I like that the fact the data lands where the data lands - without an external influence - The Corpse of Anandtech and Tom's (owned by the sme company) have a specific point to push (pro AMD - esp Alcorn)
@@oofig even with his "international" "fanbase" hes still a nobody. Only wannabe geeks with hipster tendencies like such people like this 8AUA or however this is written.
Don’t forget, back in the day no CPUs had heat spreaders. And “direct die” was all we had for all those years and things were fine. Breaking the die was insanely rare.
Even with chipped corners all of my Pentium III work just fine. Except my lone C0 stepping PIII 1000. Even undamaged it was unstable, unlike my pair of D0 PIII 1000. Still amazed that the foam pads on my Duron, Athlon XP and Sempron are intact as well (they've seen some heat).
Because they were designed to be that way. Newer chips are way easier to crack, and with the much higher density even a microcrack has a lot higher chance of killing the chip compared to old dies
@@Cypeq The package was ceramic on Durons and early Athlon, - it wasn't packing material. At best you had 4x small foam pads around the die to help prevent damage.
@@Cypeq Socket A and Socket 370 CPUs both had exposed dies. Prior to that AMD had used metal heatspreaders on the K6 family, and before that almost everyone had used ceramic packages for PGA and DIP CPUs. Sometimes with DIP you would, and these days almost always do, find plastic packages. Edit: also some Socket 7 CPUs, including some but not all Pentium MMX, had either a metal heatspreader or a metallized coating on the bare die. At this point I don't recall which it was and no longer have such old CPUs on-hand to check. Slot 1 and Slot 2 BGA CPUs had metal heatspreaders as well, while the later FCBGA models and Slot A CPUs had bare dies. Most users would never encounter either, since the slot-format CPU cards were enclosed in cassette-type packages that incorporated a (additional) full-dimension metal plate for cooler mounting - so essentially the early Intel slotted CPUs had _two_ heatspreaders, stacked.
Really interesting results... What if you use thermal pad for framing? I am thinking that 5-10W thermal transfer from the PCB could be helpful. It will be ~3% increase in cooling...
@der8auer, if you look closely at the cpu socket you'll see it has raised corners that stops the block being able to press down on the cpu. I think the kapton tape is worsening that issue. Have a look and I think you'll see what I'm on about. Best wishes from England!
What if you were to use a Panasonic PGS sheet on the water block to spread the heat over the whole block? Those PGS sheets are amazing at moving heat in the XY axis. I’m thinking Die -> liquid metal -> PGS -> water block. That would basically take the place of the IHS, but have much better thermal conductivity, and still eliminates one layer of thermal paste.
I love your videos, you push the limits. I have a suggestion. for the soldered Heat spreaders, why not make a jig to hold it flat and firm. You could use this in a cnc machine to shave off the top without stressing the CPU too much with heat and pressure. you could keep the standard mounting tabs which would make the whole mounting job easier. quite simply just remove the middle of the Heat spreader.
Pretty sure that EK Velocity block has a concave cold plate meant to match up with the IHS better. If you had the EK Magnitude block, you can switch out the concave cold plate for the flat one, which should work better in theory for direct die.
I did liquid metal on my 10850K without delidding with good temps. All I did was use 1,000 grit sand paper on IHS and water block and got really good results.
The Watercool blocks I bought have thick low oxygen copper base plates and have a lot of thermal mass. Oxygen content of copper will impede electrical resistance and thermal conductivity, not all coppers are equal in quality. There is also a quality improvement with the trademarked Plexiglass over generic acrylic.
Have u seen the channel Clock'EM UP? The guy just design a new kind of waterblock, that means the IHS now is kind of directdie mode for 10900K. Check it up u may like... he was running 10900k with very high clocks with ambient temp around 30c with GREAT temps.
it would be interesting to see at what temperature differential the deliding is efficient while considering heat flux with and without IHS. If you put a thermistor on the cooling block and compared it to the internal temperature of the chip you can probably estimate the amount of insulation experienced by the liquid metal -> IHS ->thermal paste -> cooling block. I would predict this is efficient with refrigeration and high current TECs with heavy overclocking or over volting/amping the dye since the temperature gradient will demand less and less insulation? Thanks I was thinking about this!
I would like to see more content investigating why the temperatures were better with the heatspreader in between. Would putting a small piece of silver between the die and the cooler be better?
It's probably thickness of the liquid metal layer. To me looks like direct die ends with greater LM height in the assembly than delided. Since IHS is copper, it's extremely conductive, while liquid metal is leagues below. Basically, lowest TIM height always wins because TIM materials have lower heat conductivity than any metals used in cooling(copper and aluminium). The solution is easy, reduce TIM layers total height by adding good conductor materials in the middle...oh wait: the IHS does already that when deliding, right? Solved, just delid. Other option would be designing a cooler mounting mechanism accounting for the dimensions of the assemby when doing direct die. But that's the ideal solution and it would be expensive since Intel changes PCB, die and IHS dimensions. For Intel is fine as long as these 3 total height remains the same not breaking cooler compatibilty.
yeah lap CPUs too just a little polishing it helps clean out any leftover STIM and it makes the die more flat, i had scratches on my die after a bad mounting attempt and i tried to lap them all out i couldn't get all of them but the CPU still works so they couldn't of been that deep and i am sure the lapping would of helped thermals a bit
Awesome channel Roman. It definitely needs more publicity because videos are great. I am sure you would be also popular with ladies ;) Though maybe the topic of overclocking processors is not the most popular with them ;) Anyhow, not sure others but I would love to see video about die lapping of 11900K. In my view this already naked Core i9 is basically screaming for it :)
Why not put copper shims in there to keep the cooler from every being able to rock that far down? Even covering them, given the pressure, it will push through whatever you put on them to "protect" them. Just use copper thinner than the die, but thicker than the components.
