Incredible NVIDIA RTX 5090 Founders Edition: Liquid Metal & Cooler ft. Malcolm Gutenburg

00:38:04
https://www.youtube.com/watch?v=-p0MEy8BvYY

Résumé

TLDRDans cette vidéo, Malcolm Gutenberg, ingénieur thermique chez Nvidia, présente les innovations derrière la carte graphique RTX 590 Founders Edition. Il discute des nouvelles techniques de refroidissement, notamment l'utilisation de métal liquide et d'une chambre à vapeur 3D, qui améliorent le rendement thermique. Les designs aérodynamiques, comme la conception 'blow-through', facilitent une meilleure circulation de l'air, ce qui est crucial pour maintenir des températures basses tout en réduisant le bruit. Les prototypes révèlent une progression impressionnante vers des cartes graphiques plus puissantes et plus compactes, illustrant comment chaque élément est soigneusement optimisé pour la performance.

A retenir

  • 🔥 Malcolm Gutenberg explique les coulisses de la RTX 590.
  • ⚡ Une chambre à vapeur 3D révolutionnaire pour une meilleure dissipation thermique.
  • 💧 Le métal liquide est utilisé pour optimiser la conductivité thermique.
  • 🌪️ L'architecture 'blow-through' améliore le flux d'air et réduit le bruit.
  • 🔧 Le PCB joue un rôle crucial dans la gestion thermique des composants.
  • 📏 Des designs compacts pour un rendement accroché sous le capot.

Chronologie

  • 00:00:00 - 00:05:00

    Introduction de Malcolm Gutenberg, ingénieur thermique chez Nvidia, qui présente le refroidisseur RTX 590 Founders Edition tout en rappelant ses projets précédents avec le RTX 490.

  • 00:05:00 - 00:10:00

    Discours sur les prototypes et les évolutions thermiques des cartes graphiques, y compris le système de chambre à vapeur 3D qui améliore la dissipation thermique.

  • 00:10:00 - 00:15:00

    Discussion sur la conception aérodyamique du nouveau modèle RTX 590, y compris les effets du refroidissement à double flux et la répartition de la chaleur.

  • 00:15:00 - 00:20:00

    Exploration des défis techniques liés à la meilleure orientation et au débit d'air, ainsi que des options de conception pour réduire les pertes de pression dans le refroidissement.

  • 00:20:00 - 00:25:00

    Description des tests internes avec des prototypes, y compris l'utilisation de blocs chauffants et de thermocouples pour évaluer la performance thermique et acoustique des unités graphiques.

  • 00:25:00 - 00:30:00

    Analyse approfondie de l'utilisation du métal liquide pour les interfaces thermiques, y compris les considérations de fiabilité et de durabilité du matériau.

  • 00:30:00 - 00:38:04

    Discussion des améliorations sur les plaques arrière et les designs de PCB qui optimisent la performance thermique totale, tandis que Malcolm souligne l'importance de la conception intégrée.

Afficher plus

Carte mentale

Vidéo Q&R

  • Qu'est-ce que la chambre à vapeur 3D ?

    C'est une structure qui améliore l'évacuation de la chaleur en connectant directement les canaux de chaleur au refroidisseur.

  • Pourquoi utiliser du métal liquide ?

    Le métal liquide offre de meilleures propriétés de conductivité thermique, permettant un transfert de chaleur plus efficace.

  • Quels sont les avantages de l'architecture 'blow-through' ?

    Il réduit la résistance à l'air, améliore le flux d'air et diminue les niveaux sonores.

  • Comment la conception thermique évolue-t-elle ?

    L'évolution comprend des designs plus compacts avec une densité accrue de composants tout en optimisant le refroidissement.

  • Qu'est-ce qu'un 'wick' dans la chambre à vapeur ?

    C'est un matériau qui aide à transporter le liquide de retour vers l'évaporateur par capillarité.

  • Quel rôle joue le PCB dans la chaleur du GPU ?

    La disposition et la dissipation de la chaleur des composants sur le PCB influencent directement la performance thermale.

