Polishing CPU Die

[im4gine3]

Gonna look into the temp issue again some time this week.. as my CPU delid is averaging 55c idle and ramps up to 83 on Cinebench.

This is definitely not normal under full water so perhaps I've not put enough LM on.. most videos seem to show a small pea sized blob, but I've only been applying grain of salt amount, so i'm guessing it's not 'wet' enough.

Failing that, i may polish the CPU as there seems to be a bit of residue left over which won't go away by applying LM.. but wonder in terms of polishing.. would you polish like how you would lap a cpu? - glass surface, cpu die flat and lap in figure 8 motion, or do you think it would be ok using finger pressure with a polish cloth. e.g. cape cod polish cloths (i have a few)

Specs in signature... running 24H2

 

[im4gine3]

Just checking but you have been manually spreading the LM with a cotton bud right ? Both surfaces need to be covered.

With regards to polish you get Flitz polish or similar and rub the die. You don’t lap it like you described above.

Edit: also a pea sized blob is too much. You need as thin as possible on both surfaces.

Yeah, both sides have been covered with LM with as thinly layer as possible.. to the point it 'looks dry' or not enough.

Last i delid was on a 4770K a while back, but memory is hazy with how much is needed, so i checked a few videos, and one from Derbauer looked like more than a few grains of salt worth:

The first time i LM the die, was using maybe 1 grain of salt worth for both the die and ccd, then another grain of salt on the inner heatsink. Idle temps in the high 50s, load temps mid 80s.

Second LM attempt, i put a little bit more.. 2 grains of salt on both areas, temps dropped to mid 40s, load high 70s to low 80s.

Articles suggest a slightly smaller than pea sized (quarter maybe or probs 5 grains of salt worth) to the point it looks wet.. but not as far as a puddle.. or not where when you spread there's a ripple effect.


With regards to the polish, if lapping is not the correct method, surely using the finger is not fine too.. since the result is not meant for a mirror finish, but to remove any excess residue and make the die as flat as possible. would finger polish make the die surface uneven? I was thinking of maybe wrapping the cloth around a spludger - the spreader usually sent with thermal pastes) and giving it a good rub

Edit: Grain of salt maybe not the best term to describe it, as a grain is a generous amount.. maybe a spec of dandruff worth

 

[im4gine3]

I'll get a picture tonight after work. (with current LM on it, and what it looks like without)

Yeah i'm using the TG Direct Die block... originally i thought the block was a defect as the inlet port is at the bottom (with micro fins) but after a back and forth discussion with Roman, seems the website/instructions/pictures posted elsewhere.. including OCUK website is incorrect and need fixing... as they show the inlet port at the top and outlet at the bottom.

I thought polishing was not only a mirror finish, but to sand down/remove any of the residue/idium/solder which is left on the die? to be honest, can't harm to try as i've seem some posts of reddit with slight improvements in temperature after polishing.

 

[im4gine3]

No you either scrape off the solder or dissolve (if that’s the correct word) with LM.

No point polishing, I think you have a contact issue. Try more LM or use thermal paste to see how much contact your DD block is making.

ok i'll try both tonight.. and share pictures

 

[im4gine3]

hmm guess i'll need to check contact using thermal paste.

These are my idle temps before i did anything:

Temps after a CB 10min run: (see max temp)

Temps after CB 10min run after applying more LM:

temperatures were worse by about 2 degrees after loading default Bios and enabling EXPO. Even tried dropping from -30 to -25.. made no difference.

Spoiler: DD Block and Die BEFORE

Spoiler: DD Block and Die AFTER

Spoiler: Temp Loop

 

[im4gine3]

Have you tested against performance? CPUs these days work to temperature targets and can use the extra overhead to hold turbo boosts longer, etc. in some cases it can result in higher temperatures on better cooling depending on the thresholds you are hitting under different cooling. Though 55C idle is not normal with the CPU allowed to operate under normal power states.

I used the computer for normal gaming and benchmarking using OCCT/CB.. all of which raises the temps to around 80c. GPU is fine as that never goes over 35c.. at some points i'm curious whether it's to do with the amount of QD i'm using restricting water flow, or maybe a setting i'm missing on the bios resulting in high temps

If you hsve any exposed copper after a flatening process. I'd probably consider nickel plating parts to reduce LM erosion. Home kits available.

nope.. no copper about.