Interesting, I can't help but wonder if just using regular thermal paste and covering the whole square (including Kapton tape) might work better if the die isn't perfectly straight etc and needs more help spreading the heat.
With direct die cooling only a part of waterblock surface area works thermally - after few milimeters off the core die the copper is allready cooled to ambient (water) temp. Heatspreader actually, spreads heat in this case ,)
Could he gather more info on those 12C difference after delidding? He said it's way to high and he'd wait for others to confirm or negate his findings.
Can you do this with two or three more to see if this is a fluke or if there really is this much improvement delidding these? It just seems strange. Maybe this one didn't have that good of coverage/wetting with the In solder? Delidding these just looks like a real pain, but if they have solder issues it makes sense.
So I'm guessing some of the heat from the CPU spreads out through the PCB. So in order to get the best cooling, you have to have contact between the PCB and the cooling block. I guess you could carefully cover up all the surface mount components with rubber and then use liquid metal on the rest of the PCB? Sounds like a disaster tbh, but it would be worth a shot maybe.
Sounds like road to disaster yes haha but I don’t see at all why use the effort to pcb cool CPU … I have never heard of this and no reviews out there suggesting this is a thing at all. The air compartment under the ihs is so tiny that I can’t at all see this as a bottleneck in cooling. If you really want the best I suggest custom water loop and then you can additionally add either custom copper ihs with LM or all in - direct die LM …. I just ordered nCore V1 which is truly a unique one off amazing product for direct die cooling of my 10900k . 3 weeks for shipping- can’t wait
Hello. I know you've thought about it but a stupid question. Is it possible to make IHS that will be used for water cooling? I mean IHS with holes for water and liquid will actually touch the core in order to remove the thermal paste and additional layer of metal.
i have a rocketcool direct die kit on my 10850k, and ive had super stable and equal temps accross all cores. i did have to modify a couple things (like for the water block i am using i had to use electrical tape under the block support studs instead of the plastic washers, because it created too much clearance between the block and the die) but other than that its great, and for 60 bucks for the delidding tools and the mounting stuff, not a bad deal really.
I wanted that also originally but due to asus board there is an extra row of capacitors requiring to mill or Dremel the water block. So I’m going for another one off unique product .. the nCore V1! Check it out ! F ing amazing product… ordered two days ago so can’t wait
Would you be interested in conducting some basic delidding and achieving, (with a lot more cooling and stability, of course), on my rehearsal piano ? I don't exactly know the bend and the size of PCB, but the reason is pretty, fairly simple, and the piano itself, sounds a lot like a tank of liquid metal ? Best regards.
Have you tried lapping the block coldplate on the same surface as the the CPU? I find that the LM pools, so trying to discern contact when test fitting the TIM'ed block to the CPU can be tough.
Have you ever done any sustained OC testing? I was thinking of using a micro-brew beer chiller to keep the coolant a few degrees above the dew point and using a protein folding program as a stressor.
You would have to be very careful of condensation. Mixing cold and warm can be dangerous .. but the thought have crossed my mind haha although now with direct die, liquid metal and powerful custom loop . I no longer need this… GPU max 50g…. Typical 45g. CPU 10900k 48-60g for gaming. 78-80g extreme stress test.
This is such a weird result. The lid must be cooling some other area that affects the readings. Could it be so that temperature measurements are by in-pcb devices and they only have the "pins/leads" into cores? Then this would make sense.
The result is perfectly normal, as far as I see it. The surface of the IHS is larger than the die. The heat from the die is distributed evenly through the IHS. More surface=more contact (between the IHS and the surface of the cooler)=better heat dissipation=better/lower temperatures. And of course, the opposite, the direct die cooling, as it was proven in the video.Heat pads around the core would not improve the performance, as literally the small electronic parts in the pcb produce very little to zero heat, which is irrelevant to take into account and would not influence the overall results.
@@ron200088 What's happening is more likely to be due to the total height of liquid metal used in both assemblies. Probably direct die ends with a thicker layer of liquid metal overall than delided, even with delided using 2 layers of liquid metal. It's what der8auer points at when he says the coolers aren't designed for direct die, most coolers mounts assume there's an IHS and their dimensions put distances accordingly, so in direct die you have to fill the extra gap with your TIM. Liquid metal is faster heat conductor than silicon based pastes, but it's still very far from conductivity compared to copper(IHS, heatpipes and water-cooling blocks) and aluminium(air-cooling finned heat-sinks). Basically, in this setup delided is winning because it uses less liquid metal height, that's long story short. MORE INDEPTH ABOUT HEAT TRANSFER IN PC, SKIP IF DON'T WANT DETAILS That would really require an engineer to do calculations. But heat doesn't spread evenly on the IHS: things in nature follow the less resistance path to equalization. Heat spreading around the IHS isn't the less resistance in this scenario, but actually heat going in straight line through the IHS towards the cold plate. When the straight line path becomes totally saturated heat will act in a similar way a car displaces wind: the wind slides aside the closest around the car surface. Heat wise will slide the closest around the saturated exchange area of the IHS but always going towards the cold plate(which acts as an exit) in the shortest geometrical way. Similar happens if you put a very intense single core load. No matter only 1/12th of the entire surface of the chip is generating all that heat and it could spread over the entire chip to use more surface and keep the specific core running lower temperature, there's no awareness in physics that the priority is to keep the die cool, priority is moving through the path that costs less, no matter how hot silicon gets. That core is going to overheat and the nearby cores will be like no load at all because less resistance path in CPU cooling is normally shooting straight upwards towards the cooler's cold plate. This is a confusion that comes probably from motherboards VRM metal block style heat sinks. But that's a totally different dynamic: there heat transfer to air is extremely poor, not comparable to finned aluminium heat sinks or water-cooling radiators with fans blowing at them. So in a block style VRM heat sink heat spreads around the heat sink because it acts more like a prison. But if you were to put a fan blowing straight at the metal VRM heat sink, increasing metal to air heat transfer in the air blown area, even if it's a block it will decrease by a long shot how much heat spreads over the block, because now heat will choose to go out of the metal block through the metal portion that's getting air blown at it's surface by the shortest geometrical path.
