I always think it's a shame companies like Intel/AMD don't offer a special edition version that was never 'lidded' as it's the delidding process and the cleaning up the die that is the hassle, scary part and where most of the risk is.
@@Simon_Denmark I disagree, Intel's thermal management reacts super fast now if there's an issue and if it's physically damaged then yeah of course that's not going to be covered under a warranty and I'm ok with that. fyi, Intel is offering a delidded warranty on these chips to certain companies, which again to me just says they are worried about the chips getting damaged during the delidding process more than anything else.
Roman mentions in the previous video about intel working with some SIs. About selling builds with delidded 14900ks, which will be covered under warranty.
@@Speak_Out_and_Remove_All_Doubt They’re offering that right now for their current CPU’s or do you mean 14900KS delidded for OEM’s? I know that there’s been some delidded CPU’s sold by Intel in the past right? Anyway I doubt that either of them will start selling delidded CPU’s for normal customers.
@@Simon_Denmark The warranty is only on the 14900KS and it's only for a select number of OEMs. But what I am talking about is creating a batch of special edition CPUs that were never bonded to the IHS in the first place so that delidding is not even required. That would be easy to do, just take them off the production line before the IHS is put on.
The reason the signal integrity is better when you leave the copper block further away from the memory traces is that those traces are impedance controlled as differential microstrips, which are top-layer trace pairs where the electric fields couple to the ground plane on the layer below and, to some extent, between the two traces. The impedance is based on the trace geometry and the relative permittivity (ε_r) of the dielectric materials surrounding the traces - in this case, the fiberglass between the top copper layer and second layer, and the soldermask and air above the top copper layer. When you place metal directly over the top of the traces, you're changing the transmission line type from a microstrip to a stripline, which significantly changes the characteristic impedance of the traces in that region. As an example, a 45Ω impedance single-ended microstrip (with a 0.1mm layer height, dielectric ε_r=4.1) suddenly becomes 24.4Ω impedance stripline when you have a copper block floating 0.1mm above it, e.g. if you press it to the board. That impedance discontinuity causes signal reflections, which will narrow the eye diagram and harm signal integrity, and the copper block also causes a significant rise in conductor losses - if the original trace had 0.5dB of conductor loss, you might see closer to 4.5dB of loss with the block pressed to the board. That loss alone may be as much as a third of the total loss budget of the DDR5 link, depending on the speed! No wonder it doesn't work. If you lift that copper block just 2mm off the board, the trace impedance jumps back to within 10% of the original and the conductor losses plummet.
and that's why some ppl. can run 8400+ DDR5 with no problem and some ppl. that are very experts in DDR4 fail to run even 7600MT. Things change and you need to keep up with new way to undervolt and overclock.
Because it's the "EN" channel, which he made not too long ago to upload the English versions of his videos. Before he uploaded everything to his normal channel, Der8auer, which is now used to upload the German videos exclusively (which still only has 483k subs, still too low for the quality of videos he produces tbh)
He is one of thew few "old-school" serious XOC left, actually testing the hardware with real tweaks, squeezing out everything from a system and not calling XMP profiles "tuned" settings. Always love the videos, especially the one electron microscope video was crazy, he should have gotten millions of views. I also love how the AIO is not marked as "watercooling" since AIOs are crap compared to even a simple custom loop.
Intel should just sell delidded cpus that come with a direct die frame at this point for like 30-40 usd more.Its cheaper for them to not have to give you a copper IHS and it removes the delidding risk and hassle for us while keeeping our warranty and have better temperatures.
When DD'ing I've now twice used a PTFE shim covering the PCB, of die thickness (minus perhaps 0.01mm), to ensure even clamping pressure and prevent pin contact issues from the PCB bowing up at the edges/corners. It seems to work well. As it happens I discovered this was a problem with the der8auer 9th gen frame, which is actually over 0.1mm thinner than the die and the entire mounting force/pressure of the cooler plate is exerted only on the die (hence people reporting POST, stability and memory bus issues). ETA >> bear in mind that with no ILM you have to ensure the required minimum total clamping pressure is exerted by the cooler, c. 40 lbs / 18 kg for LGA 115x (I imagine higher for 12/13/14th gen with the higher pin count) according to Intel. I modified the frame by cutting the centre out (which does nothing anyway, just bows up off the PCB) leaving only short 'tabs' so it holds the CPU in place when installing/uninstalling the cooler - without it, it would be easy to fumble when the CPU sticks to the cooler plate and trash the socket.
It would be nice to see a comparison of your DD block once it's available compared to the EK, Iceman Cooler and Supercool incumbents. I'd like to see the comparison include max core to core temperature deltas as well as an assessment on the repeatability of measurements over several installations, things people have struggled with when using DD blocks previously. I've personally tried all three and settled on the Supercool in my main rig, even with it making loop maintenance more challenging and requiring >200l/h flow. I never comment usually as a long time lurker, so wanted to take the opportunity to thank you for your content.
I have the EK block and I want to swap it out. I’m waiting for this but if it doesn’t drop soon I’m going to just get the Super Cool block. Hurry up!!!!! Lol
I really do wish we'd have some innovation in this area. Maybe with CPUs coming delidded by default like old CPUs, or some kind of different heat spreader.
94c at 370w 5.8ghz. Thats insane! That just proves, speed doesn’t equal performance. When you consider the 7800X3D delivers on average the same or better performance in gaming at 65-70c 95w 5.50ghz, this is what I’m getting on a full open loop.
Really really want to see a full copper+NI / copper & brass version of your block without rgb and the better cold plate. Basically your version of a heat killer IV pro full copper ❤. Obviously the price will be higher but it would just be awesome.