Voir plus de résumés vidéo

Accédez instantanément à des résumés vidéo gratuits sur YouTube grâce à l'IA !
Sous-titres
en
Défilement automatique:
  • 00:00:00
    today we're joined by the most excited
  • 00:00:01
    thermal engineer in the industry when
  • 00:00:04
    are you doing 43ds blue when are you
  • 00:00:06
    breaking the law can't talk about that
  • 00:00:08
    one that unreleased that's another
  • 00:00:10
    prototype you'll get this is Malcolm
  • 00:00:12
    Gutenberg from Nvidia who previously
  • 00:00:14
    joined us to cut an RTX 490 cooler and
  • 00:00:17
    half for science this time he served as
  • 00:00:19
    the lead thermal engineer on the RTX 590
  • 00:00:22
    Founders Edition and he brought a lot of
  • 00:00:25
    prototypes and a lot of vocabulary this
  • 00:00:28
    is a pure aerodynamic change so
  • 00:00:30
    coefficient of thermal expansion
  • 00:00:31
    application of thermal couples is a is
  • 00:00:33
    an art hermetic seal rubber gasket here
  • 00:00:36
    heat of vaporization you can actually
  • 00:00:38
    calculate the mass flow rate based on
  • 00:00:39
    the power even ampere we started looking
  • 00:00:42
    at liquid metal redirect the air flow
  • 00:00:44
    away from the inlet of the fan in this
  • 00:00:46
    video Malcolm shows us liquid metal on
  • 00:00:48
    the RTX 5090 the design and the
  • 00:00:50
    considerations around it a world first
  • 00:00:53
    with a horizontal 3D Vapor chamber
  • 00:00:54
    piping structure and we learn about
  • 00:00:56
    aerodynamics and thermal design let's
  • 00:00:59
    get into it
  • 00:01:01
    hey everyone I am joined by Malcolm
  • 00:01:03
    thermal engineer and thermal evangelist
  • 00:01:06
    at Nvidia Malcolm you've joined now for
  • 00:01:09
    I think three videos yep yep every time
  • 00:01:12
    Malcolm's been on he has brought a box
  • 00:01:14
    of mysteries and this time there was a
  • 00:01:17
    very large suitcase of mysteries uh so
  • 00:01:20
    this is for the 90 we're going to go
  • 00:01:22
    through a lot of stuff today but Malcolm
  • 00:01:24
    if you could help walk me through a few
  • 00:01:26
    of the things on the table sure to give
  • 00:01:28
    people a primer yeah 50 90 you know
  • 00:01:30
    thermally it is incredible what we've
  • 00:01:33
    been able to achieve and this is kind of
  • 00:01:35
    the evolution on how we got to 90 so
  • 00:01:37
    obviously 4090 we have some you know
  • 00:01:41
    very early flow and acoustic samples
  • 00:01:43
    here and then new this generation is a
  • 00:01:45
    3D VC so here is kind of the evolution
  • 00:01:48
    of the 3D VC from its very early infancy
  • 00:01:52
    through towards the final 3D VC VC here
  • 00:01:55
    meaning Vapor chamber Vapor chamber yeah
  • 00:01:58
    and this here is the incredible 3D VC
  • 00:02:01
    where the heat pipes directly connect to
  • 00:02:04
    the vapor chamber featuring a new artery
  • 00:02:07
    Wick in the evaporator area yeah I'm not
  • 00:02:10
    sure if the shadow will help the camera
  • 00:02:11
    see it but there's some really slight
  • 00:02:14
    what was the the difference the Delta
  • 00:02:17
    say 300 Micron so so there's like 300
  • 00:02:20
    Micron height difference in the is is it
  • 00:02:22
    accurate to call it cented powder yeah
  • 00:02:24
    this is cented powder and this is also
  • 00:02:26
    using three different types of Wick so
  • 00:02:29
    we have the braid Wick the centered
  • 00:02:31
    powder and then a mesh wick on the top
  • 00:02:33
    cover okay very cool uh and then this is
  • 00:02:36
    a test vehicle I guess yeah so this is
  • 00:02:38
    one of our test Vehicles here uh you can
  • 00:02:40
    see all the thermocouples miraculously
  • 00:02:42
    weaving their way through to each
  • 00:02:44
    component and then of course liquid
  • 00:02:46
    metal thermal interface material with a
  • 00:02:49
    uh triple barrier gasket there to ensure
  • 00:02:52
    it's reliable make sure you can use it
  • 00:02:54
    all orientation shot and Vibe throw it
  • 00:02:56
    around and there's more back there too
  • 00:02:59
    so so there's a lot of stuff before that
  • 00:03:02
    this video is brought to you by height
  • 00:03:03
    and the y70 case the height y70 case has
  • 00:03:06
    a lot of Polish and heavy attention to
  • 00:03:08
    detail on the finer points the case can
  • 00:03:10
    fit radiators that are massive in depth
  • 00:03:12
    to the side mount has Cooling in the
  • 00:03:14
    floor of the case for direct intake to
  • 00:03:16
    the GPU and tries to find a balance
  • 00:03:18
    between structural support of dust
  • 00:03:19
    filters without obstructing too much
  • 00:03:21
    intake the y70 is a follow-up to the
  • 00:03:24
    popular y60 which got height to where it
  • 00:03:26
    is today with its cut Corner tempered
  • 00:03:28
    glass case object for this video really
  • 00:03:30
    is just to talk about how this stuff
  • 00:03:32
    works and some of the design process but
  • 00:03:34
    same thing we did with the 40 series
  • 00:03:35
    except now there's a lot of new topics
  • 00:03:37
    and uh we just took apart that unnamed
  • 00:03:40
    card on the channel recently to me
  • 00:03:42
    there's some stuff here that's familiar
  • 00:03:44
    to that the performance on that card
  • 00:03:45
    I've been telling people at the show
  • 00:03:46
    here is really the only reason that I
  • 00:03:48
    have any faith speaking honestly that a
  • 00:03:50
    two- slot card can do 575 Watts yeah
  • 00:03:54
    because the performance on that
  • 00:03:55
    prototype was so crazy and granted
  • 00:03:57
    that's bigger uh if there's some of the
  • 00:03:59
    same principles applied yeah then I I
  • 00:04:02
    think that's where it gets interesting
  • 00:04:04
    so yeah so the whole architecture kind
  • 00:04:06
    of revolutionary concept this generation
  • 00:04:08
    is dual blow through so both sides of
  • 00:04:11
    the heat sink the air flow directly goes
  • 00:04:13
    through which basically means it doesn't
  • 00:04:16
    have to turn in order to exhaust out of
  • 00:04:17
    the graphics card and what that does is
  • 00:04:20
    it reduces pressure drop increases air
  • 00:04:22
    flow reduces temperature Rec reduces
  • 00:04:24
    Acoustics the really beautiful thing
  • 00:04:27
    about blowr and obviously you've you've
  • 00:04:29
    taken a part our prototype and there's a
  • 00:04:31
    video on that and uh you know when when
  • 00:04:34
    we look at the Prototype the reason it
  • 00:04:36
    had to be four slots wasn't because the
  • 00:04:39
    thermal solution necessarily needed for
  • 00:04:41
    slots when you look at it you can see
  • 00:04:43
    the fins are very wide they're they're
  • 00:04:44
    far apart and um and even though it is
  • 00:04:49
    for slots really thermally you don't
  • 00:04:51
    need all that it was more because the
  • 00:04:52
    PCB you know the GPU plus memory the
  • 00:04:55
    stackup was four slots so if we want to
  • 00:04:58
    rotate the PCB and have what we call
  • 00:05:00
    three thirds blow through okay or 100%
  • 00:05:02
    blowr um okay I see I was trying to
  • 00:05:05
    follow the three3 three3 why 3/3 why not
  • 00:05:09
    four fours is it because three fans this