Haha, certainly had some good times with those - how about the AMD Duron? Since the beginning of the year I've been going through my former HDD's trying to find a video my buddy and I made called "The Little Duron that Could" which ran and o\c'ed too well having part of its core physically missing. - much appreciated comment nonetheless, thank you for the nostalgic moment!
This could be due to low flow rate. Most cases it won't matter, I know, but this is an extreme case of heat concentration. Any chance you would run the test without the quick-releases?
No proper contact. That's because the EKBW water block has a convex cold plate machined for Intel's concave IHS, which doesn't make good contact with the chip's direct surface.
Did you by any chance get the Ncore V1 direct die block that is finally out? Can you try that with the 11900k by any chance? I have one I ordered like 2 or 3 years ago during the kickstarter and it finally came. I wonder if it would work on the 11900k.
Iirc the main advantage of watercooling is that the water functions as a giant heat sink, in the sense that it can collect a lot of heat (because of the mass) and use a larger surface (radiators) than regular cooling. So if you're not seeing an improvement in watercooling it's unlikely that it will benefit aircooling.
@@DCG909 you missed the part where der8auer said his waterblocks not meant for DD cooling like that, meaning that this specific scenario is not showing best possible results. (that is why DD results were worse in comparison to the delidded one) D15 might be actually better here, but the weight is another issue, you dont want 1.3KG chonk of metal sitting on your CPU die like that.
In theory, direct-die cooling SHOULD work much better than just with de-liding. Something doesn't add up there, there might still be a contact pressure or evenness issue. I've done that many years ago with an older Intel CPU that wasn't using solder under the IHS but normal paste and the difference was noticeable between just delid and direct-die.
Your T shirt... it would look fantastic without that over the shoulder broad pink stripe, The logo on the right shoulder is great, a small logo on the left chest area would be great, yes its traditional but it works. or make it less tight fitting, give it slightly raised Chinese collars and place a logo on either or both the raised collars.
Makes sense as most are concave. How where the results difference? I am curious as I have a custom cobbler ihs which is flat for my 10900k but my waterblock an older EK classic, I have never modified. I am unsure if it’s flat or concave. It will be history though in 3 weeks after almost a decade in service haha it will be replaced by unique one off nCore V1 direct die waterblock which actually have a built in ihs! And locks in the cpu locking and spring mechanism- no screws needed. Can’t wait … ohhh and that too should be perfectly flat as it’s a direct die product …. But tell me your results and difference with lap and no lap block. Also you did use LM right ?
I can't believe you are sponsering hetzner. They are WAY better than OVH lol, as a partial guy learning Linux administration I use hetzner for one server there, NO complaints whatsoever.
Why don't people direct die cool their desktop cpus more often? Majority of laptops are cooled this way. Now I know there are dimensions that will have to be met for this to be possible, however, more efficient cooling is better, I'm surprised there aren't many manufacturers don't make coolers designed to fit stuff like this(I'm not the most knowledgeable, so if I'm missing something that is a no brainer, this is why)
@@kaylaandjimbryant8258 mhm to be fair , I got my first pc last year, it's a laptop of course,( I've had other little pcs, but ones like cheap chromebooks), and when I got my laptop first thing I did was researching on liquid metal applications. And guess what my dumbass did to my laptop that costed over 2300 dollars? I immediately put liquid metal on it, with no previous hands on experience doing it too! I did it to the cpu and gpu, and I learned to do it just from watching videos, from ltt and people from channels like this. And the end result, it still works to this day, now did I almost have a heart attack doin it in the process knowing I could have a 2300 dollar paper weight? VERY MUCH SO LOL. But keep in mind , I've always liked pc hardware, and I literally never did this before, not along actually have ever built a pc nor took many apart, aside from watching a million videos over the years. (This is a sort of irreverent response but I'm bored lol)
So it basically goes to one thing: desktop computers are used in enterprises where they have IT workers which assemble, repair and reuse the computer pieces. Since time is money and this type of client cares a lot for this manufacturers like Intel have to design CPU mounting in a way that replacing it's very fast and safe. An enterprise wouldn't be happy if their IT guys took 15 minutes to replace a CPU, it has to be done under one minute and with ISH and CPU manufacturer stock coolers is doable. Same happens with RAM. If you ever worked in a software company when a RAM is faulty the IT guy comes pretty fast with a couple of modules and just replaces it ASAP until it works mixing them in any way. Since everything runs at JEDEC SPD any timing configuration will POST right off the bat and stable due to the absurd margin, no memtest needed. On another hand on computer hobbyist realm everyone agrees mixing up RAM is troublesome and leaves performance on the table because you do that and you're going to need to find a common frequency and timings that POST for both modules or you just run it a JEDEC SPD and go along with slower RAM. Then memtesting your configuration to ensure it runs stable because you aren't aiming to work with a fat error margin. You're basically disregarding the platform used in gaming has also very big clients in the enterprise world who buy much more than "individual users". Intel isn't oblivious to this.
