9900KF Delid & Direct

A

azrael.arach

Limp Gawd
Joined
Feb 17, 2011
Messages
152
Anyone yet on a 9900k or 9900kf do direct die cooling? Wednesday parts come in to try. Right now I have my 9900kf at 5ghz @ 1.295 volts in bios, load about 1.27. Current temps in realbench top 68 degrees, prime95 I hit 76 to 77. Any tips by anyone that has done it.
 
The lid is soldered on.
It can be removed but the risk is much greater.
 
Yep, did a lot of reading and even went up to performance pc shop as they are hour from me. Seems solder is soft enough and I will be using the quicksilver solder remove method instead of razor blade.
 
Nenu said:
The lid is soldered on.
It can be removed but the risk is much greater.
Click to expand...
As long as you use a proper delidder its a low chance of any issues, Cecil on here has done dozens of them with no problems (though he does liquid metal and replaces the IHS with a copper one generally).

azrael.arach said:
Yep, did a lot of reading and even went up to performance pc shop as they are hour from me. Seems solder is soft enough and I will be using the quicksilver solder remove method instead of razor blade.
Click to expand...
Correct
 
Nasgul said:
The CPU runs cool enough and if you delid it, I highly doubt that it'll get you higher clocks pass 5.4GHz. I see no reason to do that, unless you have a 7700K.
Worried about temps? Then I'd recommend an NH-D15 or Dark Pro 3. Or a Corsair 280MM RAD, which is what I'll be using my 9900KF with. And Gelid Extreme.
Click to expand...
Eh, all depends what you're happy with, I'm stable 5.0 w/2 avx with a H110i 280mm aio, it rarely hits 70c gaming and such, but it gets uncomfortably hot on stress testing 80-85c.
Some folks want temps to be under 70c full bore, thats on top of I have what I consider a good chip since I only needed 1.26v to get there.
When/if I dedlidded and copper topped, I'd expect a drop of double digit celcius temps off the load temps (Cecill claims several of them dropped 25c-30c on the high clock/voltage testing).

Cecil said:
The problem is Intel is using too much gasket material when mounting the IHS. It's not letting enough pressure on the actual die.

If anyone wants to have theirs delidded and a copper IHS put on with liquid metal just send me a message. I could do it for about $60. The 9900k I have now stock at 5ghz throttled like crazy even just in cinebench. After replacing the IHS it's 5.2Ghz Prime95 with AVX stable and cool. Dropped at least 25-30C.

Also for temps, mid 80s is perfectly fine when stressing. That would equal 70s under any other heavy load situation.
Click to expand...
 
Cooling cpu is an ekwb velocity with ek xe 360 rad. As in original post in 70's with avx - 1.
The chip has more in it just summer heat sucks and I have always thought cooler is better at any speed.
 
9900k are soldered. There is no reason to delid. You might get a couple of degrees better.

Your temps look good.

And Intel's IHS are nickel plated copper. There is no reason to buy a copper IHS. And liquid metal products are better with nickel plating than bare copper.
 
chameleoneel said:
9900k are soldered. There is no reason to delid. You might get a couple of degrees better.

Your temps look good.

And Intel's IHS are nickel plated copper. There is no reason to buy a copper IHS. And liquid metal products are better with nickel plating than bare copper.
Click to expand...
I disagree, depends on how well your chip was manufactured: https://hardforum.com/goto/post?id=1044131271#post-1044131271

Most liquid metal compounds have no adverse side effects on copper other than staining, nickel plating is just to reduce/prevent oxidation of the copper core.
The nickel plating is so thin it can often be ignored, that being said copper has over 4x better at thermal conduction/heat transfer so replacing the IHS with a copper one that part I agree is only a couple of deg difference.
 