  • 00:05:11
    was always oneir blow through we always
  • 00:05:13
    call one/ thir blow through so you work
  • 00:05:14
    up from there 2/3 blow through and then
  • 00:05:16
    3/3 blow through um when are you doing
  • 00:05:19
    four3 blow through when are you breaking
  • 00:05:21
    the law can't talk about that one that
  • 00:05:23
    unreleased that's another prototype
  • 00:05:25
    you'll get anyway so so basically when
  • 00:05:28
    when you think about designing the
  • 00:05:29
    thermal solution for any graphics card
  • 00:05:31
    you want to look at each resistance you
  • 00:05:32
    want to minimize each resistance and the
  • 00:05:35
    biggest resistance is always air flow so
  • 00:05:38
    there's two ways to really increase air
  • 00:05:40
    flow and they both both have to do with
  • 00:05:42
    reducing pressure drop so one way is if
  • 00:05:45
    you have a traditional PCB you can make
  • 00:05:47
    the card you know three four slots and
  • 00:05:49
    that gives the airf flow more space for
  • 00:05:51
    it to flow out you know as it changes
  • 00:05:54
    Direction 90 de and the other way is
  • 00:05:57
    with blowr so with blowr the beauty of
  • 00:06:00
    blowr the real peak of blowr is being
  • 00:06:03
    able to make things smaller so in this
  • 00:06:05
    one you can see compared to like uh the
  • 00:06:08
    Prototype compared to 40 series the fin
  • 00:06:10
    pitch is much thinner right and so you
  • 00:06:14
    we only have about 1.5 millim fin pitch
  • 00:06:16
    okay and we can fit a lot of fin surface
  • 00:06:19
    area and a lot of um effective fin area
  • 00:06:23
    underneath the fans so I guess the PCB
  • 00:06:25
    ends up I haven't looked at this that
  • 00:06:26
    closely yet is it here correct yeah
  • 00:06:28
    right in the middle this Ridge where the
  • 00:06:30
    uh PCB is is this um attached to the PCB
  • 00:06:34
    the whole pcie foot yeah so that pcie
  • 00:06:37
    connector oh let me get a different here
  • 00:06:41
    I can bring it back okay so this is the
  • 00:06:43
    PCB here okay and you can see it has uh
  • 00:06:47
    couple interesting things well first of
  • 00:06:49
    all incredibly small incredibly dense
  • 00:06:52
    compared to the 40 series um but also we
  • 00:06:54
    have a couple connectors here so power
  • 00:06:56
    delivery on both sides yeah Power
  • 00:06:57
    delivery on both sides so we've actually
  • 00:06:59
    lowered it into the heat sink that's why
  • 00:07:01
    you can see here there's no actual back
  • 00:07:03
    cover right and this is to help with
  • 00:07:06
    Cooling and allowing for tall components
  • 00:07:08
    on the back okay so the backside power
  • 00:07:10
    components actually increase the overall
  • 00:07:11
    efficiency of the board and it it's
  • 00:07:14
    quite incredible something we've never
  • 00:07:16
    been able to do with the traditional PCB
  • 00:07:19
    level with the back of the heat sink and
  • 00:07:21
    then this is your pcie connector so that
  • 00:07:23
    is we have a board to board connector
  • 00:07:25
    for the pcie and then this connector is
  • 00:07:27
    for the io is this custom uh yes out
  • 00:07:31
    okay or well yeah I guess because it's I
  • 00:07:33
    guess it's just standard pcie but you're
  • 00:07:35
    changing the configur the way attaches
  • 00:07:38
    so exactly exactly um and this one is
  • 00:07:41
    for the I/O here so then there's a flex
  • 00:07:43
    connector connecting the io to the the
  • 00:07:47
    main PCB so in total it's four pcbs that
  • 00:07:51
    allow okay for dual blowr I'm noticing
  • 00:07:53
    this indentation here is there any
  • 00:07:55
    meaning for that yeah yeah so I think
  • 00:07:57
    that's the most visual change this it it
  • 00:08:00
    looks really cool and there is a thermal
  • 00:08:02
    reason behind it so basically um a
  • 00:08:05
    traditional fin stack something like
  • 00:08:07
    this so when the fan is spinning when
  • 00:08:10
    the fan is spinning we have a different
  • 00:08:14
    velocity gradient along the the fan from
  • 00:08:18
    The Hub to the outer diameter so
  • 00:08:21
    basically although the the rotational
  • 00:08:23
    speed is the same the tangential
  • 00:08:25
    velocity is different at each point
  • 00:08:27
    along the the radius so you have your
  • 00:08:29
    most effective fin area or most
  • 00:08:33
    effective kind of velocity and air flow
  • 00:08:36
    at the edge of the fan so what we want
  • 00:08:39
    to do is we want to maximize the amount
  • 00:08:41
    of fin area under that section of the
  • 00:08:45
    fan so basically what we do is we
  • 00:08:47
    decrease the fin pitch between every fin
  • 00:08:50
    so instead of an optimize let's say 1
  • 00:08:53
    millimeter which would be a flat version
  • 00:08:56
    we change it to 1.5 so we increase the
  • 00:08:58
    pressure drop and then what we do is we
  • 00:09:01
    scoop out that fin under the fan so we
  • 00:09:05
    normalize the pressure drop but we
  • 00:09:06
    increase the the area the fin area in
  • 00:09:09
    that most effective region of the fan so
  • 00:09:12
    towards the outer edge of the I'll call
  • 00:09:14
    it the radius or the diameter of the fan
  • 00:09:17
    it's getting taller again and then
  • 00:09:18
    you're scooping it in towards the Hub
  • 00:09:20
    exactly exactly so of course below the
  • 00:09:23
    Hub we have almost no velocity from the
  • 00:09:24
    fan so then as we get to the outer part
  • 00:09:27
    it's the most effective area we want all
  • 00:09:29
    that effective fin area so it's it's
  • 00:09:31
    interesting I guess because um the
  • 00:09:34
    trade-off effectively it sounds like
  • 00:09:35
    you're trading off the total surface
  • 00:09:38
    area because you're losing some height
  • 00:09:40
    but obviously what you're gaining from
  • 00:09:42
    that you think is worth the trade-off of
  • 00:09:44
    the surface area yeah you gain surface
  • 00:09:46
    area in that one in the outer diameter
  • 00:09:49
    area of the fan so that's that makes up
  • 00:09:51
    for it exactly exactly so I mean it's a
  • 00:09:55
    minor change but kind of on the pursuit
  • 00:09:57
    of perfection you know everywhere every
  • 00:10:00
    every aspect of this card has been
  • 00:10:02
    designed and engineer how how granular
  • 00:10:04
    do you get with AB testing and
  • 00:10:06
    development cuz like something like this
  • 00:10:08
    for example um let's just pretend we had
  • 00:10:12
    an easy way to swap only this and change
  • 00:10:14
    nothing else right I
  • 00:10:17
    mean other than computer simulation do
  • 00:10:20
    you guys try to do practical tests of
  • 00:10:22
    individual small things like that yeah
  • 00:10:25
    for sure I think when it comes to
  • 00:10:27
    something like this it's mostly cfd
  • 00:10:29
    driven
  • 00:10:30
    but then we'll test different versions
  • 00:10:31
    with the old pitch and the new fin pitch
  • 00:10:33
    sometimes there is some sample to sample
  • 00:10:34
    variation so you need to be able to
  • 00:10:36
    eliminate that from your testing okay
  • 00:10:39
    but we do a lot of different prototyping
  • 00:10:41
    to make sure kind of where we're at it
