Delidding 1366 CPUs [Soldered]

[Dinnercore]

Hey there. I´m currently busy with 1366 CPUs and made some tests on delidding them even tho they are already soldered.

Right from the start I can tell you it is definitly NOT worth it for temperatures, I can not recommend doing it. There are other reasons to do it tho (I think some Mac Pro needs delidded chips).

The first problem I encountered was, that the current delidding tools like the one from Der8auer are too small to fit the 1366 CPUs. So I instead used the vice method. Have one edge pressing on the IHS and one on the substrate. Keep the angle as flat as possible. Gently but steadily increase pressure, avoid sudden movement and the IHS should just come off (remember that the CPU will drop, try to catch it with something soft underneath).

No heating required! I did this 3x and all three CPUs came out alive and well.
EDIT: 4/4 now, I´ll keep track of the number

Second IMPORTANT thing to know is the placement of SMDs underneath the IHS. I found that with Xeons and the i7s SMDs are placed in the same shape around the die. You do not want to push the IHS towards the SMD components, ALWAYS away from them.



As you can see, there is one side that has no caps on it to the right. So you want to push the IHS to the left if looking at it from the top, oriented with the little QR-Code towards the bottom. On the vice, if you use the top edges you will have to flip the CPU upside down, so you would want the left edge of your vice sitting against the IHS and the right one against the substrate.
Always have the edges of the vice fully cover the side of the subtrate to avoid splitting the pcb.



Once you got this done you now need to remove the solder, it is very soft so you can carefully scrape it off with a sharp knife. I used a scalpel which can actually cut the indium solder but does not scratch the die IF you are careful and use little force. This might take some time, so be patient and go slow. You do not want to damage the SMDs, scratch the die or cut into the pcb.

This will still leave some traces of the solder left on there which can now be polished off, some use a pencil-eraser (did not work for me) or polishing paste. I used a very fine micro-mesh sanding paper used for polishing metal surfaces.

After that you are nearly done, just carefully remove the black rubber from the IHS and subtrate, again be careful not to cut the PCB by accident. And that´s it. Now you can go direct die cooling or use the IHS + LM.

I took it one step further and sanded down my die to remove the diffusion barrier (WIP-pic):


I want to run a small experiment and see if that will have a major impact on the CPUs lifespan. I´m now running this i7 960 direct die and will see if it dies within a month or two. I found very little data on this topic and thought it would be interesting to test this out.

Temp improvement, if any, is marginal on those 1366 i7s. It is very hard to get an even surface under full cooler mounting pressure and if you take the pressure off, the weight of the tubes from the waterblock are enough force to skew the cooler in one direction and completly ruin your temps.
If you finally find a proper position, the idle temps are identical and the load temp improvement I saw was around 3-6°C.



What´s your take on delidding soldered CPUs? Would you consider it or have you done it before? Feel free to share your results and thoughts.
Did anyone else ever grind down a die and ran it without diffusion barrier for an extended period?

 

[sneekypeet]

Did anyone else ever grind down a die and ran it without diffusion barrier for an extended period?

I do recall a long time ago, someone here ground down an Opteron, but I cannot for the life of me remember his name.

 

[Deleted member 185158]

Hello!! Very very very nice job!!! Super impressed for the delid without heat.

Some answers to questions at the bottom of the post......
I've been de-lidding Soldered AMD processors for about a decade now. I've never gotten a chance to get a fancy 8auer tool to even try using!!! Those tools are so sweet. But none for AMD hardware....

Have lapped a core pretty far down. But there are other components is the way from actually removing all the silicon over the insulator, but you can start seeing through it better with a bright light!!
The avatar to your left is actually THE example I could display of lapping the silicon quite a bit after a de-lid. The processor depicted is an Athlon x2 5000 (no plus) ending in DGI (if memory serves) and unlocks to an FX-5000 Quad core.
See sig for most current de-lid.

Any-hootz, Fantastic job. Keep up the good work!!