Spartacus09 said:
I disagree, depends on how well your chip was manufactured: https://hardforum.com/goto/post?id=1044131271#post-1044131271

Most liquid metal compounds have no adverse side effects on copper other than staining, nickel plating is just to reduce/prevent oxidation of the copper core.
The nickel plating is so thin it can often be ignored, that being said copper has over 4x better at thermal conduction/heat transfer so replacing the IHS with a copper one that part I agree is only a couple of deg difference.
Click to expand...
Well, I supposed if your CPU is throttling in a situation it shouldn't be, then maybe delidding is still an option for some of the soldered chips. However, I would think that even a worst case solder shouldn't be worse than the greased CPUs. And those usually showed around 10 - 12c reduction. Not 20+. I was just watching some videos a couple of days ago on the soldered Intel chips and the general consensus was that it isn't worth it. Unless of course you have an anomaly, poorly made one.

Gamers Nexus in particular, showed only 1-2c difference between solder and liquid metal. *however, they admittedly do not have the best practices for delidding. They don't re-seal and sometimes don't even remove the old sealant.

Anyone have data on delidding the soldered chips?
 
Last edited:
azrael.arach said:
Well I can hit 5.2 but heat becomes a problem. Why I want to goto direct die cooling.
Click to expand...
I would look around and confirm the sort of temps people get, before you delid.

Unless you have a particularly poorly made CPU, it looks like delidding 9900K generally is not worth it. Your temps look good, to me.
 
Yes, just delidding is not worth it if going to put the ihs back on. It gives maybe 4 to 5 degrees at best.

Delid and direct die with waterblock is. The ihs is not going back on. At volts needed for 5.2ghz temps hit 85 to 87. Little hot for me. Most had drops with direct die to waterblock of about 12 to 18 degrees.

This is what is coming in.
https://rockitcool.myshopify.com/co...ucts/9th-gen-direct-to-die-frame-kit-complete
 
Last edited:
azrael.arach said:
Yes, just delidding is not worth it if going to put the ihs back on. It gives maybe 4 to 5 degrees at best.

Delid and direct die with waterblock is. The ihs is not going back on. At volts needed for 5.2ghz temps hit 85 to 87. Little hot for me. Most had drops with direct die to waterblock of about 12 to 18 degrees.

This is what is coming in.
https://rockitcool.myshopify.com/co...ucts/9th-gen-direct-to-die-frame-kit-complete
Click to expand...
Is the 87c in games or your typical dialy use? or Prime95?

Because Prime95 doesn't really matter. Its fun, but it doesn't matter for your daily use. Prime95 can be 20c+ more than gaming temps. And even somethhing like Handbrake, which is a realistic, high workload: Prime 95 can still be 10c+ more than that. Being in the 70's during Realbench or Handbrake, are good temps for peak overclocks.
 
That's prime temps. Handbrake is alittle lower. Once stuff gets here will post screenshot before and after.
 
chameleoneel said:
Well, I supposed if your CPU is throttling in a situation it shouldn't be, then maybe delidding is still an option for some of the soldered chips. However, I would think that even a worst case solder shouldn't be worse than the greased CPUs. And those usually showed around 10 - 12c reduction. Not 20+. I was just watching some videos a couple of days ago on the soldered Intel chips and the general consensus was that it isn't worth it. Unless of course you have an anomaly, poorly made one.

Gamers Nexus in particular, showed only 1-2c difference between solder and liquid metal. *however, they admittedly do not have the best practices for delidding. They don't re-seal and sometimes don't even remove the old sealant.

Anyone have data on delidding the soldered chips?
Click to expand...

It's completely down to luck. I have seen samples with so much gasket material they would throttle at stock resulting in 30ish degree drops, and I've seen some that only drop about 5c.

They aren't soldered exactly, they are using sTIM which is TIM and solder mixed and is really stupid imo. It creates a large layer between the die and the IHS.

Also the copper IHSs from Rockitcool have a larger surface area which allows either better transfer of heat to larger coolers and/or easier and more consistent TIM applications. You can put a nice long grain or rice size line down the middle and not worry about it running off the edges or not making good enough contact.
 