  • 00:10:43
    makes sense and correlates with the
  • 00:10:45
    simulation so this is our most you know
  • 00:10:47
    our earliest prototype for dual blowr
  • 00:10:50
    you can see it's just a a block here so
  • 00:10:53
    this is non-functional obviously but it
  • 00:10:55
    goes to test just airf flow and
  • 00:10:57
    Acoustics okay so we want to see what
  • 00:10:59
    kind of acoustic benefit we get so this
  • 00:11:01
    is kind of our earliest prototype it was
  • 00:11:03
    actually three slots to compare it to
  • 00:11:05
    4090 as well and then this kind of
  • 00:11:08
    refines our simulations refines our
  • 00:11:10
    design and then we can you know dig
  • 00:11:13
    deeper and figure out what works so do
  • 00:11:16
    you um what do you mount this to when
  • 00:11:19
    testing it so that one we just mounted
  • 00:11:22
    to a wind tunnel and then for the
  • 00:11:24
    acoustic we did it in the uh the anaco
  • 00:11:27
    chamber that you visited um for all the
  • 00:11:29
    rest of the samples we actually use
  • 00:11:31
    something like this just a metal
  • 00:11:33
    enclosure to simulate it or to test it
  • 00:11:36
    in a actual computer chassis okay so
  • 00:11:39
    this kind of gets more further along the
  • 00:11:41
    development cycle so are you doing
  • 00:11:43
    Thermals yet at this stage yeah this one
  • 00:11:44
    is thermally functional so we're testing
  • 00:11:47
    the thermal performance and we're
  • 00:11:48
    actually comparing all these different
  • 00:11:49
    ones and and these heat sinks are all
  • 00:11:51
    slightly different in terms of the
  • 00:11:53
    development life cycle at Nvidia I mean
  • 00:11:56
    are the PCB guys and the GPU guys even
  • 00:11:58
    ready for you yet to strap something to
  • 00:12:01
    it or like in other words how how
  • 00:12:05
    synchronous or asynchronous are you with
  • 00:12:08
    the team that's going to end up giving
  • 00:12:10
    you a PCB at the end yeah so the PCB
  • 00:12:13
    team kind of enables this entire
  • 00:12:15
    architecture without you know the
  • 00:12:16
    incredible you know this dense PCB and
  • 00:12:19
    all the connections and everything it
  • 00:12:21
    wouldn't be possible to do you know two
  • 00:12:23
    slots 575 Watts so so I guess you're not
  • 00:12:26
    you're not off doing your own thing yeah
  • 00:12:28
    exactly if we're doing our own thing it
  • 00:12:30
    kind of never it never ends up in a
  • 00:12:31
    cohesive project but of course we still
  • 00:12:34
    we don't have necessarily the chip to
  • 00:12:36
    test and all the diags all the software
  • 00:12:38
    so we do use typically like a heater
  • 00:12:41
    block to test these ones but this is all
  • 00:12:43
    kind of concurrently being designed with
  • 00:12:46
    the the PCB and the overall architecture
  • 00:12:49
    I think you have a I do have a heater
  • 00:12:51
    block yeah should I bring it over yeah
  • 00:12:54
    yeah let's do that so what is this that
  • 00:12:57
    you're bringing over so this is kind of
  • 00:12:58
    the iest sample here and we're still
  • 00:13:01
    trying to mimic all the different um you
  • 00:13:04
    know forces from all the Gap pads and
  • 00:13:06
    everything so we have a PCB that ended
  • 00:13:08
    up looking pretty similar close yeah
  • 00:13:11
    quite close but there's a couple couple
  • 00:13:13
    differences um and this will be what we
  • 00:13:16
    use to test these samples here so these
  • 00:13:19
    ones here are going to be tested using
  • 00:13:20
    heater block and then as we get further
  • 00:13:22
    along you can see this one here this is
  • 00:13:24
    one of the does we did um what does that
  • 00:13:27
    stand for basically design of
  • 00:13:29
    experiments so we change very minor
  • 00:13:31
    things especially with the evaporator
  • 00:13:33
    area which we can talk about as well um
  • 00:13:35
    we Chang you know instead of 300 microns
  • 00:13:38
    we make it 250 microns or something like
  • 00:13:40
    that and we'll test a good amount of
  • 00:13:42
    samples to get a good Distribution on
  • 00:13:44
    how one performs compared to another
  • 00:13:46
    okay so um but a lot of the initial
  • 00:13:48
    testing is done with heater block and
  • 00:13:50
    then we can compare that to our
  • 00:13:52
    simulation so so does this end up
  • 00:13:53
    basically attaching to this yeah yeah so
  • 00:13:56
    this should just fit right on there and
  • 00:13:58
    I guess like so got it cool yeah are you
  • 00:14:02
    using thermal pads at this stage this
  • 00:14:04
    one this one should fit okay okay well I
  • 00:14:06
    don't know it fits on one of them it
  • 00:14:07
    fits on one of them do you um when
  • 00:14:09
    you're running that are you also using
  • 00:14:11
    the thermal interface material yeah okay
  • 00:14:13
    yeah um just to make sure all the
  • 00:14:15
    balance all the GPU tilt is correct how
  • 00:14:18
    are you constructing the actual heater
  • 00:14:20
    cuz I I see like those these I'm
  • 00:14:22
    familiar with as like cartridge heaters
  • 00:14:24
    I guess yeah those are cartridge heaters
  • 00:14:25
    we actually have a lot of different
  • 00:14:27
    heater blocks we have some with some
  • 00:14:29
    curvature to mimic the GPU die um here
  • 00:14:32
    you can see all the cartridges in we
  • 00:14:34
    also have some that mimic the GPU heat
  • 00:14:37
    flux so we can get accurate versions of
  • 00:14:40
    uh like what the dry out point is for a
  • 00:14:43
    specific Vapor chamber got it uh how
  • 00:14:46
    about the the next ones down the line
  • 00:14:48
    yes so actually all these are a bit
  • 00:14:50
    different so this one here you can see
  • 00:14:51
    the heat pipes are flattened a little we
  • 00:14:53
    went through different designs and
  • 00:14:54
    iterations the real challenge so this is
  • 00:14:57
    the first ever 3D VC in a consumer
  • 00:14:59
    Graphics product but it's also the first
  • 00:15:01
    ever Wing type 3D VC that I've that I
  • 00:15:04
    know about um in the world so the
  • 00:15:07
    challenge with a wing type 3D VC which
  • 00:15:10
    means the heat pipes are coming out of
  • 00:15:12
    the side of the vapor chamber the
  • 00:15:14
    challenge is connecting those pipes to
  • 00:15:16
    the main part of the vapor chamber so we
  • 00:15:19
    want water to evaporate in the
  • 00:15:21
    evaporator and then travel to all these
  • 00:15:23
    different heat pipes and then condense
  • 00:15:26
    and make their way back and they do so
  • 00:15:29
    with this this braid Wick here so a lot
  • 00:15:32
    of these different prototypes here were
  • 00:15:35
    experimenting with different ways to
  • 00:15:36
    connect it this one particularly you can
  • 00:15:38
    see here we actually have the vapor
  • 00:15:40
    chamber extend the main body of the
  • 00:15:42
    vapor chamber extend all the way out to
  • 00:15:45
    the edges and then we just directly
  • 00:15:47
    connect the heat pipes right in so
  • 00:15:50
    there's a lot of different variations we
  • 00:15:52
    had different size heat pipes d8 D6 and
  • 00:15:55
    on a technicality I guess uh you said
  • 00:15:57
    there might be like one guy in the world
