DAN HSLP-48: A powerful sub 50mm heatsink

Discussion in 'Small Form Factor Systems' started by dondan, Feb 21, 2017.

  1. Your pic is my reflection

    Your pic is my reflection n00b

    Messages:
    36
    Joined:
    Apr 25, 2017
    You've got almost 10 degrees already. And you can't know exactly what the temperature on the C7 would be with the CPU no throttling. And there also are different fan speeds? The result is not relevant. I think that you at least are touching your goal if not have reached it already.
     
    darksable likes this.
  2. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    SaperPL I tested it on the table without the case. Also I made a short test with different orientation of the heatpipes. Only with motherboard vertical and heatpipe bends facing up to pcie port result in bad temps. But that was clear.
     
    Talyrius and SaperPL like this.
  3. SaperPL

    SaperPL [H]ard|Gawd

    Messages:
    1,811
    Joined:
    Aug 31, 2014
    If I understand this correctly:

    1) If someone has a board with CPU socket next to the pcie and chipset on the opposite side, then he'll have this bad orientation of the cooler in A4-SFX and Sentry?
    2) In every standard, not "inverted" chassis layout where the pci-e expansion slots are on the bottom, people with the CPU socket far from PCI-e and chipset in the middle, which is quite standard now for intel mainstream, the orientation will be bad unless there's room above the motherboard?

    EDIT:
    GOOD ORIENTATIONS (MOTHERBOARD INSTALLED VERTICALLY):

    HSLP-48.jpg HSLP-48_2.jpg HSLP-48_3.jpg

    BAD ORIENTATION (MOTHERBOARD INSTALLED VERTICALLY)

    HSLP-48_bad.jpg

    GOOD MOTHERBOARDS FOR HSLP-48 in A4-SFX/Sentry (will fit cooler in good orientation):

    [​IMG]

    BAD MOTHERBOARDS FOR HSLP-48 in A4-SFX/Sentry (will fit cooler in bad orientation):

    [​IMG]
    Do I get this correctly?
     
    Last edited: Jun 11, 2017
    Vlad502, Talyrius, kingtron and 2 others like this.
  4. Jonas Mellkvist

    Jonas Mellkvist Limp Gawd

    Messages:
    159
    Joined:
    Dec 18, 2016
    Nice to se some more test.

    How far away are the next prototypes and have you got any help from the manufacture, what to change to reach your goals ??
     
  5. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    SaperPL: You understand it right.

    Jonas Mellkvist: Today I send a lot of questions to them. The work on the new prototype drawing is nearly done.
    I think it will takes 4 - 5 weeks to get the next sample.
     
    Vlad502 and SaperPL like this.
  6. kingtron

    kingtron Limp Gawd

    Messages:
    252
    Joined:
    Feb 2, 2017
    Do you think maybe because of using thermalright tensioning kit as you said before, and maybe there is not enough contact force between the cpu and the cooler block ?
     
    Vlad502 likes this.
  7. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    The contact pressure is perfect I tested this.
     
    Vlad502 and kingtron like this.
  8. dominathan

    dominathan [H]Lite

    Messages:
    85
    Joined:
    Feb 23, 2016
    Can you get a FLIR 1 or something and check it out? Seeing the heat would tell us so much.
     
    Vlad502 and Talyrius like this.
  9. Julio Espinosa

    Julio Espinosa [H]Lite

    Messages:
    71
    Joined:
    Feb 1, 2017
    Last edited: Jun 11, 2017
  10. Noctua

    Noctua n00b

    Messages:
    8
    Joined:
    Feb 21, 2017
    It's unfortunately normal with those high clocked, high TDP models as the heat just isn't transferred away from the chip into the heatspreader fast enough, thanks to that crappy TIM.

    This is especially bad with very small coolers, even though the cooler itself would be capable of handling the heatload. I've seen this in internal testing numerous times, as I've tested quite a few of those CPUs before and after delid and the results speak for themselves. It's also very obvious when testing the LGA2011-x compatible L9x65 with an 5960X @ 140W, which results in similar temperatures than with a stock 7700K @ 100W.