Cecil said:
Also the copper IHSs from Rockitcool have a larger surface area which allows either better transfer of heat to larger coolers and/or easier and more consistent TIM applications. You can put a nice long grain or rice size line down the middle and not worry about it running off the edges or not making good enough contact.
Click to expand...
Or you can spread the grease over the whole surface and not have to guess with a pea dollop or a line.
 
chameleoneel said:
Or you can spread the grease over the whole surface and not have to guess with a pea dollop or a line.
Click to expand...

No, that isn't any better. If you really want to not have to guess how well you applied your TIM, you have to immediately assume that both your IHS and your cooler's base are imperfect, and manually lap them to as close to perfection as you can manage. No matter how much or how little TIM you use and how you apply it, if the IHS/coldplate geometries don't line up properly, you're wasting cooling potential and potentially leaving CPU performance on the table.
 
Has anyone done direct die cooling here for a 9900k or 9900kf? Like I said I am not putting the ihs back on. My ekwb block is pretty flat for all test with dial indicators. Looking around most seem only to put the liquid metal on the cpu die, none on the waterblock. Is this the best way or should I try to figure out centers and out liquid metal on both?
 
chameleoneel said:
Is the 87c in games or your typical dialy use? or Prime95?

Because Prime95 doesn't really matter. Its fun, but it doesn't matter for your daily use. Prime95 can be 20c+ more than gaming temps. And even somethhing like Handbrake, which is a realistic, high workload: Prime 95 can still be 10c+ more than that. Being in the 70's during Realbench or Handbrake, are good temps for peak overclocks.
Click to expand...

If you system cant do the same tasks overclocked as non overclocked. Your oc is not stable
trying to use less harder task to keep a blind eye to the instability is not making it go away

the reason we test with highst possibe ltaks is to ensure that changes or new optimoans or diffrent software down the road does not suddenly take the cpu above the critical line that you accepted before because you load was below it.


But in the end its your system. if you want a unstable system you can do as you want.
 
SvenBent said:
If you system cant do the same tasks overclocked as non overclocked. Your oc is not stable
trying to use less harder task to keep a blind eye to the instability is not making it go away

the reason we test with highst possibe ltaks is to ensure that changes or new optimoans or diffrent software down the road does not suddenly take the cpu above the critical line that you accepted before because you load was below it.


But in the end its your system. if you want a unstable system you can do as you want.
Click to expand...
stability in Prime and Temps in Prime are not the same thing.
 
doubletake said:
No, that isn't any better. If you really want to not have to guess how well you applied your TIM, you have to immediately assume that both your IHS and your cooler's base are imperfect, and manually lap them to as close to perfection as you can manage. No matter how much or how little TIM you use and how you apply it, if the IHS/coldplate geometries don't line up properly, you're wasting cooling potential and potentially leaving CPU performance on the table.
Click to expand...
I was talking about application method. Flatness doesn't correct poor application methods. Spreading or an "X" are the best methods. I prefer spreading, so that there is no guessing.
 
chameleoneel said:
stability in Prime and Temps in Prime are not the same thing.
Click to expand...

Agreed i dont really care for thermals directly in prime unless if you hit themal tresshold with prime you are no longer testing your CPU as it worst? ( this was not the situation debated. this is just a side info)
 
chameleoneel said:
I was talking about application method. Flatness doesn't correct poor application methods. Spreading or an "X" are the best methods. I prefer spreading, so that there is no guessing.
Click to expand...
Sorry but that has been proven incorrect many times over the years. Spreading is inconsistent and can trap air bubbles, and an X can as well. A dot or line doesn't create a spot where two areas of TIM come together to create an air pocket.
 
azrael.arach said:
I got the one in the kit from Rockit. Honestly I had used his delidding tools before and always were nice. It was a whole kit one shot and little lazy factor in there. One order done.
Click to expand...

He discusses what kind of temperature drops you can expect from direct die cooling in the video (around 6:44)--that's what I was pointing out with the video.
 