3:37 that's not why the cpu crashed. Intel has really violent transients during thermalthrottle. The cpu is crashing by itself due to massive frequency deviations and thus massive transient spikes. Its just an Intel issue that haven't changed a bit
Personally I think that if Intel was "smart" they would have always sold all the KS SKUs (starting the 9900KS or even the 8086K which is a 5 GHz out of the box edition of the 8700K) pre delidded since they are super binned chips and only enthusiast which want to direct die them should get them (the IHS should however still be in the box and sold alongisde the delidded CPU). Those CPUs have such insane power draws that direct die is the only "smart" thing to do if you purchase one, otherwise you should have gotten the regular K.
Don't really agree with the last statement. I cannot make a perfect comparison since I did not have 13900k, but do have currently 13900KS and had a 12900K. What people do not realise is that mostly KS variants are just binned K version of a chip, that being said... Running at the same let's say stock frequency they do require less voltages and if you re gonna undervolt it, you ll have better results than doing it with the K . My is currently running on - .04 v, and on LLC3 normal behaviour... Reckon could go even lower than that. SP prediction was 112, while with the K I never got anything higher than 95. Temperature wise, doesn't go nuts like 12900 did.. less jumping all over the place and takes undervoltng like a champ plus has a higher potential power draw
Really nice, last Intel CPU I did this too was a Devil Canyon, made a massive improvement on that one, the water block looks great, but should have some soft foam pads in the corners like the old AMD Athlons had to keep things level as you install it.
Hey, I've always noticed that you say that Hwinfo taking some of your CPU, but even on my little quad core CPU it barely takes any CPU usage, what is wrong with yours?
I adjust it to 500ms polling rate. Otherwise it only updates every 2 seconds which is too slow for what I want to test. If you increase the polling reate it takes more resources
While not directly related to this video, I can't wait for Ryzen 9000 to come out, will definitely delid it using your products! Let's hope they don't change anything so the old 7000 series tools still work just fine...
It's so frustrating that with power densities this high and dies this small, the majority of the temperature delta occurs between the bottom of the silicon and the top of the IHS. You can have the best loop in the world, but as long as we're talking ambient cooling you're always limited by getting the heat out to the water block in the first place. It would be lovely if AMD and Intel would sell versions of these CPUs with no IHS and a silicon die thinned as much as possible (since the thermal conductivity of the silicon itself is quite poor compared to copper). They could either charge a bit more to cover the expected warranty issues from people damaging it, or simply void warranties if there's any visible cracks/damage to the cores. It's clearly possible, both because they _still_ sell mobile CPUs this way, and because both vendors used to sell bare socketed desktop CPUs back in the Pentium 3 and Athlon XP era. Being able to place a thin layer of copper for the waterblock as close as possible to the transistors emitting the heat would do wonders to improve the situation where your processor instantly jumps 60-80c going from idle to full load (the thermal resistance you can't do anything about) and then only rises 5-10c after that (because your loop is plenty capable of dissipating the power). We could even dissipate way more, because hotter water means more efficient thermal transfer to the air but you can't normally tolerate a high water temperature as things are now because it doesn't give you sufficient headroom because you're desperately trying to keep things below throttling temperatures on the die itself.
Yeah for production having a big floating copper plane right there next to a bunch of high speed traces is going to be a nightmare for signal integrity. If you can net even a few mm on that one side I think it may mitigate the issue. The electric field should attenuate extremely quickly at such low potentials reducing cross talk and line capacitance.
If you replaced the water with a custom formulated coolant, you might get even better results. You should look into what sort of coolants are being developed.
Very good result. About the cinebench score, anyone noticed that we’ve lost a lot in multi core benchmark? I scored over 43k with my old 13900k with 5,7 GHz on P-Core, now 42K with 6 GHz…
Yeah, I have an average quality 14900k, and I can just break 41k with a manual vcore of only 1.320v. I planned to do the same when my KS arrives. Not interested in overclocking, really. I mean its fun/cool, but I have no purpose for it besides just personal enjoyment. I went through 4 14900k, and just said screw it on the 5th one. Too much of a pain chasing better silicon quality. So, I just got the 14900ks essentially because I wanted a better binned 14900k, which would be guaranteed without playing the silicon lottery. IMO, the little bit of price increase is 100% worth the tine and money you saving chasing better random 14900k.
Thermal grizzly needs there own quicksilver solder remover. It works so much better and faster then trying to use normal Liquid metal when trying to clean the CPU.
I got a small vial of galinstan from Lithuania via ebay years ago, 10mL cost 1/3 of one tube of Conductonaut, I'm still not even 1/4 way through it. Eats solder both when you want it to and when you don't 😑
Qtip silicon teflon gel onto cpu pcb get it hot with preheater then cool it down while applying more coatings then completely clean with clean qtips thermal as well as noise dampening for better response as well as efficiency
Technology is advancing alot faster then coolers can keep up . personally if I can't keep my cpu or gpu between 60-70deg cel MAX LOAD I'm not happy. Been using this template all my life and I've never had any hardware fail in the last 30 years
Sold on the new Delid-Die-Mate , went to check if i can get one now but no stock in the UK😆 Looks a much much better design than the EK one i have here that i used on 13th gen and the acrylic piece is a brilliant idea 😃 How long till they are back in stock as really want to delid this new KS ....
why not set cinebench at real-time priority on task manager? it hardware monitor will not use CPU during the bench mark, and you will get a temperature reading right after cinebench is done.
Slow pump flow or block flow completely monitored temps break it in properly as well as tune the function built in ai due its job while showing proper energy to adapt to
The CPU is definitely still throttling, but it's just not visible in HWinfo. My 13900k (even with some things running in the background, including HWinfo and FanControl) gets 42000 or very close to it in Cinebench R23, at 5.6-5.5Ghz P-core and 4.4Ghz E-core, so there's no way you get the same score if it's truly running the P-cores at 6Ghz... It can be very tricky to figure out if there's throttling under heavy load since the reporting becomea very delayed and isn't fast enough to begin with. The CPU can change clock speed many times in one polling update (even if you set polling to 100ms) and it wouldn't show normally, but when it's lagging like that because of the benchmark you can't even see what you normally would, but the score reveals lower performance than expected reliably.
