IceMan AMD Threadripper Water Block Breakdown

Discussion in 'Overclocking & Cooling' started by Kyle_Bennett, Jan 10, 2018.

Thread Status:
Not open for further replies.
  1. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    IceMan AMD Threadripper Water Block Breakdown

    AMD's Threadripper CPU is finding its way into many computer hardware enthusiasts' HEDT systems, and if you are looking to overclock it to the edge-of-the-envelope 4GHz, you are very possibly going to look to custom loop cooling. We just got in the IceMan Threadripper block and we tear it down to see how it is designed.
     
    Last edited: Feb 27, 2018
    Emission likes this.
  2. Dekar12

    Dekar12 Gawd

    Messages:
    734
    Joined:
    Oct 2, 2003
    Not sure if the water jetting makes entirely that much difference, but not having fins covering the entire cpu die under the heatspreader seems to affect cooling capacity.

    The surface area is almost large enough on these things, it almost feels like these blocks could really benefit from having two inlets to split water across each core. And by benefit, I mean like a 1-3c cooling difference, lol.
     
  3. cyberguyz

    cyberguyz Limp Gawd

    Messages:
    259
    Joined:
    Aug 28, 2014
    Having the fins running longitudinally along the length of each die would cause the ccx located at the end closest to the edge of the cpu cold plate edge to run hotter as they would be cooled by water that already picked up heat from the neighboring ccx closest to the inlet end. Having the fins running perpendicular to the dies would let each ccx to be cooled by water unheated by its neighbors.

    This of course is pretty tough to verify since it seems that the Ryzen CPUs only gives us per die (package) temperatures rather than per-core temp like Intel. I would suspect the cpu would report cooler composite temps anyway though.
     
    Last edited: Jan 10, 2018
    Kyle_Bennett likes this.
  4. owcraftsman

    owcraftsman Gawd

    Messages:
    727
    Joined:
    Feb 3, 2007
    you'll be taking that back apart to get the lighting back in there :D GL hate it when I do that.... Look forward to see the results
     
  5. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    Left it out on purpose.
     
  6. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    Well, the first outing looks to be totally FUBAR.

    Thread pitch on the hold down screws is not correct. Striped off. Silver screw is from IceMan TR4 block. Black screw is from XSPC and still threads in and holds very solid.

    Tread Pitch IceMan TR4 Water Block.jpg

    This is the mount it got. Went to 83C+ almost immediately and hard-locked the OS.

    Mate IceMan TR4 Water Block.jpg
     
  7. Elkwood

    Elkwood Limp Gawd

    Messages:
    162
    Joined:
    Sep 17, 2017
    Hmm that doesn't look good at all !! Hope it doesn't mess up the board threads. That be a real pain
     
  8. AceGoober

    AceGoober Live! Laug[H]! Overclock!

    Messages:
    22,557
    Joined:
    Jun 25, 2003
    Maybe it is just me but it seems somebody didn't form that Iceman screw all that great. There is surface pitting all over it. Not a good sign indeed.
     
  9. KarVi71

    KarVi71 n00bie

    Messages:
    15
    Joined:
    Dec 14, 2016
    Yes, I would have done the same :)

    Sadly it seems that this promising looking block is being let down by badly designed and manufactured screws.

    Too bad, it looks like it could have performed OK.
     
  10. artforz

    artforz n00bie

    Messages:
    5
    Joined:
    Mar 24, 2017
    Mk1 eyeball says that screw's UNC #6/32, SP3/TR4 is supposed to be M3.5x0.6.
    Oh well, at least they managed to get it REALLY wrong and didn't go for #6/40...
     
  11. DigitalGriffin

    DigitalGriffin [H]ardness Supreme

    Messages:
    4,401
    Joined:
    Oct 14, 2004
    Kyle,

    You're assumption about the cooling flow traversing the long length of the die versus the short length is quite correct due to Newton's law of cooling. It is a generally less efficient approach*. NOW if the heat transfer is thermal conductively limited, it will make little to no difference. (Think about walking along hot coals with your bare feet. You don't get burned because the thermal conductivity of coals is low)

    EDIT:
    My above statement is based on a similar surface velocity. However I'm going to assume the total cubic flow rate between blocks is similar. If you rotate it so that you are traveling down the long side, the surface velocity increases because of a narrower path. But you are pumping more fluid past the block in a given time period. At this point you get down to differential equations for the liquids and surface materials in question.
     