Can I also add some techniques that may help and apply to your experience with cooler mounting???
So tip one, Lap the cooler the same grit as the core. Do them both on a piece of machined flat glass, this will provide the best most even surfaces. I recommend min 2000grit ideal 3000 grit.
Two, use a 1/4" drive torque wrench for your mounting pressure. Tighten by hand as you usually do, but make sure all are spot on even with the torque wrench.
Three, practice the Thermal Interface Material or LM amount and spreading. If it pushes out, use less. (never used LM personally but would apply here)
_The reason for no push out, is because your mating surfaces are so even and flat, you will use far far less TIM. Surprisingly less....
Forth, Boot bios. Monitor cpu temp. Keep tightening the screws or nuts until you see the temp stop dropping. Use a strong backing plate.
_Then go into windows and load test. Re-adjust/test mounting @ load. Re-adjust mounting every 12-24 hours for the next three days.

 

[potato580+]

hmm never tried deliding, it cost around $50 for a task depending the difficulty level, and have a risky 70:30 chance of broken, might try my old cpu, but its not worth a$50 anyway

 

[Dinnercore]

So tip one, Lap the cooler the same grit as the core. Do them both on a piece of machined flat glass, this will provide the best most even surfaces. I recommend min 2000grit ideal 3000 grit.
Two, use a 1/4" drive torque wrench for your mounting pressure. Tighten by hand as you usually do, but make sure all are spot on even with the torque wrench.
Three, practice the Thermal Interface Material or LM amount and spreading. If it pushes out, use less. (never used LM personally but would apply here)
_The reason for no push out, is because your mating surfaces are so even and flat, you will use far far less TIM. Surprisingly less....
Forth, Boot bios. Monitor cpu temp. Keep tightening the screws or nuts until you see the temp stop dropping. Use a strong backing plate.
_Then go into windows and load test. Re-adjust/test mounting @ load. Re-adjust mounting every 12-24 hours for the next three days.

This is some very good advice! Yes I did lap both cooler and die and I did evenly tighten the cooler but its on a flat bench and the tubes sag a bit, creating a pulling force on the block. It helps to have the tubes supported mid air above the block.
I did use 2500 grit to start (all on perfectly flat glas), then when I saw the coating completly gone I went up like this 2500 -> 3000 -> 6000 -> 8000 -> 12000

Another thing for sanding those surfaces is, use very thin or rigid sanding paper. Some stuff has a lot of flex and give to it, so when you push it down even just gently you create a bump and grind more from the edges then the middle.

For applying the paste, I use a very small amount and spread it with a razor-blade as thin as possible. You do want manual spread so it reaches the corners without using so much in the middle that it squeezes out like you said.

Also, when in doubt just re-do it until it seems right.

 

[Deleted member 185158]

Right on, seems you have good practices.
However you are going way too high in grit in my opinion. You want some micro scoring for surface area.
You can also use those foam dots from socket A cpus to help stabilize your water block when youre mounting up for the first time.

I use 3 inch long screws with nuts. I run the nuts and block to the top, screw into the back plate and this allows for a straight down contact with little twisting of the water block.

If youre hoses are moving the block, the mounting is way way too loose. Should be tight enough you can move the entire board weight.

 

[Dinnercore]

[...]If youre hoses are moving the block, the mounting is way way too loose. Should be tight enough you can move the entire board weight.

I think you are right, I should really tighten it further. I´m always a bit scared to go full hand-tight because I think I might damage the socket. Those pins are so delicate but I guess since you push on all of them you can use a bit of force.

And I just tried something else, I added some 0,5mm thermal pads around the die which just happend to be the right height to be level with the die. These pads help to spread the force on the subtrate, so the waterblock is not just pushing in the center on the die.
Now I see a more consistent 6°C improvement on my load temps.

 

[Enterprise24]

My 9900K P0 saw around 6-7C dropping in temp after delidded.
My friend's 9900KF R0 saw only 2-3C dropping.
R0 is a nice improvement over P0. Probably not worth it to delid 9th gen with R0 stepping anymore.