Also Windows task manager since windows 8 doesn't show CPU busy time or utilization, but core utility (like you can also see in HWinfo64) that's capped to 100% (yes it's kinda stupid, and not normalized) so it's not 5% of available performance or CPU busy time, rather, it's compared to the work that would be possible to do if the CPU was at stock clock (no OC or Turbo Boost) except it'll not show more than 100% (HWinfo64 does).
If it's changing clockspeed then how is HWInfo not reporting it? Even if it changes many times in 1 polling rate, the chances of not even a single one of those hundreds, if not thousands, of changes being detected by HWInfo is beyond astronomical. Of course there'll be a ton of changes that HWInfo won't detect as you correctly pointed out, but how can there be 0 changes detected? If in, say, 10 seconds, the speed changes 10,000x, the chances of HWInfo not even detecting, say, 3,000 or, I don't know, let's say, 800 of those changes is beyond astronomical.
Yep thought the same thing - I have a 360 AIO and my 13900k will regularly hit 40k in c23 at 330 watts. Why is sucking so much more power without any more performance? Is the extra wattage really generating "so much" heat that it's down-clocking that much?
Also, check out Clock'EM UP (I think Vietnamese or Thai channel). He does 6.0, 6.1, and 6.2 all-core with a 13900KS and 14900K and at fairly low temps (high 60s to low 80s) yet his scores are also more inline with DerBauer's. Something doesn't make sense if a "measly" 5.5 GHz all-core is getting the same or higher score than a 6.0-6.2 GHz all-core direct-die cooled system.
@@Spinelli__ Because the clock speed you see reported is just BCLK clock times CPU core multiplier - it's not really a measurement. CPUs constantly switch states and clock speeds many times per second, even effectively turning off when doing nothing (known as core parking, though it can be disabled). What you should look at is the effective clock, though keep in mind it is an averaged value, not an instantaneous measurement, and under full load this will obviously not report every single cycle as the CPU will be fully busy. Either way this data isn't just a reading of some sensor, so you can't rely on it fully, which is why the most reliable way is to simply test how long it takes to complete some known amount of work, that doesn't rely on polling or reporting or anything like that, and also doesn't require extra software. Sometimes the clock will be lowered long enough for HWinfo64 to report the decrease, but in those cases the throttling will already be very obvious.
Can I request a direct die comparison between LM, Kryosheet and Hydronaut please? I have your delid kit and the EK direct die conversion kit but would be interested to know if a non LM option is viable as it won't then 'dry out'. I don't want to have to reapply LM every year 😢
For best conducting properties, I use the metal part of JBWELD.
6 หลายเดือนก่อน
@der8auer-en Supercoolcomputer has two direct die block for intel 14, acrylic and full copper C1100 ( what they claim). The cost for SC block & delid tool on supercool are $140 plus ~ $30 shipping fee ( DHL express), that is without any tax or hassle, front door delivery. It takes 3 days to receive in located in Asia and within a week if located in UK. I wonder how much are yours since there is 19% VAT on top on shipping from TG. I hope Roman's stuff are quite better than Supercoolcomputer and the price is cheaper, $120 for both would be great . Supercool's coldplate is quite thin so you may wanna consider that. Can't wait for some more comparison. Edit: Nvm, i don't think this is gonna fly bc no ILM, contact is gonna be big issue. You do know >8000MT, impedance is very annoying to deal with. A $50 Torque drive is far from getting it down 0.05NM precise.
Idea and question at same time: why nobody will do direct die air coolers, I remember old times of Athlon CPUs, where there was direct die, and everything was ok, and they was also super overclockable. Here I believe it will help to cool a lot to, and no water inside pc, problems with mixing metals, so super expensive water cooling ;-)
Any update on when these water blocks might release? EKWB has kept me waiting nearly 4 months on a direct die AIO pre-order. Would love to get something quality in my system.
I had to RMA my 14900ks - without adding some LLC value, I couldn't get it stable even at Intel limits (NH-D15) so hopefully the replacement will be better.
@@dragonsyph2557 unlikely. The second 14900ks was just as bad, whereas putting a 13900ks or 13900k in the same system was fine. My other system is running a 14909k and that works perfectly. I have seen one guy try 3 14900ks chips, something doesn't seem right with the sku, perhaps they just cannot work without an amount of custom tweaks. They should at least work at Intel power limits with no other changes.
@@dragonsyph2557 unlikely. The second 14900ks was just as bad, whereas putting a 13900ks or 13900k in the same system was fine. My other system is running a 14909k and that works perfectly. I have seen one guy try 3 14900ks chips, something doesn't seem right with the sku, perhaps they just cannot work without an amount of custom tweaks. They should at least work at Intel power limits with no other changes.
1) that end run with the one p core hitting 96c while the rest are 10c cooler or more is such a shame, it limits everything. Perhaps clipping that ones frequency down a bin will let the others go to 61. 2) Very low scores for the clocks being run, very bizarre. But it makes sense, because the end run the max core eff clock for p cores was only 5935. So not true 6000. But on the prior 58x/45x profile, being sub 41k points is very concerning. My 13900Ks at normal 56x and a minor bump to 44x e cores, gets 41k-41.85k points max (depending on priority and R2R variance). IE. Normal Prio is around 41.2-41.4, High Prio gets me that 41.6-41.8 range. Not delidded or anything, maxes a little over 90c on 4 hot cores.
@@toonnut1 Liquid Metal (at least conductonaut doesn't) won’t dissolve copper. It etches into it and improves thermal transfer. You just can’t ever clean it out.
Hello Roman I just want to know when the Intel's Mycro Direct Die would be available please. I just see AM5 model but no LGA1700 to purchase . Kind regards, Max.
