Gadgetory

guys build Zoid here and today we're gonna be taking a look at the GTX 20/80 FTW three PCB from EVGA so this is one of the first custom GTX 20/80 PCBs that we're gonna get to well actually one of the first gtx 28 PCBs to look at period and it's it's very impressive but that seems to be the case for like the entire 20 series lineup but even like the reference cards are getting kind of ridiculous and I wonder if it's maybe just because they're expensive and not so much because they actually need it because judging by the Titan V these things still don't scale with voltage so there's not really any reason to ram a ton of power into one of these cores unless you're on liquid nitrogen and this card right here is already excluded from liquid nitrogen use by virtue of this memory power circuit being where it is before that this video is brought to you by us and the gamers Nexus store at store dock gamers Nexus net you can pick up one of our mod mats now on backorder because they keep selling out or you can grab one of our other products like the Raglan 2-tone hoodie which we have in stock or the full gamers Nexus logo t-shirt that we just restocked in cotton and tri-blend learn more at store gamers Nexus net nice PCB it's just I wouldn't like that I wouldn't like you can't use it on liquid nitrogen just because that's where we're like that's right there and that's gonna get in the way of any a liquid nitrogen pot you're gonna try to attach to this and that's kind of a major problem in my opinion so that's your vmm VR I'm up there the card also comes with a bio switch the core is of course this giant strip right here which that vrm is so big that they've actually ended up making the PCB a lot taller which I think they kinda have a lot of empty space here so honestly they could have gone for one of the more like oddball vrm layouts where you know that they like line that put like half the vrm here and then the other half of the arm here and then they could have probably crammed the memory of erm up there somewhere though there are downsides to doing that because essentially you have one half of the vrm doing less work than the other half just by virtue of being further away from the GPU Corp as well as having some extra power loss across the power plane just because of the distance so you know it's like this is better in terms of power delivery it's just worse in terms of making the card a baker without you know really maximizing the usage of the PCB space then again most of these cards are apparently loose like very closely based on the reference PCB which kind of makes sense because this looks like they just took the reference PCB and added four more phases to it and then move the memory circuit so as well as the switch added and put it on dual eight pins so you have that there for power monitoring standard Nvidia you know you have a Texas Instruments I on a 3-2 to one down here and that also means that you can do all of your typical shunt mods with liquid metal stacked shunts soldered on top of shunts long piece of wire soldered on top of shunts which with the long piece of wire soldered on top of shunts you need to actually be kind of careful with how long that wire is if it's too short it's gonna trip safety on the card where it'll think that the there's a malfunction because the power reading is too low which is also the reason you can't just directly short over the shunts if you don't want to mess around with modifying the actual shunts there's a bunch of capacitors around the eye and a three two to one which are used for input for filtering the actual voltage drop across the shunts before it gets into the read into the sense pins of the iodine three two to one and if you want you can apply a resistor across those capacitors typically around the size of ten ohms and that'll give you a say three times increase in power consumption it really depends on how that filtering circuit is actually wired up though but on some of the ten series cards if I remember correctly a 10 ohm resistor would give you a three times increase in maximum power consumption and like it would allow the card to pull three times more power not that the card would actually go and do that because obviously it's not gonna pull power it can't use so you know standard in video right here with the current monitoring setup and that's kind of normal and then you have a bunch of minor VRMs over here so you're gonna have things like the 1.8 volts VPP rail for the gddr5 GD d r 6 I'm so used to saying Gd or 5x4 high-end Nvidia at this point but GD d R 6 which isn't that different from gddr5 X actually like in terms of power requirements it's very very similar it runs on 1.3 5 volts on the VM rail or VDD Q if you want the technical term that the data sheets use 1.8 volts of EPP then there's gonna be I assume this is gonna be pax actually 1.8 volts might be over here and then because the thing is this has a USB type-c connector and that USB type-c connector needs a 5 volt rail and you can't get 5 volts off of this because this is 12 volts and 3.3 only so yeah that that's just something worth considering that there's gonna be a 5 volt regulator somewhere on the card and this might be 1.8 volts or that could be PEX that's also another option that that could be the PAX rail which sits at one volt but all of these are basically minor rails and you don't need to worry about them unless they don't work and if you have a car or one of these cards you basically like they either put out 0 volts or they put out whatever voltage they're supposed to be putting out or your chip is dead so you know that kind of thing you don't need to worry about them that much if you're on a liquid nitrogen it can sometimes help to change the voltage on the PAX rail but even that is just like not not necessary like not always necessary so that kind of covers all the different yorams we have on here now let's get into the