 

[dont whant to set it"']

I went by the gravity asist method by placing the CPU ihs facing down with the pcb' resting on stacks of coins and inserting a soldering iron under the ihs thus letting heat and gravity do the rest, the soldering iron was not touching the ihs , so there was an air gap of 1/4 inch for a more uniform heat soaking the ihs. LE: prior to the above procedure I did tear the glue wich retains the ihs (outer perimeter "lip") to the pcb' with a razor blade.

 

[Tomgang]

That i7 980x I have now already have good temperatures, so no I have no intensions on delid my cpu. I will never delid any cpu that is already soldered. Not worth the risk or temp gain. If any thing I Will only attempt to maybe lapping the cpu heatspreader.

With thermal grizzly kryonaut, noctua nh-d14 and 3 x noctua industrial ippc fans. This cpu runs far from hot. Stock temperatures at full load stays well below 50 degrees Celsius and more like low 40 degrees on hottest core. At 4.4 ghz with 1.43 volts hottest core max out at 75-76 degrees Celsius. I even managed to push cpu to 4.75 ghz on still same air cooling at 1.55 volt a cold winter day with out cpu thermal throttle all throw it dit come above 90 degrees Celsius at those clock and voltage.

But point is that these cpu can overclock plenty high with decent cooling with out delid needed. And Intel recommends max 1.4 volts for these CPU's any way. So no point in delid if you ask me. But other wise good job on your delid.

My cpu overclock results on air cooling. No delid or lapping. This is stray out of box cpu results.

http://imgur.com/a/uHjbbMg

 

[Deleted member 185158]

I think you are right, I should really tighten it further. I´m always a bit scared to go full hand-tight because I think I might damage the socket. Those pins are so delicate but I guess since you push on all of them you can use a bit of force.

And I just tried something else, I added some 0,5mm thermal pads around the die which just happend to be the right height to be level with the die. These pads help to spread the force on the subtrate, so the waterblock is not just pushing in the center on the die.
Now I see a more consistent 6°C improvement on my load temps.

When I first started de-lids, I had the exact same issues.
First chip was an amd 9850BE Phenom Agena core.
Mounting the water block was nerve wrecking almost as bad as thr delid itself.... I used heat for removal, still do.... But yea, getting water block placement and tightness was a challenge.
The block seemed to turn from the pressure from the hoses, I started mounting the block first, then connecting hoses.
Once I realized the mount was too loose, started to crank down with some good pressure and watched the temps drop even more.

One particular scare was a no post situation I had caused where the board flexed so much the pins didnt make contact in the socket, but this amount of force caused me to buy long screws and double the backing plate. I just ran the screws through both plates and seemed to help a great deal preventing the board from bending. Something to keep an eye on.

But yes, the mounting pressure will be greater than expected. Gotta have even pressure too, its just as important.

Hardware isnt cheap. Ive killed 2 chips maybe 3... Out of maybe 20 total delids.....something like that.....Cant remember lol. But none from mounting but from mistakes in the delid process. So I think you can go ahead and easy over that fear. You'll be alright.

 

[natr0n]

I only delid if not soldered and/or if temp are absurd as well.

 

[Dinnercore]

I only delid if not soldered and/or if temp are absurd as well.

That is good practice, as I said in my first post I do not recommend doing it to anyone else. It´s just that I love to play with hardware and get the absolut limit out of it. And I love to test out things for myself.

I´m wondering when I will break my first chip from a delid. By now I did this to an i5 2400, the i7 960 and 2x Xeon X5690s. All are alive, but I won´t try it on something as expensive as an i9 9900 any time soon.
Something else I just found out by accident, heat can help you spread the paste as it makes it a bit more fluid. I had the pump not running during the first boot and saw core temps in the 90°C range... After switching the pump on my temps have now improved by another 1°C.

I guess direct die cooling is something you need to get experience with. First attempts will be a bit underwhelming and nerve-wracking but when you get the hang of it you start to see more improvement.

