AMD Hawaii series (NV 780Ti) thermal issues pin pointing theories.

Discussion in 'General Hardware' started by DeOdVue, Nov 7, 2013.

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

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    First off, I am NOT an expert! :p I do this as my hobby. I do not have any of these cards (couldn't afford to keep up anymore. Got to take care the wify and the housy! Most of all, Uncle Sam took 1/3 of that! :D), therefore, none of it is conformed nor conclusive and most worse of all, it is my opinions! :p

    Here go nothing:
    Theory 1, let me make a wild guess... maybe, probably, slight chance, tiny weeny bit of chance... the AMD's engineer team tested the GPU lying down on the test bed and called the date! In this setup (Horizontal) vapor chamber technology had the advantage over the heatpipe. Yeah, why bothered while the AIBs partner going to design their own cooler anyways. Also, resources and money is not a luxury for AMD. "Why not let the AIBs partner launched their own cooler at launch?" I heard someone asked, once. Well, Didn't you noticed the tight lips. With all the leaks, spies.. lawsuits... hint, hint...NV in the house!? ...Sometimes earlier this year... ?!
    Anyways, from the pictures (Guru3d.com and other various websites)the cooler is a vapor chamber base. You might/not know the disadvantage of the vapor chamber base is mounting it vertically. Unless, the technology had overcome this issue which I haven't followed. The issue with vapor chamber mounting vertical is that the fluid inside the chamber will draw downward by gravity and therefore not affectively cooling the components properly. This might be hinted why the aftermarket (heatpipe technology) cooler worked great!
    Kudos to AMD for thinking of using vapor chamber technology to cool the GPU. Yet, at the same time, implemented the cooling design on GDDRAMs with the same cooler. Thinking it will cooled the GDDRAMs while it cooled the GPU. The problem is, they have to overcome the height differences from various components, PCB bending, warp... and without the BACKPLATE ... to overcome this issue, they used thermal pads. In my opinion, the affectivity of thermal pad is not that good and sometimes its worked against you (just from my personal experienced). I'll guess it worked against AMD designed in this case. see theory 2 for why.

    Theory 2, judging from the thermal image (Front) provided by Guru3d.com, the cooler itself seemed to be cool... Toooo cooled for thermal transferring from the GPU. AT which, the back thermal image shown M1=94.6C, M2=79.8C and M3=85.9C. Keep in mind, M1 is at the back of the PCB, through 11(?) layers of PCB and/of coppers plates and traces. Prior to reaching the first top PCB layer is the BGA soldered balls joint on the GPU. As you can IMAGINED THE HEAT at the core of the GPU... this heat needed to be released somehow and as affectively as possible. Therefore, the core GPU must/needed to be pressing against the cooler as much as possible. How much pressure? <-- As much as possible for better thermal transferred. <-- than you cracked the core! ;p Hint, hint... the outer metal frame. The million dollars question is "What is the gap/spacing at the TOP FLAT SURFACE of the GPU and the TOP FLAT SURFACE of the metal frame?". Thermal compounds can do so much... Hint, hint Intel thermal issue.. It is very important the height of the GPU core and the metal frame is to be close as possible (tolerance?). Enough of this craps, let move on.
    As mentioned above, the cooler at the GPU core should be reading 95C PLUS and radiated out... and I'm not seeing that (front thermal image). You can see the cooler is hot but not anywhere neared 90C... The HS1=86.5C seemed to be pointed at the front layer of the PCB, GDDRAM... more than 1/2 degree hotter made sense since it is the top layer of the PCB. While the back thermal image shown PCB is trapped with heat. Hottest at M1 (GPU) and radiated out (lesser at the blower). The GDDRAMs shouldn't be at that temp. This kind of temperatures at the GDDRAMs were caused by the trapped heat from the thermal pads (can released heat fast enough at saturation) but mostly by the heat that trapped by the fiberglass of the PCB layers within them are coppers plane, pads and traces. Virtually no air circulated the top of the PCB, obviously. Another evident is the back retention and screws. It is hard connected to the cooler itself, it is part of the cooler and as you can see how cool (not hot) it is. So, to pin point it out, I would said LOOK AT THE GPU CORE AND THE METAL FRAME FOR THE FIXED!
    And then you putted the two together, we got the R9-290/X thermal issue!
    "Then why does the same basic (minor differences on the cooler and open top) designed on the NV 780 and the Titan worked but not AMD?", you asked. NO, IT'S NOT! Take a good looked at thermal status of the 780TI and you will see what I've meant. NV is suffering the same issue but at the lesser extent and the 780TI shown it limited. Hint, hint.. aftermarket cooler cooled much better than the reference.
    There will be more evidences if you view the data careful enough. But that should summed it up with my 2 theories. Again, worst of all... it is my opinion! :p How did I came up with these two theories? <-- Let's just said I giggled when I first saw the 780 and the TiTan cooler design and then I laughed hard when the R9-290 series came out! Go. Dam.it, someone have hacked into my computer and got a copy of my Sapphire Radeon 2900PRO modified pics. Excepted, my is much bigger, cooler and much more kick asses!