@@wesleyfurnish8228 everyone has to start somewhere. We also use conductonaut on our laptops and GPUs. All of the cpus here are the intel kind that were easily delidded, so it was easy once tools were available to do so (my still main rig due to licensing is a 4790K that I delidded in 2015 using the razor blade method). If the new ones all use indium solder, that complicates things drastically though and seriously increases the risk involved.
Would be awesome if intel and amd decided to produce a variant of future desktop processors that are bare die processors without IHS attached for the experienced crowd. If laptops almost always have direct die cooling, why can't desktops have the same?
Considering the thermal output that this thing can reach under Prime95, I would not be wholly surprised if part of the issue would be due to the whole die being, in itself, too small. By which I mean that, possibly, this particular chip crosses a boundary of some kind where a bigger square area of dissipation yelds a better rate of heat transfer than a die-sized area, regardless of the latter having the best thermal conductivity achievable (block contact and pressure issues being conveniently sidelined as a variable, for argument's sake). It's mostly speculation that i can't see being reliably tested, but the combo of taller die, backport of a dense design upon a less dense fabrication process and so on might give it some legs. The end result sadly remains the same, DC doesn't do anything, but knowing would be nice. I suppose that the intel manufacturing team does know, to an extent at least. Maybe in the future, with them licensing floorplans and manufacturing capacity, a chip will come out that will yeld (ha-ha) the opportunity to detect this.
I wouldn't say . Take for e.g. the 10th gen i9 . The die is smaller . And regardless , direct die cooling with a proper frame does lead to a very good performance (I am typing this on a i9-10900Kf - direct die cooled - 5.4Ghz speed on 24/7 usage .... . Temps of CPU before/after direct die were like 30deg difference ....)
Roman, I don't speak German but I like to _think_ I know a few words and phrases. When you say your sponsor's name, "Hetzner," my 76-year-old ears perk up - witch? - which witch? ("Hexe" ... ja, ich spreche deutsche perfekt ;^)
I have a thought about the direct-die results. You probably right that the block isn't meant for this application What if those notches where the water flows (in the block) are parallel to the length of the die so the heat that picked up by water travels on the whole die before its back to the pipe So if you turn the block 90 degrees you get fresh flow on the whole surface and maybe a bit better temps
Really appreciate you doing an English version of your videos. Always interested in what you have to say about recent tech.
Thanks 💪🏼💪🏼
You are one of my favorite youtube OC dudes 😎👍
@@ShaneIsbister This is more than overclocking. That's why der8auer's content is way interesting than others
I like that the fact the data lands where the data lands - without an external influence - The Corpse of Anandtech and Tom's (owned by the sme company) have a specific point to push (pro AMD - esp Alcorn)
Absolutely, always excited that I get to watch a new video.
I live the English videos, appreciate all the extra work you put in for us English only people
Otherwise he would be a absolute nobody.
@@harrison00xXx nah he gets a good amount of views on the german videos
@@oofig even with his "international" "fanbase" hes still a nobody. Only wannabe geeks with hipster tendencies like such people like this 8AUA or however this is written.
@@harrison00xXx wtf do u have against this man smh
@@oofig Hes a ugly nerd and part of the greedy, bad youtuber scene. That should be enough already.
Don’t forget, back in the day no CPUs had heat spreaders. And “direct die” was all we had for all those years and things were fine. Breaking the die was insanely rare.
Even with chipped corners all of my Pentium III work just fine. Except my lone C0 stepping PIII 1000. Even undamaged it was unstable, unlike my pair of D0 PIII 1000.
Still amazed that the foam pads on my Duron, Athlon XP and Sempron are intact as well (they've seen some heat).
was it really direct die though? most of those chips had like a thick ceramic packing material around them
Because they were designed to be that way. Newer chips are way easier to crack, and with the much higher density even a microcrack has a lot higher chance of killing the chip compared to old dies
@@Cypeq The package was ceramic on Durons and early Athlon, - it wasn't packing material. At best you had 4x small foam pads around the die to help prevent damage.
@@Cypeq Socket A and Socket 370 CPUs both had exposed dies. Prior to that AMD had used metal heatspreaders on the K6 family, and before that almost everyone had used ceramic packages for PGA and DIP CPUs. Sometimes with DIP you would, and these days almost always do, find plastic packages.
Edit: also some Socket 7 CPUs, including some but not all Pentium MMX, had either a metal heatspreader or a metallized coating on the bare die. At this point I don't recall which it was and no longer have such old CPUs on-hand to check.
Slot 1 and Slot 2 BGA CPUs had metal heatspreaders as well, while the later FCBGA models and Slot A CPUs had bare dies. Most users would never encounter either, since the slot-format CPU cards were enclosed in cassette-type packages that incorporated a (additional) full-dimension metal plate for cooler mounting - so essentially the early Intel slotted CPUs had _two_ heatspreaders, stacked.
I would like to see the performance of intel cryo cooler from previous 10900K on 11900K
Could you do that again?
this is the video I have been looking for @der8auer
You have any plans for more scanning electron microscope content?
did one last week :D Will be online soon
@@der8auer nice, looking forward to seeing it
@@der8auer noiice 💪
Damnnn You've gotten buff during lockdown
overclocking himself out of boredom
Don't worry, he already has liquid cooling built-in.