Hardware isnt cheap. Ive killed 2 chips maybe 3... Out of maybe 20 total delids.....something like that.....Cant remember lol. But none from mounting but from mistakes in the delid process. So I think you can go ahead and easy over that fear. You'll be alright.

That´s why I don´t do this on my current daily system, I do not even want to imagine what it would be like to try and delid a threadripper and kill it in the process.

 

[DR4G00N]

You should just leave soldered chips alone, if you want to lower the temps then just improve the cooling.
I've delidded many soldered chips, but it never really benefits any of them unless the solder application was defective from the factory.

Much easier to just flick on the waterchiller to simply reduce temps by 30c or more.

 

[Deleted member 185158]

That is good practice, as I said in my first post I do not recommend doing it to anyone else. It´s just that I love to play with hardware and get the absolut limit out of it. And I love to test out things for myself.

I´m wondering when I will break my first chip from a delid. By now I did this to an i5 2400, the i7 960 and 2x Xeon X5690s. All are alive, but I won´t try it on something as expensive as an i9 9900 any time soon.
Something else I just found out by accident, heat can help you spread the paste as it makes it a bit more fluid. I had the pump not running during the first boot and saw core temps in the 90°C range... After switching the pump on my temps have now improved by another 1°C.

I guess direct die cooling is something you need to get experience with. First attempts will be a bit underwhelming and nerve-wracking but when you get the hang of it you start to see more improvement.




That´s why I don´t do this on my current daily system, I do not even want to imagine what it would be like to try and delid a threadripper and kill it in the process.

I killed a 9590. Cut the tracings in the pcb. Killed an fx quad 4150, too much heat the core exploded.

Many chips run on daily Phenom Phenom II and FX around 10c drop in temps on ambient liquid, not chilling like Dragoon mentions. I use chilling now and main rig is stock air in the sig.

Older processors seem to have the most benefit from delids.

Newer tech like my 2700x, doesnt help much. A few c drop in temps, not really worth it to anyone except guys like us, enthusiasts.

Ill continue to delid even through the shun process "its not worth it.... Blah blah..." Ya ya...
Thats not why I delid. Do it because... I felt like it. Nothing more. And its fun to show off too....

 

[Dinnercore]

You should just leave soldered chips alone, if you want to lower the temps then just improve the cooling.
I've delidded many soldered chips, but it never really benefits any of them unless the solder application was defective from the factory.

Much easier to just flick on the waterchiller to simply reduce temps by 30c or more.

It is much easier to just improve the cooling, yeah but I want to go for something sustainable and for me anything sub-ambient is nothing 'sustainable'. Using a big chiller or dice / LN2 is something for bench-sessions but most would not consider running a setup like that as a daily.
I like to see how far I can go on ambient water with just a big radiator as heat exchange. My setup is something I consider to use and with my next build I will use something similar as a daily system.

I would love to use a chiller, but these things are noisy and need to be in a seperate room. Not to mention the 1000€ price tag on the unit alone + cutting holes in my walls. A waterchiller would not be a 'simple' solution for me.

And for the i7 it seems to have improved vs soldered IHS.



A quick and dirty cinebench run, my room temp is at 21°C, water temp is 23,2°C.
4GHz @ 1.28V. The hottest core hit 61°C max. The temp difference between cores tells me I might need to adjust torque again after it settled over the night.

Even so, this is a solid improvement. On the stock chip I had (with the same ambient and water temp) 61°C already with 1.22V (@ 3.7GHz). I did not have the stock chip on 4GHz which is a shame now that I can´t compare them side by side.
But, I did a stock run with no OC for both. The delidded run had 30°C idle and 50°C load, with the stock solder it had 32°C idle and 55°C load.

 

[Deleted member 185158]

Those temps are good.

I use geothermal or tap to drain these days. Very reliable stable and sustainable temps.

I dont ambient liquid cool any more. Current modern tech doesnt seem to benefit at all especially for AMD ryzen processors.

When using ambient liquid I did testing with folding at home where I could sustain a load from 12. To 48 hours to really get a good average load temp.