Pictures to determine contact: (LM on die only, clampted to socket, removed then checked DD for LM contact)

Naked die with LM removed

Since i'm getting these temps, and well the CPU is second hand anyways, i decided to polish the CPU and try again. I didn't go crazy with it since i'm not in the mood to sit there for hours rubbing away. I don't have pics of the polished CPU, but upon reapplying LM and tightening the block to the socket to the point i think the CPU wil nto break, my temps have improved slightly idling around 43c. Will give it a day or so to burn in and see how it goes

At the end of the day, i just can't work out what is diffierent.. as Kindai said his temps never reached the 70s no matter what he did... the only thing which has been changed is different set of memory albeir same model.

 

[im4gine3]

weirdly changing the loops config slightly knocked off 2degrees but still the load temp sits in the 80s.. i have another waterblock on the way to see if the temps improve.

 

[im4gine3]

Were both block and cpu die surfaces smooth to the touch before you put LM on ?

Yes

 

[im4gine3]

Do you have the existing IHS ? Worth putting that back on with LM and using a different cooler/ non DD block.

Just to see if it makes a difference.

yes i was considering doing that as i have an Alphacool Core 1 block lying about.

I also spoke with Roman again for advice about these temps, and he kindly offered to send me a Pro block to see if temps improve. (he said there have been others who have experienced similar issues and thinks it's due to the contact between the nickel oxide on the block making temps worse).

So will try with both when the TG arrives

 

[im4gine3]

after receiving a block from Debauer, i installed it and reran CB23 for 10min.

Idle and load temps improved a lot compared to the old block, but i feel it could be improved more...

i feel the cause could either be the restriction of the loop with the many QD i'm using together with 10mm ID tubing and a low pump speed
re-lm'in the dies
and/or allowing the LM to temper with repeated heat cycles

Idle flickers between 36-41c
Load doesn't go beyong 71c

Using same settings as above but i updated to the latest bios and doesn't sit stable at -30, so had to drop it to -25

 

[im4gine3]

Sorry not read the whole thread, but........

Firstly, you dont need to cover the SMD's in that nail polish stuff from TG, its just a marketing gimmick, they are already covered in a glue like substance, so already protected, if I accidently get any LM anywhere else on the CPU except the dies, I just grab a cotton bud, dip it in some 99% IPA and clean it off just leaving the dies covered.

2nd, to polish the dies and get rid of that last bit of residue, I just gently used the grey scouring pad that comes with TG's LM, very lightly, it takes quite a while, but comes off, and once off, allows the LM it adhere to the dies much easier, its not residue from the solder, its the same residue that the LM leaves on the underside of you water block, especially if you used LM for 15 to 20 mins to eat away and remove the last of the solder once you got the lid off and thats what TG supply the grey pad for.

Finally, limit the max temp in the bios and test with something like CB23, so if you test with no limit first, make a note of your score, then limit the temps to 75oC, then test again, if the score is the same, limit the temp to 65oC again, if the score is the same, then the CPU is still boosting the same as stock but you've shaved 20oC off the temps, make sure you test with something lighter load too so you make sure your are still hitting your max boost speeds, just remember one of the 3 PBO factors are temperature, so if its running cooler and knows it has the headroom available, it will boost higher for longer, so if you can get it to hit max boost speeds at a 65oC limit, then the block is doing what you brought it for and its a win win.

This was my 7900 non X, now sold you can see the glue they put over the SMD's in this picture, you can also see the residue you speak of on the die's, I never polished this one, next time I take out my 7900X3D, i'll take a pict of it, it has more glue over the SMD's and also I polished the cores on it.

Thanks for the info.. i'll do a bit more testing this week.

In terms of the glue, i purchased the CPU from MM which was already delidded and glue applied.. but polishing the die did help the temps a bit by around 2 degrees.

 

[im4gine3]

yeah after discussions with Roman, it seems more people had issues with the original TG block in terms of idle and load temps, and he mentioned it was due to the nickel plating properties, which from his video, they sourced another supplier of, and redesigned the fin density of the new block.

After polishing the die even more (as i was bored), and retesting applying more/less LM, my CPU now idles in the late 30s/early 40s which is to be expected and never goes beyond 72c on CB23, 55c on OCCT large dataset. All fans at 500rpm

That said, while -30 CO and thermal limit of 75 was stable during benchmarking, not so much during general use/gaming, with it often crashing, so i knocked it back to -28. Too lazy to do per core.. maybe when i have some free time.

I think i've repasted it enough to the point i'm not gonna get any better.. but think it could be related to the restriction of the loop.. with the number of QD i'm using temporarily, will see what temps are like when i go back to hard line.