    Last edited: Jan 11, 2018
  12. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    That is what I learned in my physics classes. :) When I discussed the same issue with Koolance about its TR4 block they were in total agreeance, but I do know if the design change or testing will be made in the future. Also, if you check our EK TR4 water block review you will see where us turning the block 90 degrees made a huge difference in performance. So I feel pretty good about thermal conductivity of the die, the solder, and the IHS being pretty damn good. Of course TIM comes into it as well and prepping the TIM and its application is extremely important. I am actually using a heat gun now to "pre-heat" the IHS as it makes the application of the Prolimatech PK1 much easier.
     
  13. mvmiller12

    mvmiller12 Limp Gawd

    Messages:
    442
    Joined:
    Aug 7, 2011

    That sounds excessive.........AND AWESOME! I would love to see that in-video sometime.

    Also (unrelated): How did YOU end up being Member #3 of the [H}ardForum, or am I totally misunderstanding what that "member: 3" meant when I quoted your post?
     
  14. mnewxcv

    mnewxcv [H]ardness Supreme

    Messages:
    5,839
    Joined:
    Mar 4, 2007
    Well if Kyle is #3, and he joined in 1997, Al Gore must be #1. But then, who is #2? :confused:
     
  15. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    It is a heavy duty hairdryer for all intents and purposes, not exactly exciting.

    Account 1 was the guy that did our software install back in '97, and I am not even sure if account 2 was anyone honestly...
     
  16. RogueTadhg

    RogueTadhg [H]ard|Gawd

    Messages:
    1,374
    Joined:
    Dec 14, 2011
    Damn. Kyle, have you spoken to the maker about the poor-ass mating? I haven't seen any blocks that bad in a long time.
     
  17. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    The block looks great, just looking at it. The screws he is using are simply the wrong thread pitch. That said those are so soft that I could not really tell they were crossthreading when screwing those in. Thankfully the mounting points on the motherboard were not damaged at all. I can still lock in the proper threaded post nice and tight.
     
  18. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    So I did get a response, here it is in full. I sent the same pictures that are shown above.

    IceMan: Hi, what's the problem?

    z1.jpg z2.jpg

    Kyle: The thread pitch is incorrect on the screws.
    Count the thread on your screw compared to the correct one pictured below...
    Your screws crossthread in the mounting holes.

    IceMan: I know,we will send again.
     
    DrezKill and RogueTadhg like this.
  19. rgMekanic

    rgMekanic [H]ard|News Staff Member

    Messages:
    5,064
    Joined:
    May 13, 2013
    Meme_-4.png
     
  20. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    OK, so we got the correctly threaded hold-downs in and I got to testing. Immediately hitting 85C die temps and hard-locking the system.

    With our proven TIM application:

    Mate spread with two dots.jpg

    Going with a single thin layer application, things got even worse.

    Mate IceMan TR4 Water Block-Single Layer Spread.jpg

    So what is the issue? Ah, the surface of the cold plate is concave. That is not good for TR.

    IMG_20180220_192336.jpg
     
    DrezKill likes this.
  21. mnewxcv

    mnewxcv [H]ardness Supreme

    Messages:
    5,839
    Joined:
    Mar 4, 2007
    wow, that is pretty bad for a block designed for threadripper. How did that even make it through in house testing?
     
  22. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    Not exactly sure, but after sending out mounting posts with the wrong thread pitch, I guess anything is possible. This one is not going on the buy list, that is for sure.
     
    DrezKill likes this.
  23. Nebulous

    Nebulous [H]ard|Gawd

    Messages:
    1,362
    Joined:
    Nov 16, 2005
    Dang. In your initial breakdown the coldplate was flat. Is this the same coldplate?
     
  24. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
    Good eye! What is shown in the video is "correct," but I was a bit too quick in my evaluation. the curve of the coldplate run long-ways down the mating surface, but NOT across the short side. Think about the shape of the cold plate being a section of a tube. This is why my straight edge did not pick it up as I laid it diagonally on the surface. Live and learn.
     
    Nebulous likes this.
  25. Nebulous

    Nebulous [H]ard|Gawd

    Messages:
    1,362
    Joined:
    Nov 16, 2005

    Ahhh, gotcha! Yeah I was alittle confused when I saw you checked with the straight edge and I saw it flat, but when you mated the block and TR the foot print was horrible. Aside from the fins not going far enough either, I doubt this block in going to be recieved well unless ICEMAN does a complete redesign. A shame as i thought it did look really nice.
     
    Kyle_Bennett likes this.
  26. magoo

    magoo [H]ardForum Junkie

    Messages:
    14,497
    Joined:
    Oct 21, 2004
    LoL.......

    Kyle: You're block ain't flat.

    IceMan: I know. We will send again.
     
  27. Kyle_Bennett

    Kyle_Bennett El Chingón Staff Member

    Messages:
    52,331
    Joined:
    May 18, 1997
Thread Status:
Not open for further replies.