Overclocking so he can play those girls at a higher success rate
he oc'd
himself
Don't delid yourself..not worth it.
Really interesting results... What if you use thermal pad for framing? I am thinking that 5-10W thermal transfer from the PCB could be helpful. It will be ~3% increase in cooling...
Merci Roman de faire ces tests pour nous ! Thanks Roman for doing those tests for us !
Don't forget the influence of the capton tape.
I bet it changes some temperature insulation's on the PCB.
@der8auer, if you look closely at the cpu socket you'll see it has raised corners that stops the block being able to press down on the cpu. I think the kapton tape is worsening that issue. Have a look and I think you'll see what I'm on about. Best wishes from England!
I love this nerdy channel. I learn a lot from it..
Der8auer Power is unstoppable! You "are" the CPU whisperer... Cheers.
Is it not direct water cooling of the bare die as the metal block is still there? Is it possible to flow water directly on the silicon?
Isn't the height difference to the DIE too small at 6:27 because of the tape on the template ?!
yea the corsair blocks hit the caps. Had the same issue with 10850 K !
The "Lid" Heat spreader also dissipates heat from the substrate along with the die, that could also be a factor.
Auch wenn intel nichts hinbekommt ist es doch schön das du zu den unnötigen cpus weiterhin spannende Videos hochlädst :)
So you essentially cooled down that kapton tape ;D
9:23 dang, I think you're breaking NDA showing off the 119000K there lol
What if you were to use a Panasonic PGS sheet on the water block to spread the heat over the whole block? Those PGS sheets are amazing at moving heat in the XY axis. I’m thinking Die -> liquid metal -> PGS -> water block. That would basically take the place of the IHS, but have much better thermal conductivity, and still eliminates one layer of thermal paste.
When I saw the title, I thought he was gonna put an O-ring on the CPU and just run water directly on top of the die lol.
If the cooler needs the heat tô be spreaded to work properly, how a vapor chamber would perform on a direct die application?
another amazing work, cheers.
I love your videos, you push the limits. I have a suggestion. for the soldered Heat spreaders, why not make a jig to hold it flat and firm. You could use this in a cnc machine to shave off the top without stressing the CPU too much with heat and pressure. you could keep the standard mounting tabs which would make the whole mounting job easier.
quite simply just remove the middle of the Heat spreader.
Useless work for 99% of us, but.. In the name of science, it's all worth it.
Respect, nice video and good luck with future projects!
Pretty sure that EK Velocity block has a concave cold plate meant to match up with the IHS better. If you had the EK Magnitude block, you can switch out the concave cold plate for the flat one, which should work better in theory for direct die.
Great video, with amazing results. I love this type of video, as I am not brave enough to Delid my own CPU, good job, you make it look easy.
I did liquid metal on my 10850K without delidding with good temps. All I did was use 1,000 grit sand paper on IHS and water block and got really good results.
The Watercool blocks I bought have thick low oxygen copper base plates and have a lot of thermal mass. Oxygen content of copper will impede electrical resistance and thermal conductivity, not all coppers are equal in quality. There is also a quality improvement with the trademarked Plexiglass over generic acrylic.
Have u seen the channel Clock'EM UP? The guy just design a new kind of waterblock, that means the IHS now is kind of directdie mode for 10900K. Check it up u may like... he was running 10900k with very high clocks with ambient temp around 30c with GREAT temps.
it would be interesting to see at what temperature differential the deliding is efficient while considering heat flux with and without IHS. If you put a thermistor on the cooling block and compared it to the internal temperature of the chip you can probably estimate the amount of insulation experienced by the liquid metal -> IHS ->thermal paste -> cooling block. I would predict this is efficient with refrigeration and high current TECs with heavy overclocking or over volting/amping the dye since the temperature gradient will demand less and less insulation? Thanks I was thinking about this!
What do you mean by simple "delete"? Please explain
I would like to see more content investigating why the temperatures were better with the heatspreader in between. Would putting a small piece of silver between the die and the cooler be better?
It's probably thickness of the liquid metal layer. To me looks like direct die ends with greater LM height in the assembly than delided. Since IHS is copper, it's extremely conductive, while liquid metal is leagues below. Basically, lowest TIM height always wins because TIM materials have lower heat conductivity than any metals used in cooling(copper and aluminium).
The solution is easy, reduce TIM layers total height by adding good conductor materials in the middle...oh wait: the IHS does already that when deliding, right? Solved, just delid.
Other option would be designing a cooler mounting mechanism accounting for the dimensions of the assemby when doing direct die. But that's the ideal solution and it would be expensive since Intel changes PCB, die and IHS dimensions. For Intel is fine as long as these 3 total height remains the same not breaking cooler compatibilty.
don't forget that you have to lap the block itself since they area relying on any imperfections to be filled by TIM over a large area
yeah lap CPUs too just a little polishing it helps clean out any leftover STIM and it makes the die more flat, i had scratches on my die after a bad mounting attempt and i tried to lap them all out i couldn't get all of them but the CPU still works so they couldn't of been that deep and i am sure the lapping would of helped thermals a bit
Would lapping the die itself have tangible results in cooling?
Funnily enough we all used to do our direct die cooling in the early Athlon days.
Poor russian guys like me still using direct die cooling for chinese "engeneering laptop to desctop" mutant CPUs like mine QTJ2
I wonder if it would get better with a lapped ihs.
Awesome channel Roman. It definitely needs more publicity because videos are great. I am sure you would be also popular with ladies ;) Though maybe the topic of overclocking processors is not the most popular with them ;) Anyhow, not sure others but I would love to see video about die lapping of 11900K. In my view this already naked Core i9 is basically screaming for it :)
I wonder how EK-QuantumX Delta TEC will perform with 11900K, if there is any Z590 board have a BIOS that supports Intel® Cryo Cooling Technology?