The 2700x is on air cooling. My clocks never changed from the delid. It did help eliminate leakage getting cold temps loaded cb at 13c and idle -30c TEC was able to run 4ghz at less than 1.2v.

Again nice job with the delidding. I wish Id done some soldered Intel for comparison with you, but unfortunately Ive not. All of the blue team Ive done where not soldered for customers looking to lower temps....

 

[Dinnercore]

@ShrimpBrime Thank you

Now to extend my results, here is the same setup on 1,45V and 4.28GHz:


Check out that power draw, 341W peak on just a CPU test. Judging by the idle from the 4GHz run (152W) that is 190W for just the CPU. Now add the idle power, as I disabled all intel speedstep and C-states I´d guess the CPU at around 230W. Still 'only' 72°C on the core with an ambient cooling solution.
If I wouldn´t run a huge radiator and direct die, I think I´d need a chiller by now. At some point I need to pair one of these Bloomfield CPUs with a GTX480 SLI-Setup

To bad my i7 960 is not a great example of overclocking potential. It struggles to go any higher. Since it came as a free extra with an OC-oriented board there is a good chance the previous owner ran it overclocked for the past 10 years and it´s showing minor degradation.

I dont ambient liquid cool any more. Current modern tech doesnt seem to benefit at all especially for AMD ryzen processors.

I can see that, modern hardware does not gain much OC-headroom from ambient cooling methods. And air-coolers have gotten really good, compared to the days when 80mm fans were standard. I just can´t help it, I´m a water enthusiast.
And ryzen first gen loves lower temps too, my 1950X is very happy on 4.05GHz with 1.32V as long as it stays below a certain temp. I can run my render workloads fine and it peaks at 56°C but as soon as I turn off the pump, it crashes around 68°C-70°C.

 

[Deleted member 185158]

That 960 is a decent clocker in my opinion. Really any of those 900 series i7 in the 4ghz range is decent. You are probably around the need for chilling plus extra volts to go more.
I feel the results are pretty good. My Buddy Sparkey247 running an SS Unit had a i7 960 at 4750Mhz at 182 reference clocks. Looking at the difference of roughly 400Mhz and a very big difference in temps. Your just at a cooling limitation is all.... But really good for Ambient liquid for sure!!

That's funny you mention your Ryzen crashes around the Cpu High temp alert of 70c. Seems accurate to me. I try to keep my 2700x below this temp when possible. You could de-lid it, but it won't help this issue at all unless you can get a significant drop in temps, which you really won't. However, it may also crash because it starts leaking hard at 70c and the v-core is just to low at that frequency.
I found that if I run too low of v-core, the reference clocks start drooping or throttling if you will. Ryzen seems pretty sensitive to low voltage and high clocks. You'd have some tweaking and testing to prove this theory wrong though....

 

[snorlaxgangs]

Nice. It's great i found this thread.
I used the vice to delid all of my i7 1366 cpus. I delided my 990x awhile ago and the temp is pretty good. Since i used liquid metal, i sand the IHS ~0.02mm shorter and lap them all with japanese 10000 sand paper. I don't have the tool to measure the die/ihs like debauer so i probably i didn't do it justice. The temp from each core is unevenly. I did try to apply more or less LM, it's still the same. Also i found that using adhesive will cause some uneven contact between the die and the IHS without using relid tool so i just dropped the IHS on top. I removed the bracket too, no reason at all, i just do it. All in all, the temperature is pretty good for me, i used custom water cooling.
And i made some art while sanding the IHS

 

[Dinnercore]

Nice. It's great i found this thread.
I used the vice to delid all of my i7 1366 cpus. I delided my 990x awhile ago and the temp is pretty good. Since i used liquid metal, i sand the IHS ~0.02mm shorter and lap them all with japanese 10000 sand paper. I don't have the tool to measure the die/ihs like debauer so i probably i didn't do it justice. The temp from each core is unevenly. I did try to apply more or less LM, it's still the same. Also i found that using adhesive will cause some uneven contact between the die and the IHS without using relid tool so i just dropped the IHS on top. I removed the bracket too, no reason at all, i just do it. All in all, the temperature is pretty good for me, i used custom water cooling.
And i made some art while sanding the IHS