  • 00:15:59
    who would bring this up but
  • 00:16:02
    technicality um these are not if I
  • 00:16:05
    understand it correctly traditional heat
  • 00:16:06
    pipes like in a CPU tower cooler exactly
  • 00:16:09
    so technically these are either risers
  • 00:16:12
    or vapor columns are Ty they're
  • 00:16:15
    typically called if they come out of the
  • 00:16:16
    top um but in this one since it's kind
  • 00:16:19
    the first of It kind you can Vapor Rose
  • 00:16:22
    but since it's you know a heat pipe is
  • 00:16:25
    you know usually one body that um this
  • 00:16:28
    one is obviously chop that one end and
  • 00:16:30
    then we braze it into the side there
  • 00:16:33
    right um so then on principle the way it
  • 00:16:37
    works I guess uh just strictly for the
  • 00:16:40
    wicking is is what I'm interested in
  • 00:16:41
    here but are you running this braid like
  • 00:16:43
    all the way down the pipe yeah okay yeah
  • 00:16:46
    it actually goes all the way to the very
  • 00:16:47
    end there is centered powder in the pipe
  • 00:16:49
    as well but it helps connect all that
  • 00:16:51
    water to come back and flow back to the
  • 00:16:54
    EVAP other principles the same I mean
  • 00:16:55
    hits the hits the condenser Rec
  • 00:16:57
    condenses and is it capillary action
  • 00:17:00
    back to the yeah all cap and we we
  • 00:17:02
    ensure it works in all orientations as
  • 00:17:04
    well so Against Gravity all orientation
  • 00:17:07
    should work got it yeah is there an
  • 00:17:09
    optimal best world you best case
  • 00:17:11
    scenario orientation yeah we always
  • 00:17:13
    optimize for the traditional user case
  • 00:17:16
    of you know horizontal in a in a chassis
  • 00:17:19
    but we also test in all orientations
  • 00:17:21
    both you know reliability testing shock
  • 00:17:23
    and Vibe but also thermal testing is
  • 00:17:26
    there a meaningful difference in
  • 00:17:27
    orientation for user it maybe a degree
  • 00:17:30
    or two something like that yeah I mean
  • 00:17:32
    it depends on exactly which orientation
  • 00:17:34
    you'll you'll uh operate it in sure does
  • 00:17:38
    um does how does dry out I guess come
  • 00:17:41
    into play uh if at all when you're
  • 00:17:45
    designing like the 3D VC yeah I mean so
  • 00:17:48
    basically it's always this push and pull
  • 00:17:50
    game of dry out versus overall thermal
  • 00:17:52
    performance because there's ways you can
  • 00:17:54
    increase the dry out point but typically
  • 00:17:57
    those ways also decrease thermal
  • 00:17:58
    performance so if you make the wick very
  • 00:18:01
    thick it'll have a lot of capillary
  • 00:18:03
    Force it'll push water back but you're
  • 00:18:05
    increasing that resistance so when you
  • 00:18:08
    increase that resistance and that
  • 00:18:09
    thickness you have lower thermal
  • 00:18:11
    performance okay so what we've done here
  • 00:18:14
    is this artery Wick which kind of Acts
  • 00:18:17
    to pull in water in the high heat flux
  • 00:18:19
    regions of the GPU so we have thicker
  • 00:18:22
    wick on those sections and then thinner
  • 00:18:24
    Wick just where we can have extra you
  • 00:18:28
    know thermal performance and and reduce
  • 00:18:31
    the net thermal performance is there
  • 00:18:33
    anything uh special going on with the
  • 00:18:34
    top half here yeah so this we use mesh
  • 00:18:37
    and mesh is typically a little bit lower
  • 00:18:40
    capillary pressure but higher flow rate
  • 00:18:42
    so we want to have high flow rate here
  • 00:18:45
    and then higher pressure here and then
  • 00:18:48
    the braid is like a a highway for water
  • 00:18:51
    to come back really good in one
  • 00:18:53
    dimension but in three dimensions you
  • 00:18:56
    need something like mesh or cented
  • 00:18:57
    powder I don't know if this question
  • 00:18:59
    even makes any sense but what is the
  • 00:19:02
    um I have no concept for how fast water
  • 00:19:08
    condenses and evaporates in a system
  • 00:19:10
    like this I mean like if you follow the
  • 00:19:12
    same whatever droplet of water or
  • 00:19:13
    something yeah yeah you know do you do
  • 00:19:16
    you do you know how that process you can
  • 00:19:18
    actually calculate the mass flow rate
  • 00:19:20
    based on the power and then you know um
  • 00:19:23
    you can just use the uh the heat of
  • 00:19:26
    vaporization to figure it out um but the
  • 00:19:30
    the speed of vapor especially in the
  • 00:19:33
    evaporator area is incredible it's like
  • 00:19:36
    you know 10 to 20 meters per second okay
  • 00:19:39
    all right that's way higher than I
  • 00:19:40
    thought just in the evaporate but
  • 00:19:41
    obviously it it slows down quite a bit
  • 00:19:43
    as soon as it leaves the evaporator that
  • 00:19:45
    makes sense yeah um yeah and there it
  • 00:19:48
    the the science of vapor Chambers is
  • 00:19:50
    incredible you know there's Sonic limits
  • 00:19:52
    like where you start getting
  • 00:19:54
    compressible flow it it gets crazy but
  • 00:19:57
    um and all of those you you need to
  • 00:19:59
    operate in a certain range in order to
  • 00:20:01
    have the optimal um thermal performance
  • 00:20:03
    very cool um anything else on these you
  • 00:20:06
    want to go over before we move to maybe
  • 00:20:09
    liquid metal or something oh I think you
  • 00:20:11
    know um yeah this is basically just the
  • 00:20:13
    evolution of the graphics card as we
  • 00:20:16
    continue to refine it and continue to to
  • 00:20:18
    determine more stuff and then finally we
  • 00:20:20
    end up with you know the final heat
  • 00:20:23
    that's the actual final final that is
  • 00:20:25
    the final final okay dot to I see so
  • 00:20:29
    there's
  • 00:20:31
    uh pedestal yeah it's about 150 microns
  • 00:20:35
    so only about 0.15 millimet pedestal
  • 00:20:37
    what what's the purpose of that um we
  • 00:20:39
    just found it worked better with the
  • 00:20:40
    barrier for the liquid metal okay so
  • 00:20:43
    just to make sure you contact the
  • 00:20:44
    Silicon I guess exactly you want to
  • 00:20:46
    contact the silicon and then push down
  • 00:20:48
    on the barrier got it so that's maybe a
  • 00:20:50
    good jumping point to the barrier so
  • 00:20:52
    yeah so the barrier um this was you know
  • 00:20:56
    something we've been working on for many
  • 00:20:58
    years at this point you know even ere we
  • 00:21:01
    started looking at liquid metal and
  • 00:21:02
    obviously the thermal performance
  • 00:21:04
    numbers are there even at Time Zero but
  • 00:21:07
    the reliability is really the key part
  • 00:21:09
    of liquid metal making sure that it
  • 00:21:12
    works in all
  • 00:21:13
    environments um it works in all
  • 00:21:16
    orientations and you know you need to
  • 00:21:18
    keep it in that GPU die region so that
  • 00:21:21
    doesn't electrically short anything or
  • 00:21:23
    oxidize or or have any issues in that
  • 00:21:26
    regard so what we've came up with is
  • 00:21:29
    this hermetic seal rubber gasket here
  • 00:21:32
    and there was a lot of different trials
  • 00:21:33
    and we had different ideas on how to how