Why not put copper shims in there to keep the cooler from every being able to rock that far down? Even covering them, given the pressure, it will push through whatever you put on them to "protect" them. Just use copper thinner than the die, but thicker than the components.
The graph at 9:00 leaked an intel confidential, the all new 119000k!
What about power consumption? It should draw less on lower temps!
Interesting, I can't help but wonder if just using regular thermal paste and covering the whole square (including Kapton tape) might work better if the die isn't perfectly straight etc and needs more help spreading the heat.
With direct die cooling only a part of waterblock surface area works thermally - after few milimeters off the core die the copper is allready cooled to ambient (water) temp. Heatspreader actually, spreads heat in this case ,)
Thanks for the vid Der8auer, keep em coming sir ! Also, when are we going to get a OC match between you and Steve from GN ? :D
I love your videos.
So informative.👍
Could he gather more info on those 12C difference after delidding? He said it's way to high and he'd wait for others to confirm or negate his findings.
Hi Roman, was the EK Velocity water block flattened / lapped? They are convex from factory which would have impacted the contact area on the die.
5800x Delid?
Can you do this with two or three more to see if this is a fluke or if there really is this much improvement delidding these? It just seems strange. Maybe this one didn't have that good of coverage/wetting with the In solder? Delidding these just looks like a real pain, but if they have solder issues it makes sense.
So I'm guessing some of the heat from the CPU spreads out through the PCB. So in order to get the best cooling, you have to have contact between the PCB and the cooling block. I guess you could carefully cover up all the surface mount components with rubber and then use liquid metal on the rest of the PCB? Sounds like a disaster tbh, but it would be worth a shot maybe.
Sounds like road to disaster yes haha but I don’t see at all why use the effort to pcb cool CPU … I have never heard of this and no reviews out there suggesting this is a thing at all. The air compartment under the ihs is so tiny that I can’t at all see this as a bottleneck in cooling. If you really want the best I suggest custom water loop and then you can additionally add either custom copper ihs with LM or all in - direct die LM …. I just ordered nCore V1 which is truly a unique one off amazing product for direct die cooling of my 10900k . 3 weeks for shipping- can’t wait
Hello. I know you've thought about it but a stupid question.
Is it possible to make IHS that will be used for water cooling? I mean IHS with holes for water and liquid will actually touch the core in order to remove the thermal paste and additional layer of metal.
What about putting a thermal pad that is the same height as the die on the PCB of the CPU?
What if a cold plate was welded directly to the IHS?
i have a rocketcool direct die kit on my 10850k, and ive had super stable and equal temps accross all cores. i did have to modify a couple things (like for the water block i am using i had to use electrical tape under the block support studs instead of the plastic washers, because it created too much clearance between the block and the die) but other than that its great, and for 60 bucks for the delidding tools and the mounting stuff, not a bad deal really.
I wanted that also originally but due to asus board there is an extra row of capacitors requiring to mill or Dremel the water block. So I’m going for another one off unique product .. the nCore V1! Check it out ! F ing amazing product… ordered two days ago so can’t wait
when was the last time direct die WAS worth it?
Would you be interested in conducting some basic delidding and achieving, (with a lot more cooling and stability, of course), on my rehearsal piano ? I don't exactly know the bend and the size of PCB, but the reason is pretty, fairly simple, and the piano itself, sounds a lot like a tank of liquid metal ? Best regards.
The neutron style!
Can you direct die cool for extended periods of time?
Have you tried lapping the block coldplate on the same surface as the the CPU?
I find that the LM pools, so trying to discern contact when test fitting the TIM'ed block to the CPU can be tough.
But how will it perform with the ekwb tec cooler ?🤔
Have you ever done any sustained OC testing?
I was thinking of using a micro-brew beer chiller to keep the coolant a few degrees above the dew point and using a protein folding program as a stressor.
You would have to be very careful of condensation. Mixing cold and warm can be dangerous .. but the thought have crossed my mind haha although now with direct die, liquid metal and powerful custom loop . I no longer need this… GPU max 50g…. Typical 45g. CPU 10900k 48-60g for gaming. 78-80g extreme stress test.
This is such a weird result. The lid must be cooling some other area that affects the readings. Could it be so that temperature measurements are by in-pcb devices and they only have the "pins/leads" into cores? Then this would make sense.
Could maybe test this by having heat pads around the core
The result is perfectly normal, as far as I see it. The surface of the IHS is larger than the die. The heat from the die is distributed evenly through the IHS. More surface=more contact (between the IHS and the surface of the cooler)=better heat dissipation=better/lower temperatures. And of course, the opposite, the direct die cooling, as it was proven in the video.Heat pads around the core would not improve the performance, as literally the small electronic parts in the pcb produce very little to zero heat, which is irrelevant to take into account and would not influence the overall results.
@@ron200088 What's happening is more likely to be due to the total height of liquid metal used in both assemblies. Probably direct die ends with a thicker layer of liquid metal overall than delided, even with delided using 2 layers of liquid metal. It's what der8auer points at when he says the coolers aren't designed for direct die, most coolers mounts assume there's an IHS and their dimensions put distances accordingly, so in direct die you have to fill the extra gap with your TIM.
Liquid metal is faster heat conductor than silicon based pastes, but it's still very far from conductivity compared to copper(IHS, heatpipes and water-cooling blocks) and aluminium(air-cooling finned heat-sinks). Basically, in this setup delided is winning because it uses less liquid metal height, that's long story short.