major sort of up well into the details of them and this is where the big upgrades start with the 20 series so with the 20 series NVIDIA has apparently gotten UPI semiconductor to make a new voltage controller say hello to the you p90 512 it is a whole plus one over the you p90 511 which you had on the entire 10 series so you p95 11 was the predecessor that was on the 10 series and there's actually a massive difference between these two chips even though the numbers got only gone up by one like I'm honestly surprised that they've not made this a whole new like lineup like a ninety six eleven but now is ninety five twelve and the changes are huge so the ninety five eleven you had a maximum switching frequency of six hundred kilohertz per phase the 9512 goes up to two megahertz per phase this means if you're running a doubling scheme you can actually still cram one megahertz per phase which is more than enough to max out any power stage or really any transistor configuration so that's a nice upgrade they still both output you know there's still both eight phase outputs so configurable up to eight phases for both of them but the U P 9512 adds in the new addition of an sm bus so fun fact about the ninety-five eleven if you have a ten series GPU and you check your V core reading that week we're reading is not actually V core at all its literal like the there's physically no way for those cards to monitor the vrm output voltage because the u p95 eleven doesn't have a digital like digital circuit to report how much voltage it's outputting at all so you don't actually know how much voltage a gtx 10 series card is actually running on you just know what the driver is telling the chip to produce so you don't know how much load line is on top of that you don't know anything really about the output voltage with the 9512 the sm bus is not fully detailed yet but it is supposed to do monitoring functionality so like the RM output current should be possible to monitor vrm operating temperature should be possible to monitor assuming that the arm has temperature sensors built in which certain power stages will have you can also just use temperate like add to temperature sensors of your own if you're the PCB manufacturer it should also support voltage readouts and may or may not support well it's also supposed to make it possible to change configurations of the voltage controller however there's no details on what exactly you'll be able to configure I'm assuming switching frequency will be configurable loadline calibration may be configurable and some other operating parameters for the vrm may be configurable but really we need the like I I really want UPI to publish the full datasheet for this chip because it'll be interesting to see what exactly will be available and I really hope they like publish the full datasheet with the actual register map for this because if they do make the register map public that would mean you would have access to things like switching frequency through software essentially so that would be really really neat functionality to have on these cards and that like makes me excited about this new chip because the 9512 is essentially on par in terms of features at this point with things like the International rectifier 3595 which was basically the voltage controller of choice for cards like the gtx 980ti kingpin edition the lightening of the galaxy all of fame all of those cards were on a high AR 3595 a and at this point the 9512 basically matches that could match that on the the feature set and that would be really really cool if it actually did because well that would mean your reference PCBs would be just that much better to start with so that's the 95 12 you know major major like that's the big different like as far as I'm concerned that's the big difference between the V RMS of a GTX 10 series card and the GTX 20 series this chip like that because it's the it's the brain of the V R M so of course that's a major difference the other interesting difference is is that apparently they these cards are so damn expensive and so profitable to produce that it makes sense to put this lovely new high end voltage controller literally everywhere because this this chip right here is actually controlling the two phase memory of erm which is just like you know yeah let's smack an eight phase voltage controller on a two phase the there's another up4 you p90 512 on this card for the V curve erm because it's actually single output like it goes up to eight phases but it's single output there's some other chips where you can go like eight phases and they'll support things like 6 + 2 5 + 3 4 + 4 7 + 1 but the 9512 is single output so it's you know anything from one to eight phases but only one output so if you're not using half of the phases well they're just kind of don't get used to do anything the other ninety five twelve is roughly in this area of the PCB I unfortunately do not have a full board like full PCB shot off to the back of the card and that one's for controlling the V curve erm so somewhere in that area we have another one and that one is running in six phase mode and this V R M as much as it wants to look like a twelve phase is not a twelve phase there is certainly twelve you know twelve sets of inductors and twelve sets of power stages but there is no doublers on this and the 9512 can't control twelve phases it only goes up to eight so what you essentially have is that you have groups of 2 power stages and two inductors at the same time that are basically synchronized and since they're not out of phase with each other it's not fair to call them sexpert phases alright because the whole point of phases is is that it's the state of a signal relative to another signal and if you have two says if you have two phases on us it like if you have two phase power that means you have to offset power signals basically well power outputs that are offset from each other