View attachment 135345View attachment 135346View attachment 135347View attachment 135348View attachment 135349View attachment 135350View attachment 135351View attachment 135352View attachment 135355

Really nice results, thank you for sharing! And the lapping on that IHS is so flawless, it´s a piece of art. I know many don´t care to lap a perfect mirror because it does not gain much if any, but it looks so nice.
Did you lap the die itself too? If so, did you remove the diffusion-barrier? For how long have you been using this CPU now, any degradation yet?

I want to try LM on an upcoming build, but I´m not sure how well it would work when going direct die with the cooler.

Anyway, I have my next candidates ready:


An X5650 + X5660. This time I will run them with a proper before and after test at a fixed ambient temperature.

The I7 960 will have to pause for a few days, so far it seems to feel comfortable with 1.42V and 4.2GHz, I use it daily.

My Buddy Sparkey247 running an SS Unit had a i7 960 at 4750Mhz at 182 reference clocks.

Yeah I see his scores on HWbot. I can not quite get to his results due to cooling limitation BUT I managed to come close to his CB15 score of 710 (Him using SS).
My score:


Just 8 points behind, I managed to squeeze my RAM a bit further. I´m very happy with this result for ambient cooling at comfy 21°C room-temp.

 

[Deleted member 185158]

Haha nice! Get that ambient colder.... you know the drill! Sparkey built the SS, I watched him build a few SS units actually. A true benching monster!

Yes mirror finishes.

Looking forward to more of your de-lids!!

 

[snorlaxgangs]

@Dinnercore

Did you lap the die itself too? If so, did you remove the diffusion-barrier?

I only lapped it a bit with 5000-7000 and finished w mirror effect with wet 10000. However i did try direct cooling but the temperature wasn't what i expected. IIRC it was 5*C higher than the current one, And weird enough the minimum core temp was still 10*C lower than the highest. I only have the razor blade to check how flat & even it is. I also have tested it with a Koolance and another EK ( all full nickel) to make sure the result is certain enough. I think because the die is a lot smaller than the block so it doesn't transfer the heat effectively. Maybe a custom direct cooling block or custom IHS would give better result. I have around 300g pure gold and 400gr of pure silver, lol i can make a custom IHS i think but i need to do some research to find anyone does it before.

For how long have you been using this CPU now, any degradation yet?

It's almost 5 months since i have 990x. In the beginning after all the "cooling optimization", it wasn't stable at 4.6 HT on with 1.425V. But now with same Uncore/QPI/Ram setting, it is blender stable 24/7 @ 4.6 HT on, same Vcore, 1.462~65 under max load. With HT off, i can push it to 4.8GHz w 1.45Vcore, but with LLC on max load was 1.5V so i don't know if it will get any degrading. All in all, i'm running stable at 4.6 HT off, 1.3750V.

@ShrimpBrime IDK if it's just me or AMD chip always look a lot tougher than Intel. Btw i saw you using 980Ti, i just bought msi 980ti RE with EK Titan X block for $200, do u mind if i ask what's the safe Voltage on this card? Also when i added 260MHz but heaven benchmark boost it to like 1721MHz or something, keep getting crashing. but i render blender for a couple hours and it seems stable at 1465MHz w stock VCore.

 

[Deleted member 185158]

@ snorlaxgangs
I have a standard GTX 980.... But wont be much different, you have more vram is all really.
So that said, mine is voltage locked at 1.212 and would require a hard mod to exceed that.
So as long as you can cool the card, around 1.212v is ok, but not entirely sure what your card tops out at, but you wont be able to change it higher.

Dont beat the card with blender. Adjust your oc to each benchmark you run. Vram overclock will carry your score a little further.

 

[Dinnercore]

I have done some more delidding. This time threw some C2D E8400 in, because I want to try them out too.