MORE INDEPTH ABOUT HEAT TRANSFER IN PC, SKIP IF DON'T WANT DETAILS
That would really require an engineer to do calculations. But heat doesn't spread evenly on the IHS: things in nature follow the less resistance path to equalization. Heat spreading around the IHS isn't the less resistance in this scenario, but actually heat going in straight line through the IHS towards the cold plate. When the straight line path becomes totally saturated heat will act in a similar way a car displaces wind: the wind slides aside the closest around the car surface. Heat wise will slide the closest around the saturated exchange area of the IHS but always going towards the cold plate(which acts as an exit) in the shortest geometrical way.
Similar happens if you put a very intense single core load. No matter only 1/12th of the entire surface of the chip is generating all that heat and it could spread over the entire chip to use more surface and keep the specific core running lower temperature, there's no awareness in physics that the priority is to keep the die cool, priority is moving through the path that costs less, no matter how hot silicon gets. That core is going to overheat and the nearby cores will be like no load at all because less resistance path in CPU cooling is normally shooting straight upwards towards the cooler's cold plate.
This is a confusion that comes probably from motherboards VRM metal block style heat sinks. But that's a totally different dynamic: there heat transfer to air is extremely poor, not comparable to finned aluminium heat sinks or water-cooling radiators with fans blowing at them. So in a block style VRM heat sink heat spreads around the heat sink because it acts more like a prison. But if you were to put a fan blowing straight at the metal VRM heat sink, increasing metal to air heat transfer in the air blown area, even if it's a block it will decrease by a long shot how much heat spreads over the block, because now heat will choose to go out of the metal block through the metal portion that's getting air blown at it's surface by the shortest geometrical path.
@@deidian635 Thank you for the in depth analysis and explanation. Very much appreciated !
Every time I see one of these I think back to the AMD Athlon Thunderbird/XP days and people arguing whether you needed the copper shim or not.
The shim was 4$ I used it.
Lol, and those bare chips were much smaller back then.
Don't forget the apprehension felt when you had to rotate the heatsink across the CPU to get the clips to connect.
Haha, certainly had some good times with those - how about the AMD Duron? Since the beginning of the year I've been going through my former HDD's trying to find a video my buddy and I made called "The Little Duron that Could" which ran and o\c'ed too well having part of its core physically missing. - much appreciated comment nonetheless, thank you for the nostalgic moment!
Can you make a video on direct water on die cooling?
9:01 119000K. When can we get our hands on those?
Do OC with the cryo cooler just for lols?
1:40 uuuuuff.. matey use the drill for corner cuts lexan or plexi :) nice vid
can you test the 11900k with the cryo cooler please.
This could be due to low flow rate. Most cases it won't matter, I know, but this is an extreme case of heat concentration.
Any chance you would run the test without the quick-releases?
If you tighten the CPU block with regular M4 screws it will be much better.
Obviously, you don't have enough pressure contact like now.
If the pins contact and the cpu boots it`s enough pressure....
That was indeed unexpected.
can you add another experiment by lapping/sanding the IHS itself?
No proper contact.
That's because the EKBW water block has a convex cold plate machined for Intel's concave IHS, which doesn't make good contact with the chip's direct surface.
Does direct die cooling work well for LN2 or is the heatspreader still useful there as well?
what can you recommend for 4790K?
Did you by any chance get the Ncore V1 direct die block that is finally out? Can you try that with the 11900k by any chance? I have one I ordered like 2 or 3 years ago during the kickstarter and it finally came. I wonder if it would work on the 11900k.
Good shit 💪
what about high end air cooling like a D15? Would DD be a benefit?
Iirc the main advantage of watercooling is that the water functions as a giant heat sink, in the sense that it can collect a lot of heat (because of the mass) and use a larger surface (radiators) than regular cooling.
So if you're not seeing an improvement in watercooling it's unlikely that it will benefit aircooling.
@@DCG909 you missed the part where der8auer said his waterblocks not meant for DD cooling like that, meaning that this specific scenario is not showing best possible results. (that is why DD results were worse in comparison to the delidded one)
D15 might be actually better here, but the weight is another issue, you dont want 1.3KG chonk of metal sitting on your CPU die like that.
In theory, direct-die cooling SHOULD work much better than just with de-liding. Something doesn't add up there, there might still be a contact pressure or evenness issue. I've done that many years ago with an older Intel CPU that wasn't using solder under the IHS but normal paste and the difference was noticeable between just delid and direct-die.
P.S.: I think you should check the flatness of the block and the die itself and maybe lap them if they're not perfectly flat.
Guys, do you know a nice glue to relid a 9th gen?
I'm pretty sure when I did direct die the cooler standoff Heights need to be lower. EK have shorter ones
Your T shirt... it would look fantastic without that over the shoulder broad pink stripe, The logo on the right shoulder is great, a small logo on the left chest area would be great, yes its traditional but it works. or make it less tight fitting, give it slightly raised Chinese collars and place a logo on either or both the raised collars.
I learned German just to watch your vids
How do I find the PSU for my PC
bin deutsch haha
I've had to lap my cpu blocks when I used for direct die.
Makes sense as most are concave. How where the results difference? I am curious as I have a custom cobbler ihs which is flat for my 10900k but my waterblock an older EK classic, I have never modified. I am unsure if it’s flat or concave. It will be history though in 3 weeks after almost a decade in service haha it will be replaced by unique one off nCore V1 direct die waterblock which actually have a built in ihs! And locks in the cpu locking and spring mechanism- no screws needed. Can’t wait … ohhh and that too should be perfectly flat as it’s a direct die product …. But tell me your results and difference with lap and no lap block. Also you did use LM right ?