  • 00:21:35
    to address all the concerns with with
  • 00:21:37
    liquid metal yeah and we came up with
  • 00:21:39
    this triple barrier um rubber gasket
  • 00:21:42
    here that basically seals to the heat
  • 00:21:44
    sink when you say triple barrier you
  • 00:21:46
    talking about these ridges where there's
  • 00:21:48
    three of them yeah yeah exactly so
  • 00:21:49
    there's three ridges and with this
  • 00:21:53
    particular barrier we've tested all
  • 00:21:55
    orientation shocking Vibe throwing the
  • 00:21:56
    card around and certainly no leaking no
  • 00:22:00
    degradation of thermal performance did
  • 00:22:01
    you have a loose barrier over here there
  • 00:22:04
    you go thanks so I guess this is what
  • 00:22:07
    you end up sticking on there around the
  • 00:22:09
    GPU is this basically an adhesive like a
  • 00:22:11
    rubber yeah high temperature adhesive
  • 00:22:13
    and high temperature high temperature
  • 00:22:15
    low temperature silicone rubber okay and
  • 00:22:18
    then it covers all the capacitors around
  • 00:22:20
    the GPU as well um no local shorting no
  • 00:22:24
    no PCB shorting um and no oide formation
  • 00:22:30
    right okay um is pitting a concern at
  • 00:22:33
    all I mean if you're contacting
  • 00:22:35
    nickel-plated copper uh are there
  • 00:22:40
    any relevant observable effects to
  • 00:22:44
    that to the Finish yeah yeah so um
  • 00:22:47
    there's none with nickel plating um
  • 00:22:50
    there's there's debate in the industry
  • 00:22:52
    of whether copper is okay bare copper um
  • 00:22:54
    but we went with nickel plating because
  • 00:22:56
    there's no real reaction there be copper
  • 00:22:59
    there is a a visual yeah you know black
  • 00:23:04
    mark So a bit of a teaser for our
  • 00:23:06
    thermal Grizzly tour later but Roman was
  • 00:23:08
    just talking to me about this because uh
  • 00:23:11
    uh he's got some kind of I don't know
  • 00:23:13
    something in the works related to Copper
  • 00:23:15
    specifically as it contacts the metal
  • 00:23:18
    and um if I remember correctly I think
  • 00:23:20
    he was saying that there is observable
  • 00:23:22
    performance drop with time on exposed
  • 00:23:25
    copper yes yeah uh nickel plating you
  • 00:23:27
    know I think what he said aligns with
  • 00:23:29
    what you said but yeah CU it is the
  • 00:23:31
    gallium in there that reacts with the
  • 00:23:33
    copper and the interesting thing is it
  • 00:23:37
    it also dries out the liquid metal
  • 00:23:39
    because you're taking out the Gallant
  • 00:23:40
    right so uh that's why we went with
  • 00:23:43
    nickel plating where there's really no
  • 00:23:45
    reaction what's the application process
  • 00:23:48
    like is is this machine applied at this
  • 00:23:50
    point yeah this is machine applied so we
  • 00:23:53
    can specify the exact amount okay um the
  • 00:23:56
    amount is really important to make sure
  • 00:23:58
    you know nothing spills out obviously
  • 00:24:00
    yeah I guess it's been a while since
  • 00:24:01
    we've made these liquid metal explainers
  • 00:24:03
    but one of the big things we would bring
  • 00:24:05
    up uh back when delting was very common
  • 00:24:09
    was a lot of people go a little too
  • 00:24:10
    heavy on it and you can actually end up
  • 00:24:12
    with worse performance yeah if you make
  • 00:24:13
    it thicker if you make it a lake of
  • 00:24:15
    liquid metal yeah exactly exactly yeah
  • 00:24:18
    yeah so it's a very fine balance between
  • 00:24:21
    you know too little too much but um we
  • 00:24:24
    have really accurate dispensing process
  • 00:24:26
    to make sure we have the exact right
  • 00:24:28
    amount we're talking about endurance you
  • 00:24:30
    know and endurance and then I guess
  • 00:24:32
    orientation being important for you guys
  • 00:24:35
    so uh how do you approach endurance
  • 00:24:39
    evaluation for something like this and
  • 00:24:42
    um yeah I guess what were sort of the
  • 00:24:44
    the conclusions what when you've made
  • 00:24:46
    the final decision for liquid metal what
  • 00:24:48
    were the pros and cons you guys were
  • 00:24:49
    weighing yeah I mean basically this is
  • 00:24:52
    what took us so long to to make sure
  • 00:24:55
    that this was a really reliable solution
  • 00:24:57
    um we went through a lot of different
  • 00:24:59
    prototypes we tested in you know every
  • 00:25:02
    every kind of environment you know high
  • 00:25:05
    humidity low temperature high
  • 00:25:07
    temperature um we cycle it you know
  • 00:25:11
    thermal shock and then of course all the
  • 00:25:13
    shock and Vibe and all that um fun stuff
  • 00:25:17
    we've done kind of all of it we've tried
  • 00:25:19
    different barriers and this is really
  • 00:25:22
    the one that proved to kind of do well
  • 00:25:25
    in every single how does how does the
  • 00:25:26
    difference uh manifest itself I guess
  • 00:25:30
    from your best barrier versus like a
  • 00:25:34
    single walled one I what are the types
  • 00:25:36
    of differences you see actually the the
  • 00:25:38
    triple wall uh it is a bit of an
  • 00:25:41
    Overkill I think um we want to make sure
  • 00:25:43
    you know it's it's not going anywhere so
  • 00:25:46
    uh we actually had a version with two
  • 00:25:48
    and um saw no issues but you know when
  • 00:25:51
    you can go for three why not go for
  • 00:25:52
    three U we did try a couple different
  • 00:25:55
    ones you know um I have some samples
  • 00:25:57
    over there
  • 00:25:58
    uh but you know you start to see if
  • 00:26:01
    there's any kind of hole in the barrier
  • 00:26:04
    or even if it's even if it can hold the
  • 00:26:06
    liquid metal but it can't hold air then
  • 00:26:08
    you you start to see either oxide or you
  • 00:26:11
    know in worst case it could it could
  • 00:26:12
    leak out which is definitely what we
  • 00:26:14
    don't want so that's why we kind of went
  • 00:26:17
    with this triple barrier even when two
  • 00:26:20
    would have been enough right on the PCB
  • 00:26:23
    anything additional to cover I mean
  • 00:26:25
    power delivery we're talking about a
  • 00:26:27
    little bit power deliver is incredible
  • 00:26:28
    actually you know even that it's so
  • 00:26:31
    interd designed with the mechanical and
  • 00:26:33
    thermal design that even the heat pipes
  • 00:26:35
    travel through the spots between the
  • 00:26:37
    inductors that's why the the inductors
  • 00:26:39
    are where they are so if you put that
  • 00:26:41
    cool so they just run right there yeah
  • 00:26:43
    exactly it's it's really a thermal
  • 00:26:46
    mechanical electrical solution um
  • 00:26:49
    obviously a really dense PCB lots of
  • 00:26:53
    memory the chip has gotten bigger more
  • 00:26:55
    memory and yet it's still like half the
  • 00:26:57
    size do you does the PCB layer count
  • 00:27:00
    factor into your equation when you're
  • 00:27:02
    designing the thermal solution yeah that
  • 00:27:05
    and how much copper is in the in the PCB
  • 00:27:10
    okay so we have some ways to simulate
  • 00:27:13
    that and how you know if you add more
  • 00:27:15