    And that is my two cents...Errggg... Theories...
     
  2. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
    that was very hard to read so I really didnt get past a couple of sentences.
    my theory is that the cards originally were designed to run at lower frequencies, but the performance would not be there vs the titan and 780 both released months ago.
    either that or they really cheaped out on the stock coolers
     
  3. eclap

    eclap Banned

    Messages:
    31,468
    Likes Received:
    4
    GPU:
    Palit GR 1080 2000/11000
    +1 wtf wall of text, nobody is going to read that OP, try making your posts a bit more readable.
     
  4. Loophole35

    Loophole35 Guest

    Messages:
    9,797
    Likes Received:
    1,161
    GPU:
    EVGA 1080ti SC
    The stock cooler was not engendered for the Hawaii core (probably not even the Tahiti either) it's one they've been using for a while. And yes I also get the feeling that they weren't originally supposed to be clocked at 1000Mhz maybe 850Mhz seeing as that is where they like to down clock to in quiet mode.
     

  5. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    My apology for the wall of text. Obviously, English is not my major. That wall of text are the evidences, data and technical inform. pin pointing to the reasons why the cooler failed to cool the GPU properly in the Hawaii series specific and others as a general, whether that NV or AMD products.

    Secondly, You're right. I should learned how to make a post, as you can see, I'm a newbie! :) I'll learned from you and others, from now on. I thank you. I'll summed those up in a couple of words from my previous post.

    Failure #1: Vapor chamber does not affectively worked vertically! <-- Go experimented.. come back later and let me know the resulted!

    Failure #2: The height of the silicon (GPU core) and the height of the metal frame have to me as closely as possible. If the height of the silicon is higher than the metal frame the cooler might crushed! An if it's to low, the cooler won't affectively cooled the silicon but cooling the frame metal itself.

    Among other failures such as no airflow on the PCB surface which evidently shown in the thermal image the Hilbert provided in the reviews. The thermal image clearly shown the heat been trapped and I'm pointing out all the obvious which you might/not see and what's might has caused the issue.

    All these issues will resolved with the aftermarket coolers except for failure #2. Water and heatpipe is the way to go in the vertical setup vs vapor chamber. (vapor chamber performed better than heatpipe in the horizontal position).

    Failure #2 is relating to the controlled height of the metal frame in relation to the height of the silicon (GPU core). I don't have the actual chip itself to measure it. Can some one provide me that inform.?

    THERE, is that how I supposed to post?! :p Much better now? :)

    BTW: @eclap... my advice to you. DO NOT go to the library! Those walls of book will DROWNED you! learned to be patient and find that needle in the hays!

    Again, those were my two theories... Errrgg... cents!
     
  6. teleguy

    teleguy Maha Guru

    Messages:
    1,363
    Likes Received:
    221
    GPU:
    GTX 1070/Vega 56
    Didn't AMD promise to redesign the cooler a while back?
     
  7. k3vst3r

    k3vst3r Ancient Guru

    Messages:
    3,702
    Likes Received:
    177
    GPU:
    KP3090
    my theory, they simply couldn't be bothered designing new cooler

    cooler for 780/titan costs fair bit of money for NV to design/make like $80

    so AMD just decided launch it does the job
     
  8. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    ^^ Read theory #1 in regarding to vapor chamber cooling technology. NV is not exempted from this issue. Note: 780TI is the evident of the vapor chamber hit it diminishing returned. 780TI will have that issue when oc... thermal limitation!

    P.S. ~250W - 300W is their limitation and the diminishing returned... begin :)
     
    Last edited: Nov 8, 2013
  9. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    Why bothered when you (AMD) had limited resource and finance, while the AIBs partner will do that anyway!.

    Why reinvented the wheel while the wheel is already at it best!
     
  10. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    Quote: --> my theory, they simply couldn't be bothered designing new cooler

    cooler for 780/titan costs fair bit of money for NV to design/make like $80

    so AMD just decided launch it does the job

    Sources please?
    It might cost more but not a fair bit.

    Hell, I even think they were made by a same company!
     

  11. CPC_RedDawn

    CPC_RedDawn Ancient Guru

    Messages:
    10,413
    Likes Received:
    3,078
    GPU:
    PNY RTX4090
    Anyone considered they are still using the same 28nm process for this complete and utter beat of chip. With that many transistors on 28nm and a 512bit bus its GOING to draw a ton of power and produce a ton of heat!

    Aftermarket coolers will sort it out no problem, and once they drop down to 22nm or further for Volcanic Islands we should hopefully see a nice drop in temps as well.
     