I find it interesting that having a larger gap between the direct die block ILM and the memory traces improves the signaling enough to make a large difference. I have tried icemancooler block from aliexpress which was around 80 USD and has a thick plastic skirt and it works fine at 8200+ on apex encore even with a relatively weak IMC. I then tried using a direct die frame from them and a modified optimus foundation block with a hole dremeled for SMD capacitor clearance and it would struggle to even boot 8000 and would error really fast in ycruncher at any torque values that I tried. I have not used liquid metal with either as im using a chiller with the capability of going subzero but the best paste results I have had were PTM7950 and any really thick paste like thermalrights lineup. Kryonaut and KPX were both fairly bad for core-core delta and water-core temp deltas. I got similar temperature results to the ones in your video with 14900kf but about 5-10c worse on the hottest core due to not using liquid metal. Also that cinebench score of 42k is a bit low for 6ghz but it can be explained by lower E core clocks probably.
Oh my, the flowmeter from aquacomputer is way smaller than I thought. I'm going to buy this one also, together with the quadro interface to steer my PWM fans, and my EK pump. But I'm not sure to use the DRGB option, because my case will already have an overload of ARGB. I must say the temperatures dropped significantly then your other video. But it's concerning you have to use water cooling to cool the damn thing. For regular users, this will become a problem.
Awesome results, I will be looking for your cooler, my 13900K is delided and direct die cooled, but it is a plexie unit, I like the looks of your better and hopefully will see even a few more C improvement, but I am already pretty low as I also use a chiller on a tank.
HI. I also delid on 149 kf. I DID EVERYTHING SAME AS YOU. I used the grizly tool IL LOOP EK DIRECT DIE. the temperatures are 30 degrees lower...but if I DO BENCHMARK IT CRASHES INSTANTLY. can you give me some advice. WHAT SHOULD I CHANGE FROM BIOS? I'M ASKING FOR HELP. I CAN NO LONGER USE THE DESKTOP PC
I don't understand why Intel don't sell the KS variant with direct die. It's practically a given that you need the direct die to fully utilise the potential. For more regular customers there is the K,for extreme one the KS, simple. Like who buy a KS to using on a aio? It's like buying a rally car and don't race.
10:10 At the beginning he said he keeps the pump speed constant near 150L/h (it was like 142 though) now it's at 172L/h & I'm stuck wishing it was constant across his testing so it doesn't skew the data
I always think it's a shame companies like Intel/AMD don't offer a special edition version that was never 'lidded' as it's the delidding process and the cleaning up the die that is the hassle, scary part and where most of the risk is.
You’ll never get a warranty for a delidded CPU/CPU without IHS from manufacturing. Are you fine with that?
@@Simon_Denmark I disagree, Intel's thermal management reacts super fast now if there's an issue and if it's physically damaged then yeah of course that's not going to be covered under a warranty and I'm ok with that.
fyi, Intel is offering a delidded warranty on these chips to certain companies, which again to me just says they are worried about the chips getting damaged during the delidding process more than anything else.
Roman mentions in the previous video about intel working with some SIs. About selling builds with delidded 14900ks, which will be covered under warranty.
@@Speak_Out_and_Remove_All_Doubt They’re offering that right now for their current CPU’s or do you mean 14900KS delidded for OEM’s? I know that there’s been some delidded CPU’s sold by Intel in the past right? Anyway I doubt that either of them will start selling delidded CPU’s for normal customers.
@@Simon_Denmark The warranty is only on the 14900KS and it's only for a select number of OEMs.
But what I am talking about is creating a batch of special edition CPUs that were never bonded to the IHS in the first place so that delidding is not even required. That would be easy to do, just take them off the production line before the IHS is put on.
7:06 As a Polish man i agree we are good at this.
Polska!
I was going to ask if you could use a Slovak instead.
And the labour is surprisingly cheap
7:14 .... I don't think that's legal, Roman...
Lmfao caught that too
The reason the signal integrity is better when you leave the copper block further away from the memory traces is that those traces are impedance controlled as differential microstrips, which are top-layer trace pairs where the electric fields couple to the ground plane on the layer below and, to some extent, between the two traces. The impedance is based on the trace geometry and the relative permittivity (ε_r) of the dielectric materials surrounding the traces - in this case, the fiberglass between the top copper layer and second layer, and the soldermask and air above the top copper layer. When you place metal directly over the top of the traces, you're changing the transmission line type from a microstrip to a stripline, which significantly changes the characteristic impedance of the traces in that region.
As an example, a 45Ω impedance single-ended microstrip (with a 0.1mm layer height, dielectric ε_r=4.1) suddenly becomes 24.4Ω impedance stripline when you have a copper block floating 0.1mm above it, e.g. if you press it to the board. That impedance discontinuity causes signal reflections, which will narrow the eye diagram and harm signal integrity, and the copper block also causes a significant rise in conductor losses - if the original trace had 0.5dB of conductor loss, you might see closer to 4.5dB of loss with the block pressed to the board. That loss alone may be as much as a third of the total loss budget of the DDR5 link, depending on the speed! No wonder it doesn't work. If you lift that copper block just 2mm off the board, the trace impedance jumps back to within 10% of the original and the conductor losses plummet.
Dang, I knew the tolerances were important but didn't know it could change that much, thank you for the detailed explanation!
I am not awake enough to even begin to understand this, but thanks 😂
Me too
and that's why some ppl. can run 8400+ DDR5 with no problem and some ppl. that are very experts in DDR4 fail to run even 7600MT. Things change and you need to keep up with new way to undervolt and overclock.
This is some next level comment, thank you
Why is this guy not over a million subs... I have no idea.
Because it's the "EN" channel, which he made not too long ago to upload the English versions of his videos. Before he uploaded everything to his normal channel, Der8auer, which is now used to upload the German videos exclusively (which still only has 483k subs, still too low for the quality of videos he produces tbh)
He is one of thew few "old-school" serious XOC left, actually testing the hardware with real tweaks, squeezing out everything from a system and not calling XMP profiles "tuned" settings.
Always love the videos, especially the one electron microscope video was crazy, he should have gotten millions of views.