    Of course one has also to consider that the heat density (W/mm²) of current processors is quite high, especially with the 14nm Skylake/Kaby Lake parts, where ~2/5 of the already small chip isn't used when the load is on the IA cores alone. This makes it even more important to ensure there is good heat transfer from the chip to the heatspreader.

    I've also seen 4770K and 4790K CPUs degrading over time from a temperature point of view, probably from the TIM getting bad, since these CPUs were never overclocked.
    I expect to see this with Skylake and Kaby Lake in 2-3 years as well.

    The fact that Intel seems to be using thermal paste even on the higher end parts (at least with the LCC based models) doesn't make things any better ...
     
    MrLahn, illram, dPunish3r and 7 others like this.
  11. QuantumBraced

    QuantumBraced Gawd

    Messages:
    584
    Joined:
    Nov 21, 2015
    I'm sorry if my comment has been addressed or is asinine, but aren't vapor chambers strictly better at this size than heatpipes? A vapor chamber is essentially one large flat heatpipe that can spread heat very evenly across a low finstack, and that saves more surface area for more fins. And they're not much more expensive than heatpipes. Apparently the HSLP-48 outperforms the T318 (which is a 150W-rated vapor chamber 1U cooler), but is that because heatpipes are better or is it because the design is better and could be improved if a vapor chamber was used? nVidia also uses a vapor chamber for their reference coolers.
     
    Last edited: Jun 12, 2017
    JackGLemmon and Julio Espinosa like this.
  12. SniperCzar

    SniperCzar n00b

    Messages:
    27
    Joined:
    Nov 17, 2016
    Vapor chambers are almost certainly more effective at heat transfer, I remember the difference when NVIDIA started using them for their reference coolers. I'm not sure how effective the vapor chamber base wicking is from the condensation side, but from what I've seen quality vapor chambers have no issues with different orientations. Combination of a vapor chamber with the larger surface area/airflow of Dan's prototype should be a noticeable improvement over a basic vapor chamber (Dynatron)

    Cooler Master are actively designing chamber+pipe combos for this very reason. HP has an advanced VC design with their Z cooler - http://www8.hp.com/us/en/workstations/zcooler.html . Thermacore also researched this, they have a page up here from some contest showing a 1000W load in a FLIR https://contest.techbriefs.com/2014/entries/medical/4010

    I wouldn't be surprised if we see more of these in the 1.5-2U form factor in the future that can dissipate a future 200W+ Xeon load.
     
  13. Jann5s

    Jann5s [H]Lite

    Messages:
    98
    Joined:
    Jul 5, 2016
    One advantage that the HSLP-48 has, is that it's fin-stack is free on both sides allowing air to blow straight through. This aspect of the design also helps draw in fresh air from the outside of the case, which is obviously an advantage over recirculating air inside the case. Perhaps this design can be combined with using a vapor chamber?

    P.S. i'm not promoting a vapor chamber, and I'm actually quite happy with the current design.
    P.P.S. the white paper from HP from one of the sides linked by SniperCzar is quite interesting (direct link to pdf)
     
    Last edited: Jun 12, 2017
    bAMtan2 likes this.
  14. Diverge

    Diverge Gawd

    Messages:
    536
    Joined:
    Nov 22, 2005
    You're forgetting the IHS on pretty much all CPUs, which will spread the heat to the outer heatpipes. I'm not sure the reason for the poor performance of the prototypes... but my modified Nexus Low does an awesome job of cooling, and I haven't even delidded it yet to change the TIM. I will say that I had to modify the mounting brackets, and my heatsink is mounted with a lot of force to the IHS.

    https://photos.google.com/share/AF1...?key=OHR5Rkcwel9mSGYxUV9PYU95VE92T2htTDJXSDNB
     
    Last edited: Jun 13, 2017
    richiegore likes this.
  15. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    UPDATE:

    So my work on the 6 heatpipe design is nearly done. Here are the new dimensions:



    Specifications:

    Socket Support: 1155. 1151, 1150, 1156, 2011 square and narrow Ilm, AM4
    Total Dimensions: 48 (H) mm x 121 (W) mm x 143 (L) mm - including heatpipes
    Fin Area Dimension: 18 (H) mm x 121 (W) mm x 116,5 (L) mm, 59 fins
    Material: copper and aluminum base
    Heatpipes: 4x 6mm heatpipes
    Total Surface: ~128500mm²


    So as you maybe see I changed also the surface size for compatibility. The width is 9mm smaller and I reduced the fin count from 60 to 59 for easier clip mounting with a 120mm fan.
    On the rednerings I used a 120x120x15mm Silverstone fan. I also made some changes to the fan clip mount holes. Now we have holes on the outside for 120mm fans with a thickness of 15mm.
    The inner mount holes are for 15, 14 and 12mm fans.


    Pictures:


    [​IMG]

    [​IMG]

    [​IMG]


    [​IMG]
     
    Last edited: Jun 13, 2017
    MrLahn, solidus2, Talyrius and 18 others like this.
  16. masteralef

    masteralef n00b

    Messages:
    61
    Joined:
    Apr 4, 2017
    Still needs VLP ram with a 120mm fan, correct?
     
  17. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
  18. richiegore

    richiegore Limp Gawd

    Messages:
    322
    Joined:
    Feb 1, 2017
    So I have been thinking and I can't imagine spending so much money on vlp ram. I will likely use the 92mm noctua and I would guess a lot of other people will too. It is by far the most compatible option and I think you should consider making it your priority.

    I was thinking that the outside of the heatsink either side of the fan won't get much/any airflow with the 92mm. See my first rubbish sketch, it is not remotely to scale.

    I was thinking you could profile the heatsink and add a cowl so the 92mm fan acts across the full heatsink. My second image shows one idea but there are lots of ways it could be done.
     

    Attached Files:

    bAMtan2 likes this.
  19. dominathan

    dominathan [H]Lite

    Messages:
    85
    Joined:
    Feb 23, 2016
    Did you manage to get an infrared camera so we can see the heat distribution? We need data!
     
    Talyrius and Rysen like this.
  20. Chapeau

    Chapeau Gawd

    Messages:
    750
    Joined:
    Jul 17, 2016
    Dondan, thats a beautiful render right there. I love that model of the silverstone fan... Nice.
     
  21. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    This is not possible, because on some boards you can't center the fan 100% if you not use vlp ram.

    Currently I don't have the plan to get one.
     
  22. ng4ever

    ng4ever [H]ard|Gawd

    Messages:
    1,441
    Joined:
    Feb 18, 2016
    Oh how I wish this would come out already!

    I hope it is ready to buy within a year or less!
     
  23. Jonas Mellkvist

    Jonas Mellkvist Limp Gawd

    Messages:
    159
    Joined:
    Dec 18, 2016
    Heatpipes: 4 x 6 mm feels not correct ?

    What diameter are the 6 Heatpipes ??
     
  24. Jonas Mellkvist

    Jonas Mellkvist Limp Gawd

    Messages:
    159
    Joined:
    Dec 18, 2016
  25. kingtron

    kingtron Limp Gawd

    Messages:
    252
    Joined:
    Feb 2, 2017
    according to Asrock ... YES
    check the link, memory section - 4th line
    - Supports ECC UDIMM memory modules (operate in non-ECC mode)

    http://www.asrock.com/MB/Intel/Fatal1ty Z270 Gaming-ITXac/index.asp#Specification



    and this will work also:
    https://www.server-hardware.com/de/shop/mta18adf2g72az-2g3b1,227593.html
     
    Julio Espinosa likes this.
  26. richiegore

    richiegore Limp Gawd

    Messages:
    322
    Joined:
    Feb 1, 2017
    So on some boards you can't have a 92mm fan centered? I thought the Intel "keep out zone" was that size...

    I still think you should really consider ways of getting airflow to the whole heatsink with a 92mm fan.
     
  27. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    Asrock X99E-ITX and I think for every AM4 board.
     
  28. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    I had a bit of time and played with my render software. A nickel plate version of the HSLP-48. I choose a olive colored type:

    [​IMG]
    (Click for 3K)
     
    davaeron, Jspr, Woxys and 9 others like this.
  29. Rysen

    Rysen Limp Gawd

    Messages:
    407
    Joined:
    Jan 10, 2017
    What do you use for rendering? Keyshot?
     