  12. k3vst3r

    k3vst3r Ancient Guru

    Messages:
    3,702
    Likes Received:
    177
    GPU:
    KP3090
    Yawns :infinity:

    http://www.tomshardware.com/reviews/nvidia-history-geforce-gtx-690,3605-2.html

    TL : DR read whole thing show's how much effort goes into making reference heatsinks for 780/titan
     
    Last edited: Nov 8, 2013
  13. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
    what are you talking about in this post?
     
  14. Redemption80

    Redemption80 Guest

    Messages:
    18,491
    Likes Received:
    267
    GPU:
    GALAX 970/ASUS 970
    k3vst3r nice article, never saw it and very interesting.

    Nvidia really do put a lot of time and money into the coolers now, arguably too much when AMD are not spending much to keep the cards as cheap as possible.
     
  15. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    QUOTE:

    Yawns

    http://www.tomshardware.com/reviews/...90,3605-2.html


    Quote:
    In my reviews, I’ve also commented on the fact that Nvidia went with a magnesium alloy fan housing for GeForce GTX 690, and then ditched it for Titan, 780, and 770. My understanding was that this had something to do with cost. But Andrew told me there really wasn’t much difference between magnesium and aluminum—at least as far as the metals go. More significant was the manufacturing difficulty.

    Quote:
    But the biggest issue was that manufacturing yields started off around 10%. That’s not the GPU, mind you. Just the coolers. Because they were going from raw magnesium to plating, with no brushing or sandblasting in between, all of the dents from the casting line showed through. Someone had to manually sort through everything coming off the line and pick out the pieces without those imperfections. Even then, if you look closely enough, it’s still possible to pick out tiny divots in the metal.

    TL : DR read whole thing show's how much effort goes into making reference heatsinks for 780/titan

    ------

    OK, awesome, that's shown NV made expensive cooler. So therefore, we can assumed AMD made cheap cooler? <--- I don't understand that logic

    Just like saying I'm stupid, therefore you are smart! in some case, VISE VERSA.

    P.S. I do read Tom'sHardware and with respected! But don't you think they have been bias if not little?! :) Especially, the timing of that exact article? Just wondered.
     
    Last edited: Nov 8, 2013

  16. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    It's become obvious to me that quite a few people do not understand how the basic heatpipe and vapor chamber worked.

    Let me go deeper into theory #1, how vapor chamber worked in THEORY and in this case. Actually, this would applied to all heatpipe or vapor chamber coolers out there in the market.

    In theory, the vapor chamber works better than the heatpipe technology, since the vapor chamber had more contacted surface area than heatpipe. (Google them for more detail information). Basicly, when heated (WHERE ?), the liquid in the chamber or pipe will vaporized, travel upward 'till it cooled enough to turn back to liquid form and repeated the cycle. How to cool it and channel the liquid back as fast as possible is the name of the game.

    In reality, the bottom cooler is not where the GPU core located at. Only when you laid your video card flat on the surface then the GPU core is truly at the bottom, otherwise it is at elsewhere. So, as most reviewer tested (vertically) as shown in Guru3D.com, from the image that Hilbert posted, the GPU core is somewhere in the middle. Worst of all, in our tower case , it would be upside down! <-- think about that! As it is upside down, the liquid is at the cool side (drawn by gravity) and the heated side (GPU core) which will be at the top. <-- This is the reason why!

    {@}
    ^)^ <-- Reality doesn't look like that
     
    Last edited: Nov 8, 2013
  17. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    Ok, might as well. Let's going deeper into theory #2: I am questioning and pin pointing out the height of the metal frame located around the GPU. This metal frame was designed to protect the silicon (GPU core) been crushed by the cooler when installed. The height of the metal frame must be controlled. As the metal frame is higher than the silicon than the cooler won't be cooling the silicon affectively. And if, the silicon is higher than the frame metal (how much higher?), the chances are, the cooler might crushed the silicon. This is important to how affectively the cooler released the heat. And how affectively the cooler worked is by the design.

    The last time I checked, AMD is a CPU/GPU company and so is NV. Their cooling thermal management, SUCK!
     
  18. k3vst3r

    k3vst3r Ancient Guru

    Messages:
    3,702
    Likes Received:
    177
    GPU:
    KP3090
    :confused: Results speak for them self's. AMD cheaped out on cooler everyone knows it. That's why 290x/290 hit 95c an 780/Titan/780 Ti don't
     
  19. ---TK---

    ---TK--- Guest

    Messages:
    22,104
    Likes Received:
    3
    GPU:
    2x 980Ti Gaming 1430/7296
    never have I saw anybody type so much and say so little. Of course nvidia stock coolers are better at cooling, the info is all over the interwebs
     
  20. DeOdVue

    DeOdVue Guest

    Messages:
    48
    Likes Received:
    0
    GPU:
    AMD, HD5850
    better resulted does not automatic concluded the others are cheap out. Again, where is the source that AMD made it cheap? Until than, I'm not judging.

    The 290 PERFORMED better than the 780... does that mean the 780 "cheaping" out?
     
Thread Status:
Not open for further replies.

Share This Page