I also love how the AIO is not marked as "watercooling" since AIOs are crap compared to even a simple custom loop.
@@marios3761 It's not the quantity of subs that counts, it's the quality. 😉
I never even notice that he had so little subs. Really just assumed he had more than a million. Kinda crazy how youtube works...
one of my favorite aswell, straight to the point and with good products
13 degrees is just insane.
yeah 400 watts too :D
@@thomasvennekens4137 i dont think this channel is trying to sell you intel. Hes just showing you the insane performance of the cooling technique.
Woo, that is very hot and power hungry sand.
This CPU should be Special Edition (water cooling required)
It pretty much is. You would be a fool to buy a KS and air cool it.
You will see a lot better temps even on an air cooler if you're going bare die, don't think water cooling would be necessary in this situation
Don't give intel any ideas next thing they'll do is branch off and make 15(1-9k) and a second run of 15(1-9w) for waterblock edition lol
@@nickshaffer7659 haha yea you never know..rolleyes. profit profit profit.
@@trixter192You would have to be a fool to buy this cpu in the first place
Intel should just sell delidded cpus that come with a direct die frame at this point for like 30-40 usd more.Its cheaper for them to not have to give you a copper IHS and it removes the delidding risk and hassle for us while keeeping our warranty and have better temperatures.
Aye
When DD'ing I've now twice used a PTFE shim covering the PCB, of die thickness (minus perhaps 0.01mm), to ensure even clamping pressure and prevent pin contact issues from the PCB bowing up at the edges/corners. It seems to work well. As it happens I discovered this was a problem with the der8auer 9th gen frame, which is actually over 0.1mm thinner than the die and the entire mounting force/pressure of the cooler plate is exerted only on the die (hence people reporting POST, stability and memory bus issues). ETA >> bear in mind that with no ILM you have to ensure the required minimum total clamping pressure is exerted by the cooler, c. 40 lbs / 18 kg for LGA 115x (I imagine higher for 12/13/14th gen with the higher pin count) according to Intel. I modified the frame by cutting the centre out (which does nothing anyway, just bows up off the PCB) leaving only short 'tabs' so it holds the CPU in place when installing/uninstalling the cooler - without it, it would be easy to fumble when the CPU sticks to the cooler plate and trash the socket.
We need an AIO Water Cooler for Direct Die cooling
I'd love to see gaming benchmarks for this beast, overclocked and direct die
that acrilic piece for removing the glue is amazing
9:37 really crazy to think about, when you’re on the bleeding edge, every little bit counts
I am impressed. Well done. Thanks.
Roman this is the type of temp testing and research development in this industry I love!
Excellent result and great testing like always. Sorry we didn't get to see the nice cats this time...
It would be nice to see a comparison of your DD block once it's available compared to the EK, Iceman Cooler and Supercool incumbents. I'd like to see the comparison include max core to core temperature deltas as well as an assessment on the repeatability of measurements over several installations, things people have struggled with when using DD blocks previously. I've personally tried all three and settled on the Supercool in my main rig, even with it making loop maintenance more challenging and requiring >200l/h flow. I never comment usually as a long time lurker, so wanted to take the opportunity to thank you for your content.
Great that you're showing every step of the delid process including cleaning the die! 😊👍
I hope I live to the day when Thermal Grizzly makes 420 AIOs :D
when is direct die for sell for 13th/14th gen intel ?
I have the EK block and I want to swap it out. I’m waiting for this but if it doesn’t drop soon I’m going to just get the Super Cool block. Hurry up!!!!! Lol
Been waiting for this as well & will also get a supercool block if this does not release soon.
I really do wish we'd have some innovation in this area. Maybe with CPUs coming delidded by default like old CPUs, or some kind of different heat spreader.
94c at 370w 5.8ghz.
Thats insane! That just proves, speed doesn’t equal performance.
When you consider the 7800X3D delivers on average the same or better performance in gaming at 65-70c 95w 5.50ghz, this is what I’m getting on a full open loop.
im getting old pentium 4 netburst wibes from these new intel space heaters
Yep, the old days are coming back... Little generational improvement for much more power.
(inserts the Captain America "I understood that reference" meme)
Funny enough, the 1.525v core voltage he sets is the stock voltage for Northwood Pentium 4s.
I do miss the 1+ GHz overclocks we used to see though.
They were "only" around 200-250W, not 400W.
Meme chips, excited for the new gen tho, also zen 5.
I'm using the AM5 Mycro myself, couldn't be happier! Great to see this line of product expanding!
Thanks :) Happy to hear that
Really really want to see a full copper+NI / copper & brass version of your block without rgb and the better cold plate.
Basically your version of a heat killer IV pro full copper ❤. Obviously the price will be higher but it would just be awesome.
3:37 that's not why the cpu crashed. Intel has really violent transients during thermalthrottle. The cpu is crashing by itself due to massive frequency deviations and thus massive transient spikes. Its just an Intel issue that haven't changed a bit
that got fixed with 11th gen AFAIK. The core ratio changing no longer causes massive transients like it did on 10th gen and older parts.
Personally I think that if Intel was "smart" they would have always sold all the KS SKUs (starting the 9900KS or even the 8086K which is a 5 GHz out of the box edition of the 8700K) pre delidded since they are super binned chips and only enthusiast which want to direct die them should get them (the IHS should however still be in the box and sold alongisde the delidded CPU).
Those CPUs have such insane power draws that direct die is the only "smart" thing to do if you purchase one, otherwise you should have gotten the regular K.
Don't really agree with the last statement.
I cannot make a perfect comparison since I did not have 13900k, but do have currently 13900KS and had a 12900K.
What people do not realise is that mostly KS variants are just binned K version of a chip, that being said...
Running at the same let's say stock frequency they do require less voltages and if you re gonna undervolt it, you ll have better results than doing it with the K .
My is currently running on - .04 v, and on LLC3 normal behaviour... Reckon could go even lower than that.
SP prediction was 112, while with the K I never got anything higher than 95.