  30. richiegore

    richiegore Limp Gawd

    Messages:
    322
    Joined:
    Feb 1, 2017
    I've been looking at the l9a and the new am4 mounting kit, it all looks symmetrical, is there actually a slight offset to clear the ram or something? I can't figure it. Can you clarify?

    I'll take your word about the asrock as you own it! :)
     
  31. dominathan

    dominathan [H]Lite

    Messages:
    85
    Joined:
    Feb 23, 2016
    So you're just gonna keep making prototypes until one works?
     
    Talyrius and Rysen like this.
  32. Julio Espinosa

    Julio Espinosa [H]Lite

    Messages:
    71
    Joined:
    Feb 1, 2017
  33. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    TDP is a useless value. You can cool a 140W TDP 5820k easier than an 91W TDP 7700k.

    The reason is soldered vs. TIM connectiom between DIE and hestspreader.
     
  34. backfeed

    backfeed Limp Gawd

    Messages:
    480
    Joined:
    Jun 22, 2010
    Correct me if I'm wrong, but a heatsink having a reflective metallic surface could be pretty hard to measure using an IR camera because of IR reflections. There's a simple trick to avoid the reflection problem, by placing a piece of thin matte black tape where measurements are desired and waiting a while until the tape temperature matches the surface it covers as closely as possible, and then take the measurement on the tape. But that could be pretty impractical with a heat sink, especially when it comes to the fins, and covering every surface of the entire heat sink would be infeasible.

    Also, the heat source would preferably need to be stable, something like an adjustable heat plate (adjustable in order to simulate different heat levels).
     
    Last edited: Jun 15, 2017
    richiegore likes this.
  35. Woxys

    Woxys [H]Lite

    Messages:
    84
    Joined:
    Oct 31, 2014
    Do you plan to test your current sample on some RYZEN processors before changing the design? I fell like the soldered CPU's will help a lot :)
     
  36. dondan

    dondan [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Apr 15, 2013
    My 5820k is also soldered.

    Maybe it will perform better on a Ryzen CPU but also the competitor heatsinks will perform better on it. This means the CPU that I use for testing doesn't matter.

    Only the temp distance to competitor products is interesting for me. I mentioned it earlier i like to be ~15°C better as the C7 or L9i.
     
    MrLahn, Woxys, Talyrius and 3 others like this.
  37. Jonas Mellkvist

    Jonas Mellkvist Limp Gawd

    Messages:
    159
    Joined:
    Dec 18, 2016
    That i can understand but you are shure your mounting kit from Thermalright got right pressure on the CPU to Hlsp48 ?

    Did you get Nexus version with correct 2011 mounting kit and tryed it before order the new version ?
     
  38. zerodeefex

    zerodeefex Limp Gawd

    Messages:
    264
    Joined:
    Apr 8, 2008
    I'm very excited to see what is possible when running fanless. I have a de-lidded, CLUed 6100 running below 6100T voltages that I'm hoping to run passively with this cooler.
     
  39. pedrofan

    pedrofan n00b

    Messages:
    55
    Joined:
    Feb 10, 2016
    After comparing myself t318 vs cooljag falcon II (which is almost the same as HSLP-48 V1) and both with several fans, on my 155 watts 22 cores xeon; the only combination that did not lead to +100ºC while rendering is t318 with Thermalright ty100 fan in push (closing the case did not affect the result, which is very confusing, and no other fan performed better, including noctua, and gamerstorm gs120)

    Falcon failed with all fan combinations.
    T318 holds 88º for one 4k render and +95º for concatenated renders wich is not common in my workload. For me it's enough to work, since gaming is not a problem at all. Maybe topping at 68º or so.


    Let's see if V2 enhance the results, I'm very excited.
     
  40. Steve0Kinev0

    Steve0Kinev0 n00b

    Messages:
    51
    Joined:
    Jan 9, 2017
    Does this Xeon come with crappy TIM on the die or is it soldered? That is... If you were to delid, overall curious to see how much that would be holding it back?