How about letting the water contact the die directly, then it will be super direct die cooling, idk if it's a good idea or not
question: where can i get that shirt? hehe i like it! thanks mate!
what QDC fittings do you use?
interesting results
I can't believe you are sponsering hetzner. They are WAY better than OVH lol, as a partial guy learning Linux administration I use hetzner for one server there, NO complaints whatsoever.
Wow, that's unexpected
Why don't people direct die cool their desktop cpus more often? Majority of laptops are cooled this way. Now I know there are dimensions that will have to be met for this to be possible, however, more efficient cooling is better, I'm surprised there aren't many manufacturers don't make coolers designed to fit stuff like this(I'm not the most knowledgeable, so if I'm missing something that is a no brainer, this is why)
Dies are fragile, and most people where I live would Fk it up. They have to idiot-proof it for the unwashed masses.
@@kaylaandjimbryant8258 mhm to be fair , I got my first pc last year, it's a laptop of course,( I've had other little pcs, but ones like cheap chromebooks), and when I got my laptop first thing I did was researching on liquid metal applications. And guess what my dumbass did to my laptop that costed over 2300 dollars? I immediately put liquid metal on it, with no previous hands on experience doing it too! I did it to the cpu and gpu, and I learned to do it just from watching videos, from ltt and people from channels like this. And the end result, it still works to this day, now did I almost have a heart attack doin it in the process knowing I could have a 2300 dollar paper weight? VERY MUCH SO LOL. But keep in mind , I've always liked pc hardware, and I literally never did this before, not along actually have ever built a pc nor took many apart, aside from watching a million videos over the years. (This is a sort of irreverent response but I'm bored lol)
So it basically goes to one thing: desktop computers are used in enterprises where they have IT workers which assemble, repair and reuse the computer pieces. Since time is money and this type of client cares a lot for this manufacturers like Intel have to design CPU mounting in a way that replacing it's very fast and safe. An enterprise wouldn't be happy if their IT guys took 15 minutes to replace a CPU, it has to be done under one minute and with ISH and CPU manufacturer stock coolers is doable.
Same happens with RAM. If you ever worked in a software company when a RAM is faulty the IT guy comes pretty fast with a couple of modules and just replaces it ASAP until it works mixing them in any way. Since everything runs at JEDEC SPD any timing configuration will POST right off the bat and stable due to the absurd margin, no memtest needed.
On another hand on computer hobbyist realm everyone agrees mixing up RAM is troublesome and leaves performance on the table because you do that and you're going to need to find a common frequency and timings that POST for both modules or you just run it a JEDEC SPD and go along with slower RAM. Then memtesting your configuration to ensure it runs stable because you aren't aiming to work with a fat error margin.
You're basically disregarding the platform used in gaming has also very big clients in the enterprise world who buy much more than "individual users". Intel isn't oblivious to this.
@@wesleyfurnish8228 everyone has to start somewhere. We also use conductonaut on our laptops and GPUs. All of the cpus here are the intel kind that were easily delidded, so it was easy once tools were available to do so (my still main rig due to licensing is a 4790K that I delidded in 2015 using the razor blade method). If the new ones all use indium solder, that complicates things drastically though and seriously increases the risk involved.
Would be awesome if intel and amd decided to produce a variant of future desktop processors that are bare die processors without IHS attached for the experienced crowd. If laptops almost always have direct die cooling, why can't desktops have the same?
Considering the thermal output that this thing can reach under Prime95, I would not be wholly surprised if part of the issue would be due to the whole die being, in itself, too small. By which I mean that, possibly, this particular chip crosses a boundary of some kind where a bigger square area of dissipation yelds a better rate of heat transfer than a die-sized area, regardless of the latter having the best thermal conductivity achievable (block contact and pressure issues being conveniently sidelined as a variable, for argument's sake).
It's mostly speculation that i can't see being reliably tested, but the combo of taller die, backport of a dense design upon a less dense fabrication process and so on might give it some legs. The end result sadly remains the same, DC doesn't do anything, but knowing would be nice.
I suppose that the intel manufacturing team does know, to an extent at least. Maybe in the future, with them licensing floorplans and manufacturing capacity, a chip will come out that will yeld (ha-ha) the opportunity to detect this.
I wouldn't say . Take for e.g. the 10th gen i9 . The die is smaller . And regardless , direct die cooling with a proper frame does lead to a very good performance (I am typing this on a i9-10900Kf - direct die cooled - 5.4Ghz speed on 24/7 usage .... . Temps of CPU before/after direct die were like 30deg difference ....)
Roman, I don't speak German but I like to _think_ I know a few words and phrases. When you say your sponsor's name, "Hetzner," my 76-year-old ears perk up - witch? - which witch? ("Hexe" ... ja, ich spreche deutsche perfekt ;^)
I have a thought about the direct-die results. You probably right that the block isn't meant for this application
What if those notches where the water flows (in the block) are parallel to the length of the die so the heat that picked up by water travels on the whole die before its back to the pipe
So if you turn the block 90 degrees you get fresh flow on the whole surface and maybe a bit better temps
Shouldn't a company come up with a direct die cpu block? I would assume there is enough market to make profit so why not?
Hetzner using custom cables (I mean, it's just cablemod, but still) in data center servers lmao
Would love Roman to make ram overclocking and 11600K/700K cpu with the Gigabyte Aorus Tachyon mainboard ! 🥰