Temperature wise, doesn't go nuts like 12900 did.. less jumping all over the place and takes undervoltng like a champ plus has a higher potential power draw
This is so impressive. You are very inspiring!
Really nice, last Intel CPU I did this too was a Devil Canyon, made a massive improvement on that one, the water block looks great, but should have some soft foam pads in the corners like the old AMD Athlons had to keep things level as you install it.
Always liking videos first then proceed to watch them. Never regretted doing so. Thank you Roman for greatest tech content.
Hey, I've always noticed that you say that Hwinfo taking some of your CPU, but even on my little quad core CPU it barely takes any CPU usage, what is wrong with yours?
I adjust it to 500ms polling rate. Otherwise it only updates every 2 seconds which is too slow for what I want to test. If you increase the polling reate it takes more resources
I came here for the cats but couldn't find them.
Increased spacing off the board improving the memory signaling is probably not about heat, but about inductive coupling with the block.
How do you connect the water block to a cooler? Do you desmantle another one? If yes, don;t you lose liquid?
in Mykro can please provide seal / gasket for Liquid Metal like Minisforum did on EM680 for added peace of mind
While not directly related to this video, I can't wait for Ryzen 9000 to come out, will definitely delid it using your products!
Let's hope they don't change anything so the old 7000 series tools still work just fine...
i wish we had the same test with the new contact frame
It's so frustrating that with power densities this high and dies this small, the majority of the temperature delta occurs between the bottom of the silicon and the top of the IHS. You can have the best loop in the world, but as long as we're talking ambient cooling you're always limited by getting the heat out to the water block in the first place.
It would be lovely if AMD and Intel would sell versions of these CPUs with no IHS and a silicon die thinned as much as possible (since the thermal conductivity of the silicon itself is quite poor compared to copper). They could either charge a bit more to cover the expected warranty issues from people damaging it, or simply void warranties if there's any visible cracks/damage to the cores. It's clearly possible, both because they _still_ sell mobile CPUs this way, and because both vendors used to sell bare socketed desktop CPUs back in the Pentium 3 and Athlon XP era.
Being able to place a thin layer of copper for the waterblock as close as possible to the transistors emitting the heat would do wonders to improve the situation where your processor instantly jumps 60-80c going from idle to full load (the thermal resistance you can't do anything about) and then only rises 5-10c after that (because your loop is plenty capable of dissipating the power). We could even dissipate way more, because hotter water means more efficient thermal transfer to the air but you can't normally tolerate a high water temperature as things are now because it doesn't give you sufficient headroom because you're desperately trying to keep things below throttling temperatures on the die itself.
They should really just make a special version where the IHS doubles as a waterblock.
No delidding, no tricky mounting/thermalcompound.
I wonder what kind of performance we can see with liquid metal, copper ihs, frame corrector, and the ek delta tec block 🤔
Yeah for production having a big floating copper plane right there next to a bunch of high speed traces is going to be a nightmare for signal integrity. If you can net even a few mm on that one side I think it may mitigate the issue. The electric field should attenuate extremely quickly at such low potentials reducing cross talk and line capacitance.
If you replaced the water with a custom formulated coolant, you might get even better results. You should look into what sort of coolants are being developed.
Very good result. About the cinebench score, anyone noticed that we’ve lost a lot in multi core benchmark? I scored over 43k with my old 13900k with 5,7 GHz on P-Core, now 42K with 6 GHz…
Interesting
1.52V is nuts, as are the gains from this (temp drops, clock speeds, and power draw). Nothing like direct-die!
It's not 1.52v. Remember, that's the voltage before vdroop. It's probably in the 1.40-1.47 V area.
You can get a cinebench r23 score 42k+ with all core 5.7 I'm confused why that score is so low for 6ghz
Hwinfo will influence results. Also, he might be recording using obs
i think this cpu is saying 6ghz but not really reaching it, after all it is a overclock over a overclock over a overclock
Yeah, I have an average quality 14900k, and I can just break 41k with a manual vcore of only 1.320v.
I planned to do the same when my KS arrives. Not interested in overclocking, really. I mean its fun/cool, but I have no purpose for it besides just personal enjoyment. I went through 4 14900k, and just said screw it on the 5th one. Too much of a pain chasing better silicon quality. So, I just got the 14900ks essentially because I wanted a better binned 14900k, which would be guaranteed without playing the silicon lottery. IMO, the little bit of price increase is 100% worth the tine and money you saving chasing better random 14900k.
EXACTLY, my stock 13900ks scores 42.5k. Can reach almost 45k OC. On an LT720 AIO.
Thermal grizzly needs there own quicksilver solder remover. It works so much better and faster then trying to use normal Liquid metal when trying to clean the CPU.
That's not in their best business intrest 😂
@toonnut1 Yea, having to buy 2 or 3 things of liquid metal because it will take a lot to clean it. Then you need more to use as thermal paste 😂
I got a small vial of galinstan from Lithuania via ebay years ago, 10mL cost 1/3 of one tube of Conductonaut, I'm still not even 1/4 way through it. Eats solder both when you want it to and when you don't 😑
You should have tried single core 6.5 GHz.
Qtip silicon teflon gel onto cpu pcb get it hot with preheater then cool it down while applying more coatings then completely clean with clean qtips thermal as well as noise dampening for better response as well as efficiency
Technology is advancing alot faster then coolers can keep up . personally if I can't keep my cpu or gpu between 60-70deg cel MAX LOAD I'm not happy. Been using this template all my life and I've never had any hardware fail in the last 30 years
Sold on the new Delid-Die-Mate , went to check if i can get one now but no stock in the UK😆 Looks a much much better design than the EK one i have here that i used on 13th gen and the acrylic piece is a brilliant idea 😃 How long till they are back in stock as really want to delid this new KS ....
sooom nerdy, sooo not needed, SOOO GREAT!!!
why not set cinebench at real-time priority on task manager? it hardware monitor will not use CPU during the bench mark, and you will get a temperature reading right after cinebench is done.