There may be my first dead CPU, as on one of the E8400s I got careless and did what I told everyone NOT to do. I did not check if the substrate was fully covered by the vice and pushed only on the top of it, the CPU broke free and ripped the substrate on one side.
It does look like mostly chipping on the surface but one corner shows a split pcb ~1,5mm deep. Very likely that this CPU is dead, but I´ll try it. Filled the broken part up with CA-glue to prevent further damage.

For now tho I have focused on the X5650. Here are my before temps with the stock soldered IHS:

All tests were done in an 18°C ambient environment. The water-temp as measured in-line from the loop can be found at each result. I heated up the loop before each test by running the CPU-Z multicore stress test for 10 minutes. It may not be fully saturated, but it is a reproducable number and saves me from waiting an hour each time for the giant radiator to soak.
To finish it off I ran 3x CB15 back to back right after the heatsoak. During the whole time I kept HWMonitor open so you can see voltages and temps.

Baseline at fixed speed and voltage. 3GHz @ 1.25V in Bios and 50% LLC. Water reached 21°C.
Idle: ~28°C / Load 44°C


3,5GHz @ 1.25V in Bios and 50% LLC. Water reached 21.2°C.
Idle: ~32°C / Load 49°C


4GHz @ 1.3V in Bios and 50% LLC. Water reached 21.6°C.
Idle: ~36°C / Load 54°C


4,41GHz @ 1.381V in Bios and 50% LLC. Water reached 22°C.
Idle: ~42°C / Load 63°C



After this I delidded it like before, again no heat needed. Just out in the cold of a shed and a gentle exponential pressure curve.
I wanted to test paste first but encountered a weird mounting problem that I could not identify at first. Temps were shooting on average 10°C higher with the delidded chip and direct die paste. I questioned my sanding technique so I reworked the surface 3x within 4 hours. I started to grow tired and decided that maybe paste direct die just is a lot worse than paste -> IHS and tried liquid metal. This is when I suddenly saw half of my cores improve a lot and the other half spiking up to the 70°C range. My mount was uneven the whole time. I redid it atleast 10x and could not figure out what it was.

So I laid the board flat on the table and immediatly saw I mixed two sets of screws from different waterblocks.... One was a bit longer, the others a bit shorter. I felt so stupid but atleast I found the issue.
Looking a lot better now.

Only problem now is, I stained the block with liquid metal so I felt like it is no longer fair to use it for thermal paste results which is why I went straight to direct die + LM:
The ambient dropped just a tiny bit as you can see in the lower water temps on average.

3GHz @ 1.25V in Bios and 50% LLC. Water reached 20.8°C.
Idle: ~28°C / Load 38°C


3,5GHz @ 1.25V in Bios and 50% LLC. Water reached 21°C.
Idle: ~31°C / Load 41°C


4GHz @ 1.3V in Bios and 50% LLC. Water reached 21.5°C.
Idle: ~34°C / Load 46°C


4,41GHz @ 1.381V in Bios and 50% LLC. Water reached 21.8°C.
Idle: ~37°C / Load 52°C


The higher delta T got, the more improvement can be seen (no big surprise eh), especially on the idle temperatures this becomes visible. During the 3GHz run idle stayed nearly identical, but on the 4,41GHz setting I got a 5°C improvement just with idle temps. And 11°C for the hottest core during load, I like that. I think I´ll use liquid metal more often now, this was the first time I used it.

Why did I choose that odd 4,41GHz number? Because this was the point I would consider a max. stable and save 24/7 OC. At load voltages of just 1.349V we are right at the recommended 24/7 limit. I personally would use it like this as daily.

Now I have to take a few days before I can look into the E8400 and test it, there is something big coming up for me that will grab my whole attention. The waterblock for my SR-2 is done and will arrive soon, I will not sleep much this week and focus only on that build to finally get it done.


EDIT: Oh and just to add, the 1.349V I mentioned during load are what I measured on the board with my DMM. As you can see the software readouts are a bit different.