hold them together and then see if that quartic or Delta really vanishes because at that point you've eliminated the variable VHS we could even if we wanted to risk it we could deal with our not United txt and swap it so we would have eliminated that variable and then we would have eliminated the consideration of the interface below the IHS and also the gap caused by the silicone adhesive so that'd be what we'd really have to do at this point to try these things they try to find the difference or try to eliminate the difference for the quartic or deltas another potential theory is that because a lot of these cold plates for liquid coolers are made with the expectation that they're going on a cpu there are some some exclusions from that but the expectation that they're going on the CPU might dictate that a lot of the cold plates for the liquid coolers on the market are also going to be concave or convex to accommodate whatever you service they're trying to fit and if that's the case then that could also potentially introduce some of the thermal Delta that we're seeing core to core but that's not not particularly is it's not the biggest consideration the biggest consideration we have is that solder under the IHS that's what we think is more likely the issue so we'll see it's I mean if there's a ton of interest then we can laugh the subnet ID XE maybe I can set it out to cane pin and ask him to do it for us just that the IHS we have and then test that again with a lap digest and liquid metal and no sealant and see what happens if the deltas go down at that point then we've found our issue on the ninety nine eight exe in theory would have been the interface between the two if it doesn't go down then there's something else at play and maybe it's just normal even if it's not particularly great we our understanding is that this Delta is on the higher end of things and in working with their Bauer previously we learned that the the core to core Delta of something like a case Q desktop enthusiast CPU might be twenty degrees and then direct IO pretty down closer like ten or something so yeah it's it's looking like maybe the interface at this point either way what we can take away from this is that a perfect lap on the IHS cuts it's about two to three degrees Celsius reduction and speaking with kingpin and ten this is expected and usual so if you wanted an extra two to three degrees that's what you can do to get it and then the core to core Delta is something that we need to investigate more and we don't have a firm conclusion on what specifically is causing the large gaps and hours but they are consistent now and they've been been seen between multiple runs including multiple batches of thermal paste and even different their own taste we're able to reproduce this issue and even some of the different coolers too so that's the big question mark now if there's a ton of interest we'll keep looking into it but this is this is getting to be pretty pretty low down at this point and not sure how many people care about this so if you really do care let us know that way we can take note of it and continue that's it for this one though thank you for watching sorry if the results were disappointing to you for the or to court deltas but we've learned a few important things which is that there's a lot more to consider than the IHS and also this whole process just to really reinforce this is more about improving the contacts for liquid nitrogen cooling not really for ambient so that's why Kane pin does it kingpin does it because if he uses the cpu as is where it's got that slight curve in the IHS then that causes problems with the thermal paste can crack and then your liquid nitrogen overclocking gets shut down temporarily while you either reapply the paste or try to heat it up and melt it and and fix the solute that the issue so that's really why he does it should you also lap your IHS well depends on do you want two to three degrees improvement if you're doing air cooling or liquid as the case may be so that's it for now thank you for watching subscribe for more you can go to patreon.com/scishow his accent stops that rightly or store doc gamers access net to pick my shirt like this one we'll see you all next time