VRM's dissipating maybe 40W on their own, must need a bit of cooling themselves.
Direct Die Cooling always wins.
It's almost a necesity these days with the Intel heaters and AMD super thick heatspreader.
you can count me in on grabbing one of those blocks im gonna compare it to ek direct die velocity 2
When i want factual videos. This i where i come. Thanks
This water block is literally the last piece to my 1000d build.
Any update on when it's releasing?
Lot's of fun doing these things...
Slow pump flow or block flow completely monitored temps break it in properly as well as tune the function built in ai due its job while showing proper energy to adapt to
Didn't know about using the thermal paste to polish trick, will be trying that
I don't know why Intel bothers even putting an IHS on it.
It will tile adapt better due to having a range of memory timings running independently
This looks promising - have 14900ks in hand - looking forward to the mycro 1700 block
The CPU is definitely still throttling, but it's just not visible in HWinfo. My 13900k (even with some things running in the background, including HWinfo and FanControl) gets 42000 or very close to it in Cinebench R23, at 5.6-5.5Ghz P-core and 4.4Ghz E-core, so there's no way you get the same score if it's truly running the P-cores at 6Ghz...
It can be very tricky to figure out if there's throttling under heavy load since the reporting becomea very delayed and isn't fast enough to begin with. The CPU can change clock speed many times in one polling update (even if you set polling to 100ms) and it wouldn't show normally, but when it's lagging like that because of the benchmark you can't even see what you normally would, but the score reveals lower performance than expected reliably.
Also Windows task manager since windows 8 doesn't show CPU busy time or utilization, but core utility (like you can also see in HWinfo64) that's capped to 100% (yes it's kinda stupid, and not normalized) so it's not 5% of available performance or CPU busy time, rather, it's compared to the work that would be possible to do if the CPU was at stock clock (no OC or Turbo Boost) except it'll not show more than 100% (HWinfo64 does).
If it's changing clockspeed then how is HWInfo not reporting it? Even if it changes many times in 1 polling rate, the chances of not even a single one of those hundreds, if not thousands, of changes being detected by HWInfo is beyond astronomical. Of course there'll be a ton of changes that HWInfo won't detect as you correctly pointed out, but how can there be 0 changes detected? If in, say, 10 seconds, the speed changes 10,000x, the chances of HWInfo not even detecting, say, 3,000 or, I don't know, let's say, 800 of those changes is beyond astronomical.
Yep thought the same thing - I have a 360 AIO and my 13900k will regularly hit 40k in c23 at 330 watts. Why is sucking so much more power without any more performance? Is the extra wattage really generating "so much" heat that it's down-clocking that much?
Also, check out Clock'EM UP (I think Vietnamese or Thai channel). He does 6.0, 6.1, and 6.2 all-core with a 13900KS and 14900K and at fairly low temps (high 60s to low 80s) yet his scores are also more inline with DerBauer's. Something doesn't make sense if a "measly" 5.5 GHz all-core is getting the same or higher score than a 6.0-6.2 GHz all-core direct-die cooled system.
@@Spinelli__ Because the clock speed you see reported is just BCLK clock times CPU core multiplier - it's not really a measurement. CPUs constantly switch states and clock speeds many times per second, even effectively turning off when doing nothing (known as core parking, though it can be disabled).
What you should look at is the effective clock, though keep in mind it is an averaged value, not an instantaneous measurement, and under full load this will obviously not report every single cycle as the CPU will be fully busy.
Either way this data isn't just a reading of some sensor, so you can't rely on it fully, which is why the most reliable way is to simply test how long it takes to complete some known amount of work, that doesn't rely on polling or reporting or anything like that, and also doesn't require extra software.
Sometimes the clock will be lowered long enough for HWinfo64 to report the decrease, but in those cases the throttling will already be very obvious.
I'd love to see a video with the 14900KS cooled by the Delta Tec²!!!
what a great video derbauer keep it up mate :)
Can I request a direct die comparison between LM, Kryosheet and Hydronaut please? I have your delid kit and the EK direct die conversion kit but would be interested to know if a non LM option is viable as it won't then 'dry out'. I don't want to have to reapply LM every year 😢
Maybe a package with the die mate too?
6ghz all core without overheating? Awesome, well done
At 1.52v😂
@@bubbamcgee5953It's not 1.52 V. That's just what's set in BIOS. It's prob somewhere between 1.38 and 1.47 V.
For best conducting properties, I use the metal part of JBWELD.
@der8auer-en Supercoolcomputer has two direct die block for intel 14, acrylic and full copper C1100 ( what they claim). The cost for SC block & delid tool on supercool are $140 plus ~ $30 shipping fee ( DHL express), that is without any tax or hassle, front door delivery. It takes 3 days to receive in located in Asia and within a week if located in UK. I wonder how much are yours since there is 19% VAT on top on shipping from TG. I hope Roman's stuff are quite better than Supercoolcomputer and the price is cheaper, $120 for both would be great . Supercool's coldplate is quite thin so you may wanna consider that. Can't wait for some more comparison. Edit: Nvm, i don't think this is gonna fly bc no ILM, contact is gonna be big issue. You do know >8000MT, impedance is very annoying to deal with. A $50 Torque drive is far from getting it down 0.05NM precise.
dont run currrent at 140 and run llc7, youll be able to run 6ghz at 1.4v, this is what i run on my 14900k that has a 87sp
Idea and question at same time: why nobody will do direct die air coolers, I remember old times of Athlon CPUs, where there was direct die, and everything was ok, and they was also super overclockable. Here I believe it will help to cool a lot to, and no water inside pc, problems with mixing metals, so super expensive water cooling ;-)
Any update on when these water blocks might release? EKWB has kept me waiting nearly 4 months on a direct die AIO pre-order. Would love to get something quality in my system.