    copper below a certain component you'll
  • 00:27:17
    actually reduce that component's
  • 00:27:18
    temperature significantly that makes
  • 00:27:20
    sense so when we're doing our thermal
  • 00:27:21
    solution we can feed that back to the
  • 00:27:24
    the PCB designers and then they can add
  • 00:27:27
    more copper in certain places and then
  • 00:27:29
    how about if we move over to this test
  • 00:27:31
    board so uh this I mean we've done stuff
  • 00:27:35
    uh adjacent to this I don't want to say
  • 00:27:36
    like this because this I for anyone in
  • 00:27:39
    the audience who hasn't run thermocouple
  • 00:27:41
    wiring I find it to be incredibly
  • 00:27:44
    frustrating and I'm sure whoever did
  • 00:27:46
    this must be in like a Zen State when
  • 00:27:49
    they're doing it but you know cuz for me
  • 00:27:52
    uh I guess the things we look at are you
  • 00:27:54
    want to attach it ideally centrally on
  • 00:27:57
    wherever the heat is generated on the
  • 00:27:58
    component y you don't want to disturb
  • 00:28:01
    the flushness of the mount uh you don't
  • 00:28:04
    want to disturb the thermal interface
  • 00:28:05
    too much and uh and then you also have
  • 00:28:09
    this challenge of you got a bunch of
  • 00:28:12
    wires yeah and so this is like very
  • 00:28:15
    difficult to do with this level
  • 00:28:16
    Precision but
  • 00:28:18
    what what types of things do you end up
  • 00:28:21
    deciding to probe I do you just probe
  • 00:28:22
    every component you know yeah actually
  • 00:28:25
    this is one of our earlier tests so for
  • 00:28:27
    the final version will actually have
  • 00:28:28
    about twice as many thermocouples uh
  • 00:28:31
    along the entire PCB but um again the
  • 00:28:35
    things you were mentioning any tilt or
  • 00:28:36
    imbalance the application of
  • 00:28:38
    thermocouples is a is an art you know um
  • 00:28:41
    but how do we look at each component we
  • 00:28:43
    have simulation obviously so we want to
  • 00:28:45
    try and correlate to that simulation and
  • 00:28:47
    also we want to we want to validate that
  • 00:28:51
    you know each component is below right
  • 00:28:53
    given temperature so when we look at it
  • 00:28:56
    we kind of look at areas we want we want
  • 00:28:58
    to look at and then specific components
  • 00:29:00
    and of course in different workloads
  • 00:29:02
    each component is going to have a
  • 00:29:03
    different amount of power going through
  • 00:29:04
    it so we we meticulously look at each
  • 00:29:08
    one and then thermocouple it just to
  • 00:29:11
    make sure that at the end of the day
  • 00:29:12
    every single component on this board not
  • 00:29:14
    just GPU not just the memory is well
  • 00:29:17
    within spec so do you I I know at a
  • 00:29:20
    consumer level even we have uh some
  • 00:29:22
    limited exposure to GPU like GPU
  • 00:29:25
    temperature right hotspot temperature
  • 00:29:28
    and one that they just call memory
  • 00:29:29
    temperature uh and then I I know the
  • 00:29:32
    memory ic's themselves have temperature
  • 00:29:35
    sensing that you can get access to so
  • 00:29:37
    where I'm going with this is do you guys
  • 00:29:40
    um are you able to just rely on the
  • 00:29:43
    software pulling of that and not have to
  • 00:29:45
    probe say the back of the GPU or
  • 00:29:48
    something yeah we we obviously trust it
  • 00:29:52
    but we verify it as well you know
  • 00:29:54
    especially with gd7 we need to make sure
  • 00:29:56
    everything is is good there right and we
  • 00:29:58
    verify it and the readings are
  • 00:30:00
    incredibly accurate you know when you
  • 00:30:02
    compare it obviously we're measuring
  • 00:30:04
    case temperature and there's Junction
  • 00:30:05
    temperature there's a difference there
  • 00:30:07
    but whenever we correlate it back you
  • 00:30:09
    know for G7 or for the GPU die itself
  • 00:30:12
    it's been incredibly accurate and then
  • 00:30:14
    also on you know we can we can look at
  • 00:30:17
    the chip for example through IR Imaging
  • 00:30:20
    just to make sure each each sensor is
  • 00:30:22
    correct right if you do that
  • 00:30:26
    uh h how do you how do you tackle that
  • 00:30:31
    without the problem of the heat sink
  • 00:30:33
    itself is in the way of what I mean do
  • 00:30:34
    you just shoot the back of the board
  • 00:30:36
    there's there's a method I don't know if
  • 00:30:38
    I could tell the method to be honest
  • 00:30:40
    fine there's a method is enough it's
  • 00:30:42
    quite incredible there's a whole there's
  • 00:30:44
    a whole Lab dedicated just to that I'll
  • 00:30:46
    believe you because last time I was
  • 00:30:48
    there uh you were measuring sound with
  • 00:30:50
    lasers that is was not something I knew
  • 00:30:53
    you could do so I'll believe you um I
  • 00:30:56
    let's just walk over here for the last
  • 00:30:58
    couple things you have where should we
  • 00:31:00
    start with this stuff so this one was
  • 00:31:02
    disassembled for you obviously I haven't
  • 00:31:04
    I haven't told you anything about it so
  • 00:31:07
    hopefully uh hopefully it goes goes well
  • 00:31:09
    goes better than the Prototype but this
  • 00:31:11
    I've seen that sticker before Oh I
  • 00:31:13
    wonder I mean I don't wonder where that
  • 00:31:15
    one was I have not seen anyway yeah so
  • 00:31:19
    so you didn't tell me how you took this
  • 00:31:20
    apart which means we'll still do a tear
  • 00:31:22
    down and I'll still figure it out blind
  • 00:31:24
    but um no drill is necessary by the way
  • 00:31:28
    hammers
  • 00:31:30
    but that's up to my discretion each to
  • 00:31:32
    their own each to their own these pads
  • 00:31:35
    and vide use these forever so you must
  • 00:31:38
    like something about them or the team
  • 00:31:40
    what is it like what why these pads
  • 00:31:43
    mostly the reliability so when you know
  • 00:31:46
    the when we heat up the GPU the whole
  • 00:31:48
    board is kind of flexing you know the
  • 00:31:50
    GPU is flexing there's CTE mismatches or
  • 00:31:53
    coefficient of thermal expansion so you
  • 00:31:55
    have all this you know movement inside
  • 00:31:57
    there and we've found that these pads
  • 00:32:00
    specifically are incredibly resilient
  • 00:32:02
    they last a very long time so that's why
  • 00:32:05
    we continue to use them um cool for this
  • 00:32:07
    generation I guess you're trying to
  • 00:32:08
    allow some of that movement yeah yeah
  • 00:32:11
    and we also we did you know last
  • 00:32:14
    generation we reduced it to 1.5 mm Gap
  • 00:32:17
    so we kept that the same this generation
  • 00:32:19
    just to make sure that temperature still
  • 00:32:21
    stay low okay but this is the most
  • 00:32:24
    reliable Tim we have and we ship it on a
  • 00:32:26
    lot of products um um what about the so
  • 00:32:30
    the pcie slot here we saw the other side
  • 00:32:32
    of this over there I guess that's what
  • 00:32:34
    sockets into it and that's a custom
  • 00:32:37
    Solution that's correct and then we have
  • 00:32:39
    this part here that that finally you
  • 00:32:42
    know provides some uh to just clamp it