Cecil said:
Sorry but that has been proven incorrect many times over the years. Spreading is inconsistent and can trap air bubbles, and an X can as well. A dot or line doesn't create a spot where two areas of TIM come together to create an air pocket.
Click to expand...
Do people still think air can get trapped in soft/wet thermal grease, which is under pressure and heated?

Spreading is one of the only ways to ensure that all of the contact points are covered. Especially for direct contact heat pipe designs.

Der8auer has a video where he tries various application methods with a glass plate. Spreading and X were the only two which reliably covered the entire contact area.

Some brands now have applicator tips designed for you to apply grease to the entire surface. Similar to spreading. I've seen it with Coolermaster, Thermal Grizzly, and one or two others. and many include a little spreading tool. Such as Gelid GC Extreme.


 
Last edited:
Those 2 videos are best I have seen for spreading thermal paste.

Still wondering if I should put liquid metal on waterblock and die or just the die.
 
azrael.arach said:
Those 2 videos are best I have seen for spreading thermal paste.

Still wondering if I should put liquid metal on waterblock and die or just the die.
Click to expand...
I don't think it makes a difference if you put it on both or just the die. However, if you do both, be careful and don't put on too much!
 
azrael.arach said:
Still wondering if I should put liquid metal on waterblock and die or just the die.
Click to expand...

Liquid metal doesn't like to "stick" very well unless spread. Use the qtip thing that comes with it and really tiny dots on the die and the block, and spread it around. On the block try to do an equal size rectangle just barely bigger than the die.

It's going to stain the copper, but won't damage it. The flitz polish can be used to clean most of the dark staining off later if you want to.


As for spreading TIM, I've had a ton of tools come with different TIMs over the years. The tests I've seen with slides over the years show air pockets and when removing them they don't appear as equal as a line or dot would.

But the difference between any method is usually 1-2C at most, and the only real concern is just that is at least a decent quality TIM. Doesn't have to be super expensive or fancy, but not junk either. I usually just use MX-2/4 or EK's stuff.
 
Well had decent drop in temps, next test is more voltage and higher clock and see.

Before
temps.JPG

temp2.JPG

temps.JPG
After


temp2.JPG
temp2.JPG
 
Currently stress testing at 5.2 no avx offset at 1.32 volts in bios. 10 minutes in and temps didn't even go up past 5.1 test.
 
Had to up voltage to 1.34 in bios. Still at 5.2 no avx offset. Average temp under real bench is still 66 to 69. One stable in realbench will move to prime95.
 
I gave up on using Prime95 for avx testing. It's pretty inconsistent. I just use OCCT linpack now and it gets to peak temps faster and does it every time.
 
Cecil said:
I gave up on using Prime95 for avx testing. It's pretty inconsistent. I just use OCCT linpack now and it gets to peak temps faster and does it every time.
Click to expand...
The thing which I find useful about Prime95 is each pass gets more difficult.

I have found several times that an overclock may be stable for a few small FFT passes, but will eventually fail a pass. Usually, if I can run Prime95 to pass 21, then my overclock is probably rock solid, as far as Prime95 is concerned. But that's why some people do over night or even 24hr Prime95 testing.

As far as temps go, small FFT hits max temps pretty dang quick ;)
 
Last edited:
chameleoneel said:
The thing which I find useful about Prime95 is each pass gets more difficult.

I have found several times that an overclock may be stable for a few small FFT passes, but will eventually fail a pass. Usually, if I can run Prime95 to pass 21, then my overclock is probably rock solid, as far as Prime95 is concerned. But that's why some people do over night or even 24hr Prime95 testing.

As far as temps go, small FFT hits max temps pretty dang quick ;)
Click to expand...
That's why I like OCCT now. It goes straight to max stress in less than a minute. It'll usually find any errors in less than ten. I used to run Prime95avx for an hour or more and sometimes it wouldn't actually trigger avx and give load temps in the 70s when OCCT would hit 85-90 right away.
 
You must log in or register to reply here.
Back
Top