I had to RMA my 14900ks - without adding some LLC value, I couldn't get it stable even at Intel limits (NH-D15) so hopefully the replacement will be better.
User Error.
@@dragonsyph2557 unlikely. The second 14900ks was just as bad, whereas putting a 13900ks or 13900k in the same system was fine. My other system is running a 14909k and that works perfectly. I have seen one guy try 3 14900ks chips, something doesn't seem right with the sku, perhaps they just cannot work without an amount of custom tweaks. They should at least work at Intel power limits with no other changes.
@@dragonsyph2557 unlikely. The second 14900ks was just as bad, whereas putting a 13900ks or 13900k in the same system was fine. My other system is running a 14909k and that works perfectly. I have seen one guy try 3 14900ks chips, something doesn't seem right with the sku, perhaps they just cannot work without an amount of custom tweaks. They should at least work at Intel power limits with no other changes.
1) that end run with the one p core hitting 96c while the rest are 10c cooler or more is such a shame, it limits everything. Perhaps clipping that ones frequency down a bin will let the others go to 61.
2) Very low scores for the clocks being run, very bizarre. But it makes sense, because the end run the max core eff clock for p cores was only 5935. So not true 6000.
But on the prior 58x/45x profile, being sub 41k points is very concerning.
My 13900Ks at normal 56x and a minor bump to 44x e cores, gets 41k-41.85k points max (depending on priority and R2R variance). IE. Normal Prio is around 41.2-41.4, High Prio gets me that 41.6-41.8 range. Not delidded or anything, maxes a little over 90c on 4 hot cores.
Pretty sweet tool that lasercut plexi
- Mad respect to taming the 14900KS .
This makes me wonder if there’s even a need for IHS if it worsens the CPU temp as a whole …
where do u get that pusher to get off the glue? its in the direct die mate?
0:18 You know der8auer is German because he said "we will today" instead of "today we will" :D
good for you to flog good products which are really needed
I've heard theyre working on 10ghz using liquid fertilizer
You should offer a bare copper version as well!
What would be the benefits of that 😂😂😂😂😂 especially if using liquid metal 😂😂😂😂😂😂
@@toonnut1 Liquid Metal (at least conductonaut doesn't) won’t dissolve copper. It etches into it and improves thermal transfer. You just can’t ever clean it out.
Hello Roman
I just want to know when the Intel's Mycro Direct Die would be available please.
I just see AM5 model but no LGA1700 to purchase .
Kind regards, Max.
I find it interesting that having a larger gap between the direct die block ILM and the memory traces improves the signaling enough to make a large difference. I have tried icemancooler block from aliexpress which was around 80 USD and has a thick plastic skirt and it works fine at 8200+ on apex encore even with a relatively weak IMC. I then tried using a direct die frame from them and a modified optimus foundation block with a hole dremeled for SMD capacitor clearance and it would struggle to even boot 8000 and would error really fast in ycruncher at any torque values that I tried. I have not used liquid metal with either as im using a chiller with the capability of going subzero but the best paste results I have had were PTM7950 and any really thick paste like thermalrights lineup. Kryonaut and KPX were both fairly bad for core-core delta and water-core temp deltas. I got similar temperature results to the ones in your video with 14900kf but about 5-10c worse on the hottest core due to not using liquid metal. Also that cinebench score of 42k is a bit low for 6ghz but it can be explained by lower E core clocks probably.
See comment by gsuberland
6GHz OC 14900KS looks nice and i think it can be 6.2 Ghz all cores in gaming . This is insane
Oh my, the flowmeter from aquacomputer is way smaller than I thought. I'm going to buy this one also, together with the quadro interface to steer my PWM fans, and my EK pump. But I'm not sure to use the DRGB option, because my case will already have an overload of ARGB. I must say the temperatures dropped significantly then your other video. But it's concerning you have to use water cooling to cool the damn thing. For regular users, this will become a problem.
Nothing more than a 13900k, special bin.
No shit.
Awesome results, I will be looking for your cooler, my 13900K is delided and direct die cooled, but it is a plexie unit, I like the looks of your better and hopefully will see even a few more C improvement, but I am already pretty low as I also use a chiller on a tank.
My stock 13900ks scores higher then his OC 14900ks, something wrong in this video.
@@dragonsyph2557 it's pretty much the same chip you probably just got a nice bin
HI. I also delid on 149 kf. I DID EVERYTHING SAME AS YOU. I used the grizly tool IL LOOP EK DIRECT DIE. the temperatures are 30 degrees lower...but if I DO BENCHMARK IT CRASHES INSTANTLY. can you give me some advice. WHAT SHOULD I CHANGE FROM BIOS? I'M ASKING FOR HELP. I CAN NO LONGER USE THE DESKTOP PC
My ryzen 7950x make 38000 point with cheap Deepcool AS500 air cooler after undervolting
I costs way to much money to buy this processor, cool it and run it properly but damn it is always fun to see how much it can be pushed.
You're not the target market, sorry sparky
You will have better results on Cinebench if you set the process priority to Realtime in the Windows task manager
40000 pts with 400 watt in cinebench :-) crazy. I get 36000 pts withj a 7950x at about 150w (curve optimized to -20).
I don't understand why Intel don't sell the KS variant with direct die. It's practically a given that you need the direct die to fully utilise the potential. For more regular customers there is the K,for extreme one the KS, simple.
Like who buy a KS to using on a aio? It's like buying a rally car and don't race.
It's true ks should be overclockers only
G'day Sheik, Makita & Roman
7:38 🤔I thought you were going to say KPx 😂😂😂😂😂
10:10 At the beginning he said he keeps the pump speed constant near 150L/h (it was like 142 though) now it's at 172L/h & I'm stuck wishing it was constant across his testing so it doesn't skew the data
If you use different blocks you will get different flow rates at constant pump speed. Because some are far more restrictive than others.
The Pump is running at the same speed. But it depends on the components you use in the loop how it affects the resulting water flow
@@der8auer-en thank you for explaining that
No kitty this time?