  • 00:32:46
    yeah to clamp it in with with the screws
  • 00:32:49
    okay and then this part obviously
  • 00:32:51
    connects the main PCB and then the io is
  • 00:32:53
    actually connected here so this is a
  • 00:32:56
    flex so it'll go like that and then plug
  • 00:32:58
    in to the main PCB so you can see that's
  • 00:33:01
    the io and that's the PCB are there any
  • 00:33:03
    special considerations because like IO I
  • 00:33:05
    haven't seen done this way I've always
  • 00:33:07
    seen IO attached to the the rest of the
  • 00:33:10
    card right yeah
  • 00:33:13
    so is is it challenging to run it this
  • 00:33:16
    way I mean is signal Integrity a big
  • 00:33:18
    concern or this is an incredible feat of
  • 00:33:21
    engineering honestly uh the the signal
  • 00:33:23
    Integrity is a concern obviously you
  • 00:33:25
    know you have your mechanical concerns
  • 00:33:27
    but this is you know uh br20
  • 00:33:30
    dp21 incredible that we were able to put
  • 00:33:32
    a connector you know all the specs and
  • 00:33:34
    everything are are made expecting no
  • 00:33:37
    connector so when you completely
  • 00:33:39
    revolutionize the cooling and the PCB
  • 00:33:41
    architecture you have to totally change
  • 00:33:44
    you know everything so a very highspeed
  • 00:33:46
    flex and you know the io over there is
  • 00:33:49
    there anything special to these are they
  • 00:33:51
    just cover plates actually these are are
  • 00:33:53
    a special so basically this is a pure
  • 00:33:55
    aerodynamic change so okay what what we
  • 00:33:58
    have this generation is we have
  • 00:34:00
    incredibly effective fins beneath these
  • 00:34:03
    covers so these ones here you know if
  • 00:34:05
    you look at 40 series or so those were
  • 00:34:07
    structural so they had to be structural
  • 00:34:09
    and basically the fin pitch wasn't
  • 00:34:11
    optimized here it's optimized but we
  • 00:34:15
    cover them with these with these uh with
  • 00:34:18
    these covers that are structural and
  • 00:34:20
    then the beautiful thing is they're
  • 00:34:21
    angled to reduce recirculation so
  • 00:34:24
    basically a pure aerodynamic change to
  • 00:34:27
    redirect the air flow away from the
  • 00:34:29
    inlet of the fan so actually air flow
  • 00:34:31
    does go in under there so then the air
  • 00:34:34
    flow because it's not you know the PCB
  • 00:34:37
    is slightly larger than the fan so then
  • 00:34:40
    there is airf flow here that is then
  • 00:34:42
    redirected those directions and then
  • 00:34:44
    when that air flow exhausts we have you
  • 00:34:48
    know we have these um angled covers
  • 00:34:52
    which then direct air flow away that'll
  • 00:34:54
    be cool we'll try to see ifen can show
  • 00:34:56
    it awesome it' be awesome it's actually
  • 00:35:00
    you know it depends on your case how
  • 00:35:02
    close your your exhaust is to the the
  • 00:35:05
    let's say the glass the glass but the
  • 00:35:08
    benefit can be huge if it's really close
  • 00:35:11
    your benefit is going to be huge from
  • 00:35:12
    that if it's going straight out then it
  • 00:35:14
    kind of just you know if it's further
  • 00:35:16
    far enough away it won't recirculate but
  • 00:35:18
    if the glass is right there it'll go and
  • 00:35:20
    it'll split 50/50 half will go back in
  • 00:35:23
    so the reduction in Inlet temperature to
  • 00:35:25
    the fans is like two or three degrees
  • 00:35:27
    okay incredible actually just from
  • 00:35:29
    adding these changing the angle from
  • 00:35:32
    from like straight that's actually
  • 00:35:33
    really cool this might be one of the
  • 00:35:35
    only pieces we haven't explicitly talked
  • 00:35:37
    about so this is the back plate I guess
  • 00:35:39
    this is the back cover yeah the PCB back
  • 00:35:41
    cover so the the interesting thing here
  • 00:35:44
    is we've increased the fin height on the
  • 00:35:46
    back now of course that is an active fin
  • 00:35:49
    area but it is you know passively
  • 00:35:50
    Cooling and and radiating I feel it's
  • 00:35:53
    got this rubber on it too yeah yeah so
  • 00:35:56
    what's the story with that is that just
  • 00:35:57
    just a soft uh connection basically or
  • 00:36:00
    exactly exactly and to make sure that
  • 00:36:02
    you know the fins aren't damaged during
  • 00:36:04
    assembly okay um and basically this we
  • 00:36:08
    can connect the the mosfets here and
  • 00:36:11
    then we have all this area for cooling
  • 00:36:13
    the backside yeah it's funny I've been
  • 00:36:15
    complaining for years in videos about uh
  • 00:36:18
    back plates that don't make use of the
  • 00:36:20
    surface area yeah so I'm happy to see it
  • 00:36:24
    this is a really incredible one I mean
  • 00:36:25
    it's the thickest one it's the highest
  • 00:36:28
    performance back plate basically because
  • 00:36:31
    the PCB is is lower in there we started
  • 00:36:33
    off this is actually an ampere card and
  • 00:36:35
    we tried different barriers and this
  • 00:36:37
    goes back into the the barrier
  • 00:36:39
    discussion you know we evolved that this
  • 00:36:40
    40 series 40 series 50 Series but we
  • 00:36:43
    wanted to make sure that was the perfect
  • 00:36:45
    solution you know something that was
  • 00:36:47
    really reliable really really met all of
  • 00:36:51
    our criteria so um this is kind of just
  • 00:36:53
    the walking through different prototypes
  • 00:36:55
    similar to how over there we have
  • 00:36:57
    different Heat prototypes um I kind of
  • 00:36:59
    wanted to show the evolution yeah this
  • 00:37:02
    is going to look really cool with the
  • 00:37:03
    first water block that can actually get
  • 00:37:05
    to size oh it's going to be awesome it's
  • 00:37:07
    going actually be a cool display piece
  • 00:37:09
    so cool that is the Fairly uh Deep dive
  • 00:37:14
    at least from a public perspective look
  • 00:37:17
    at the 50 Series Cooling and uh I guess
  • 00:37:20
    on our side we just need to get to
  • 00:37:22
    testing it once you know once that
  • 00:37:24
    embargo lifs so we'll have that data
  • 00:37:26
    soon um Malcolm thank you very much for
  • 00:37:29
    your time and patience a pleasure a
  • 00:37:31
    pleasure yeah 5090 two slots I think you
  • 00:37:35
    know even five years ago if you would
  • 00:37:37
    asked me I would have thought it was
  • 00:37:38
    impossible but you know when you look at
  • 00:37:41
    each resistance along the way optimize
  • 00:37:43
    each one you come up with something
  • 00:37:45
    that's really beautiful so I uh yeah
  • 00:37:48
    it's an incredible card yeah I'm excited
  • 00:37:50
    to test it and then if you want to see
  • 00:37:51
    other videos Malcolm's Ben and I'll link
  • 00:37:53
    them below uh all of them are just as
  • 00:37:57
    interesting even if the products have
  • 00:37:58
    aged a little bit so you should check
  • 00:37:59
    them out but thank you for joining me
  • 00:38:02
    and we will see you all next time
Tags
  • Nvidia
  • RTX 590
  • Conception thermique
  • Chambre à vapeur
  • Métal liquide
  • Ingénierie thermique
  • Refroidissement GPU
  • Technologie 3D
  • Aérodynamique
  • Prototype