by you know phase separation and you don't have that here for 12 phases you only have six so this is a basically a ginormous six phase when I say ginormous I mean ginormous because same as the GTA as the Titan V I keep forgetting that they were if that thing is not a GTX Titan but the Titan V uses a very nice 70 amp power stage from Fairchild Semiconductor called the F DMF 31 70 the 20 series cards at least the reference PCB for the GTX 20 80 and the reference PC VG for the gtx 28 ET i as well as this PCB here for the FTW three card all use this power stage and there's 12 of them in parallel so this is ridiculous overkill um this this is really really really really be powerful as far as GPU VRMs go honestly this is not this this is comparable to some of the top-of-the-line cards on the like top-of-the-line GTX 10 ATT eyes now I do wonder if this is actually necessary because as far as as far as I know 12 nanometer TSMC still doesn't scale with voltage so if you're on our air cooling or water cooling you're never actually gonna get anywhere near using the capability of this VR M but you know there's nothing wrong with having way more viewers at least if the half of the phases failed you'd still have more of erm than you need redundancy yeah which actually I do think the FD MF 31 seventies these are like smart power stages if I remember correctly these actually do support like drop it like if they've malfunction they're just gonna drop out and the VR M can keep running like nothing happened so that's you know that's kind of neat you basically have a nice redundant six phase vrm as a end result now then in terms of actual heat output figures for this VR M it gets a bit complicated because as much as I like the FBM F 30 170 the data sheets for this thing respect entirely a 1.8 volts out 502 kilohertz switching frequency which the 500 killer switching frequency is not a problem the 1.8 volts output is because that's gonna drive up the heat output of each power stage for any given current level so if you're pushing you know say 50 amps at 1.8 volts it's gonna produce more heat than if you're pushing 50 amps at 1.2 volts out so that means that all of the heat output figures I'm gonna give you right now are a bit high or maybe very high I don't know because the datasheet doesn't say how these scale with output voltage they do scale with input well like there is a graph for scaling with input voltages like not exactly useful international like there's a lot of other power stages which actually give you details for scaling with input ball with output voltage and the differences are quite significant most of the time but it also depends on the internal design of the stage you could theoretically make one that doesn't really care how much output voltage you have or that would actively get worse as you lower the output voltage if you had a really weak low side MOSFET and a really strong high side then you'd slowly run into an issue of the since as as you out lower the output voltage more and more of the current goes through the low side MOSFET and if the low side most that sucks well then your power your heat output is gonna go up and up and up as as you lower the output voltage so you know the the are basically the vrm heat output figures here are gonna be off by some amount in some direction I'm not sure probably upwards I think they're probably a bit too high but unless you know I get a beta better datasheet detailing how these actually scale with output voltage I can't really do anything about this so let's go through the various current figures so I don't know how much these cards are actually going to pull Sorge is gonna go in increments of a hundred starting in two hundred amps because that seems like a like reasonably close to where this probably runs and I'm assuming these probably run on around a hundred and eighty amps at one volt but I'm not sure so two hundred amps output you're gonna be looking at about 21 watts of heat output three hundred amps output you're gonna be looking at about 35 watts of heat output four hundred amps output you're gonna be looking at about 50 watts of heat output now this like all of this is still very cool about and especially if you were at a lower output voltage in the the heat outputs go down like this vrm should have no problem and Allah 400 amps if you just throw enough airflow at it and reasonably sized heatsink so I really wonder what's gonna happen with the 28 e TI v RMS for like the extreme overclocking parts like the galaxy all of fame or the kingpin addition I'd really love to know what they're gonna do if this is what they're putting on an FTW 328 e cuz that's the main thing like this chip is much much smaller than the 28 e TI one so you know one could one would assume that the 28 e TI is gonna pull more like a lot more power than a 20 80 once you you know take the chains off so to speak 500 amps output this is where the heat output starts getting like pretty significant but again keep in mind 1.8 volts out so it could be it would probably be lower if the output voltage wasn't so high 500 amps output you'd be looking at about 72 watts of heat and it's 600 amps output you'd be looking at a little over a hundred watts at 102 watts of heat so realistically anything up to 400 amps should be perfectly doable for this vrm assuming you know you still have enough cooling if you don't have enough cooling you're probably gonna be limited to around the 200 amp range but like don't have any active airflow you're probably gonna end up around the 200 amp range about otherwise up to 400 to be totally doable I really do want to know what they're gonna do for the 2080 TI's like because this is ridiculous so yeah really really impressive six-phase but at this point like after the Titan V I'm not sure that I'm surprised anymore because the Titan we had basically like same power stages massive eight phase where there was two power stages