Announcement

Collapse
No announcement yet.

Results: 9900k Die Lapping; A Wonderful Disaster!???? (Super long read)

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Results: 9900k Die Lapping; A Wonderful Disaster!???? (Super long read)

    Sharing another story and results....

    Decided to cut down the height on the silicon for my 9900k...

    Original height was measured using a replica Mitutoyo caliper at ~ 0.87mm, the goal was to reach ~ 0.6mm. I royally messed up but was able to recover.

    Grits Used 150 to 7000
    Polish Used: Car Pro Fixer 1 step nano

    Was I concerned about potential die contamination from LM? Originally No... (any actual reports of this being an issue? I just found speculations) But now I am somewhat concerned... (thus the ? at the title) I will definitely follow up on the behavior of the CPU in a few weeks, 6 months or a year or whenever the CPU decides to act up.

    Originally started with 600 grit, but the amount shaved off the first few passes indicated it was inefficient.




    So using the most abrasive I had available, I started at 150 grit and worked my way to 3500, (Anywhere from 20-30 passes rotating 90 degrees x 4 per grit)

    During the progression, I noticed a scratch that I couldn't level out, rather than back tracking grit, I remembered something and just continued my progression to the higher grits and observed that small scratch grow. Yup there was a crack in the silicon. A hairline crack that ran length wise edge to edge in the middle. I did recall reading about this on older die lapping cutting write-ups in the past. My assumption would be due to applied pressure on coldplates/heatsinks causing a break. I dunno whatever. No need to panic, 1 step forward 1 step back… Just got back to work to level it out.



    Eventually at around 3000-3500 Grit, I measured a die height of ~ 0.62mm, This was perfect! Until I realized I F*CKED UP ROYALLY! I was not paying attention to the corners and rounded them off! I couldn't believe I missed it, I suppose my previous die and hsf experience years ago made me cocky, (the bit of alcohol in my system didn't help i bet) It was after posting a pic on FB when I zoomed in on the upload to see the clarity of the reflection. My jaw dropped and felt like my balls got ripped off.

    What can I do? F*CK IT! GO BALLS OUT to the point of no return!

    Using 800 grit, I started cutting but the damage was too deep, it required me to drop by down to 600 and work my way back up. I think I caught my mess up. I started lapping with my middle and ring finger and somewhere along the lines my index came in to play (at least this is what I believe)

    Anyways, a perfectly squared mirrored finished die! BUT when checking contact vs the rockit mount via a razer, a clear thin line of light was shinning right through the contact point! I couldn’t stop laughing! I really should have taken a photo but that was the last thing I was thinking about!



    Hoping the bow on the ekwb velocity would permit coverage I did a patch test with some random thermal paste i had lying around and contact was there..., I super under volted my bios and tried running prime large (least power hungry) and saw cores immediately jump to 70C and 92C.

    The contact pressure was just not there!

    The only choice now was to shave down the rockit mount. I worked on the rockit starting with 600 and got it prepped and ready for paint, once completed, things started to look promising.



    Confirmed contact patch test with some TIM , looked good, cleaned and applied what little conductonaut I had remaining, threw on the waterblock and secured it while keeping in mind just how thin the silicon is. As soon as I felt resistance on the posts, 1/8th of a turn and called it.

    NEW CPU DETECTED! (WHAT? I didn't get this the first time I slapped the CPU in)

    Ok ok, I just had to load up my saved profiles, the mem training took longer than ever but it eventually booted and I started my stable stress testing routine to see if it will hold up....

    Turns out it all worked out!






    Memory First: 1000% Pass OK!
    OCCT Medium non AVX 2 hours OK!
    Prime 95 non avx Small 8 hours OK! (after hour 5 or 6, SA/IO usually plays a role from my findings)
    Prime 95 non avx Large 1 hour OK! (to supplement IC Mem testing)
    OCCT Large AVX1 2 hours OK!
    IGPU? Bah who cares! skip!
    PCIE - GPU? i think this is nb/sb? I don't remember, anyway Timespy extreme and ffxv looped. OK!

    Results (Stock Height Direct Die vs Shaved Height Direct Die)
    *this is a case study on it's own, not to be compared with the direct die write up


    5.2ghz sampling was for my own comparison, did a second sample for those that are around the 5.0 or lower mark (see power draw numbers and volts)

    *my original intent was to provide a reference by lowering my RPMs down to compare to larger AIOs / small to medium dual rad setups.... but this info can be pretty much tossed out because I went so damn low that no one will ever consider attempting this willingly LOL. My estimate would be to maybe increase values by 10-15%? (i dunno sorry? bah whatever)

    Hardware
    Case: TT View 71 (don't use the back side window)
    Cooling: HWL 420GTX, 360 GTX and a 360 GTS, EKWB Velocity Block
    Fans: 6x ML 120, 3x ML 140 on the rads (exhausting), and 1 of each for intake

    *Please pay attention to RPM % (all fans) and Ambient where listed
    *CPU Power is based on Average as it fluctuates
    *Voltages are from Vcore (Current read)
    *Temps are Package read outs

    Prime Small FFT (1.5 Hour) Ambient: 24-24.5C
    5.2Ghz @~1.337 (~215W) 25% RPM

    Before
    Max: 88C
    Avg: 84C

    After
    Max: 80C
    Avg: 72C

    5.0Ghz @~1.215 (~176W) 25% RPM

    Before
    Max: 75C
    Avg: 72C

    After
    Max: 71C
    Avg: 64C

    OCCT Small AVX 2 (Do not try this! (or ensure you have a stop limit enabled)
    *note, avx voltages fluctuates a lot, so I didn't bother to log them

    5.2Ghz (265W) (15mins LOL) Ambient: 22-22.5C, 100% RPM (front side window off)

    Before
    Max: 96C
    Avg: 91C

    After
    Max: 87C
    Avg: 81C

    5.0Ghz (225W) (30mins) Ambient: 23C , 65% RPM (front side window off)

    Before
    Max: 89C
    Avg: 85C

    After
    Max: 83C
    Avg: 77C

    Other findings
    Before all of this, the delta between the cores was about 9-10C when stressing. Now I see no more than a delta of 3C between the Cores.
    My hottest cores (0,2,4) before are now on par or lower than my lowest cores.

    Final Reflections
    I am an idiot!
    I am bad at write ups, man this one sucks!
    I really shouldn't have bothered with this write up, it's not worth it for you folks (jk lol)
    I shoulda woulda coulda taken progression pics and log measurements but this should suffice.
    If this 9900k dies due contamination within a year from the LM, I probably wouldn't do it again... or maybe I would have....and gone with kryonaut. I dunno.... Who am I kidding, I probably would!
    Damn this write up took a lot longer than I thought!
    Did I mention I am an Idiot?

    TLDR
    Die lapping, I messed up, rounded edges, needed to sand down to correct, ended up being to low for the die guard to make contact. Die guard got shaved down, It works, I am happy with the results... Till it dies (TBC)



    Followup notes
    cpu still alive, temps still good
    i did not measure height of silicon and rockit mount afterwards cause by this time I really just wanted to get it done with
    will obtain the measurements on next tear down when i do some maintenance on the loop
Working...
X