per phase two inductors per phase so it just kind of looks like an Nvidia has just decided you know what everybody gets 70 M power stages and they can just build whatever they want out of them package deal you buy a you buy a t you die and you get yourself a bunch of 79 power stages to use though use with it and so yeah it'll be interesting to see what the other manufacturers come up with because this is really really impressive and massive massive overkill now then for the memory vrm they're still using those 70 M power stages which just is like it's insane it's a memory of erm this thing's never gonna push more than like 20 amps if it's even gonna push 20 M because GDD are six is supposed to pull less power than gddr5 so a little bit less power or I think them maybe the same amount of power is gddr5 X for the same amount of you know bandwidth so on maybe even a higher bandwidth so otherwise the efficiency wouldn't really go up would it if it pulled more power at higher frequencies so yeah and even gddr5 which is more power hungry than GDD are six or gddr5 x really doesn't pull that much current either so this is ridiculously insane overkill and that straight-up doesn't need a heatsink at 20 amps output that vrm would generate about 2 watts of heat and I don't think it would actually ever have to push 20 amps and that's again assuming 1.8 volts you know so at one point 3 5 volts that would very likely be well under 2 watts and this thing is huge if you just have some moderate airflow over it you should have no problem cooling itself and if you went all the way up to 40 amps which I really don't think G DDR 6 is ever gonna pull even if you started over vaulting it and overclocking it to the moon it'll produce about 4 watts of heat we just still nothing for such a you know lot like it's a two-phase it's huge there's a lot of surface area on the basis of just being two phases so basically this is this this is really really impressive that's all there really is to say about this PCB is this is really really impressive it's just so unfortunate that they stuck that memory of erm right where they did because there's no way you're ever gonna run this on liquid nitrogen because you need to move those inductors I assume they do not clear lnto pulse because they're probably too damn tall yeah that makes it like this this thing makes me pretty excited to see the rest of the twenty eighty twenty eighty PCBs because if this is what you know EVGA is managed to come up with for the FT w3 I mean maybe all the other manufacturers decide to just like cut costs because it's like you don't need this you're just gonna get a six phase or maybe they'll just all roll with reference PCBs for their cards because the reference PCB as far as I'm aware is just a straight-up eight phase which already with this 70 M power stage the like freaking eight phase is already ridiculous for you know for your typical air-cooled or even water-cooled overclocking so yeah very very impressive from EVGA here it'll be interesting but it's just like it's ridiculous like it's so impressive it's ridiculous at this point because this honestly puts to shame a loss of extreme overclocking cards from from the past and you know especially we're now with the upgrade of the you p90 512 of the upgrades to the you p90 512 like you may not need a kingpin Edition 20 atti to break World Records at this point like it's well no you you might want like the main difference between like what I expect to happen is like the main difference between a kingpin Edition 20 atti and like a lightning 20 80 TI and a hall-of-fame 20 TTI is that they have better software volt like they have full software controls for the VRMs assuming Nvidia allows those to go public but there's gonna be internal tools for that kind of capability for those cards whereas other cards won't have that and that's the only difference I can really think of that that's gonna occur because as far as the actual component tree goes like this is more than enough that this is plenty like I said you couldn't you don't need more so yeah I'm really well I like I'm really interested to see what the rest of the cards look like at this point because you know for for the first time in ages it's like improving over the the reference design from Nvidia like this is still a step up from the reference design but I think at this point it's really I'm like unnecessary so yeah it'll be just very interesting to see the rest of the cards in my opinion because we might see some cards which actually opt to just down cost the vrm to try drive down the price point because honestly I think what's coming out at the the price point that the founders additions are at this is massive overkill like you you don't need this with all probability but it might be too early to judge who knows maybe TSM sees his new 12 nanometer process scales to 1.5 volts and the card pulls 800 watts on air cooling that would be uncool below 500 watts on air cooling that would be cool so and at that point this VR I might make a modicum of sense but otherwise it's just like well even then 1.5 volts for 500 watts you'd be looking at like little over 300 amps output which would still be very you know very reasonable in terms of the heat so yeah that's it for the video I mean there's nothing left to say this is insane thank you for watching if you like share subscribe leave a comment any comments or questions suggestions down in the comments section below if you'd like to support gamers Nexus here then we have a patreon there's also a store which I think goes stored gamers Nexus dotnet but either way you can find a links links to both of those things down below in the description or the comments section and also if you'd like to check out more extreme overclocking content like this I have a channel called actually hardcore overclocking where I do a lot of stuff like this so yeah that's it for the video and goodbye