Gadgetory

guys build Zoid here and today we're going to be taking a look at the RT X xx atti founders edition PCB from Nvidia and we're just gonna dive right into it because there's no point in talking about sort of stuff ahead of time that I'm gonna address anyway so then let's start off with the V RMS starting with the largest and most important one first the V Corps vrm which is actually split into two separate parts before that this video is brought to you by thermaltake and the view 71 enclosure the view 71 is a full tower case that's capable of fitting three video cards and most configurations it's also one of the better cooling cases in our recent case testing bench lineup the view 71 has hinged a tempered glass doors on either side that make it easy to open and show off and it comes with at least one rain fan though you can get the RGB version if you prefer learn more at the link in the description below so you have one half well it's not really a half because they're not the same size but you have sort of one group of v core phases over there and you have the other group of ecore phases over here above that above the larger v course the core vrm you have the memory VR m so v mm or v GD DDR or whatever you want to call it that powers the GDD r6 memory chips scatter around the die up here you have what is probably the 1.8 volts rail this rail is necessary for the GDD r6 memory chips as well as Nvidia's BIOS chips they they run off of 1.8 volts as well down here I assume this is for USB C power because there is a USB type-c port in the i/o section of the GPU over in this area there should be a PEX rail though the pecs rail might also be here I'm not sure it's it's hard to say without the card in hand so the pecs rail is going to be located in this area and there's also going to be a 5 volts v RM in that area as well and the 5 volts is for powering the actual power stages because those will not run off of 12 volts or 3.3 volts and the PCIe slot only provides 12 and 3.3 volts to the GPU so it's necessary for the GPU to generate its own 5 volts rail and that's gonna be happening somewhere over there again I have a pictures of the card not the actual card in hand so hard to say which regulator exactly is doing that now then let's get into the details on the V curve erm starting with the funky layout well it's not really that funky but it is a bit special we've you know there's the Titan v has a similar layout but most cards generally you just have one strip of phases and they don't do this kind of to separate lines layout and there's actually kind of two major benefits to doing this layout so first of all it spreads the thermal thermal load of the vrm across a much larger area so instead of having you know all 13 phases crammed into one line or into let's say a rectangle of that size on this side of the card and having all of the heat concentrated in sort of this area you have the heat of the VRMs sort of you know spreading out over those sort of areas and so thermal that makes this slightly easier to cool so that's that's one benefit and electrically you get the benefit that with your standard vrm layout when you have a vrm entirely located on like one side of the card like if we just ignored this one's existence issue with most cards that use that normal layout is that you actually get voltage drop at the very high currents just from the power planes electrical resistance and that is so that that like at the really high currents you're gonna be looking at voltage drops where basically if you measure the capacitor on the back of the GPU core so if we go here right so on a lot of cards if you measure the voltage onlet's i'm not sure that this capacitor is v core but I assume it's v core so assuming this one's v core if you measured the voltage on this one and then measured the voltage on say this one you would actually see a difference where this one would measure a couple millivolt slower and depending on how much current you're pushing that difference might be say twenty twenty millivolts or or even more in some cases so there's definitely so with the layout that in V that's actually not as much of an issue because they're basically shoving voltage into the GPU core from both sides and so the area of the GPU core where your voltage will be lower should be this line of capacitors right down the middle and so in theory that gives you a slightly more even supply of power to the GPU core and could lead to a better overclocking results compared to the RM layouts where you just have one strip of phases on one side another reason why this may be done is just because of routing the power planes through the PCB necessitates this layout to not well there there have definitely been cards where you've seen like two rows of vrm phases but that just like massively exaggerated that voltage drop issue over over distance so this layout definitely has some benefits and in that you basically get the voltage regulator as close to the GPU core as possible for all of the phases so that covers that look the oddball layout is actually like that that's a really cool thing to see and I think going forward for really large GPUs we're gonna see this kind of layout more and more often just because you need so much current and so many phases that this is the most reasonable layout to power them you can sort of see Vega doing the same kind of thing with the l-shaped vrm that it has that's similar reasoning there they wanted 12 phases and there's no and if you put the phases in a line then the power delivery from the phases to the actual GPU core well you you take some power losses over the distance that you need to push the current so yeah that's it's a pretty neat vrm layout we have here now in terms of actual phase count this thing is an absolute like this is a mess as far as I'm concerned because this has one two okay let's not count the chokes one two three four five six seven eight nine ten eleven twelve thirteen power stages and thirteen chokes for them and the problem with this is the voltage controller on this card which is for the vcore VRM is this chip right over here and that is a monolithic power systems MP 2888 and that's a ten phase so I don't actually and the thing is so right off the bat you might be like okay so it might be like a six phase getting doubled but there's no doublers on here and nvidia like the the thing is they have a lot of very fancy 12 volt power rail balancing circuitry on here so I'm actually so I have the sneaking suspicion that they do actually have all ten phases hooked up I just have no idea how and there's like three and I'm assuming that there's like three option phases that are being used for changing power balance situations for the the eight pins making sure that both eight pins are pulling an equal amount of power is taken care of by the sort of extra three phases and I'm not sure how that's hooked up what's definitely certain is this card can't have 13 phases interleaving the voltage controller definitely does not support that it's it only goes up to ten phase output but still that is a very like that is a very high end voltage controller right there from monolithic power systems it supports up to five megahertz switching frequency the power stages in video is using only go really or specs like you can run them beyond that but it's not recommended because their efficiency will go down the drain generally that you wouldn't want to run them above one megahertz so this is a very high-end you know voltage controller and actually recently and up until recently really getting a ten phase voltage controller wasn't really possible whatsoever I think around 2000 wellnot around the five 5870 HD Radeon series UPI semiconductor was actually making a 12 phase controller but those all stopped being like nobody that thing Goll like that's things stopped being produced and since then it's basically been only eight phase voltage controllers for high-end applications and more recently with basically between Intel skylake Zeon's which pull an insane amount of power and in India Nvidia's ginormous new GPUs on 12 nanometer we like the 10 phase voltage controller seems to me coming back because this is not the only one that I'm aware of there's also one from Infineon that also exists so yeah this this is a this is a high-end part that's nice to see has a nice square see interfaces uses PWM vid as all NVIDIA voltage controllers in recent history have I am there's no public datasheet for the chip unfortunately so modifying this thing is going to be a pain but anyway so that's the voltage controller and the end result is that I actually don't know how the phase phases are set up I think there's ten phases being interleaved and then there's like three phases that come and go depending on what kind of power balancing situation you're in also they're probably going to be dropping in and out of being run just based on how much for for like for power capability reasons as well but I am really like it's really hard to like I can't give you specifics because it's just like this is a really oddball vrm layout now then for the actual power stages where we're looking at and videos favorite new power stage the Fairchild Semiconductor FDM f31 70 power stage this is a 70 amp part there's actually a bunch of these new 70 amp power stages coming out right now this is the Fairchild Semiconductor one and it's actually a smart power stage so it integrates things like thermal protections over current protection current monitoring temperature monitoring all of that's integrated directly into the power stage and that is then you know read back by the by compatible voltage controllers that's that's like the special thing is like all of these extra features actually need support from the voltage controller itself and yeah these things are incredibly powerful the only issue with them is that the entire datasheet inspected at 1.8 volts output so for however so for actual vrm efficiency figures I'm gonna list both what they would produce like heat what they produce at 1.8 volts output as well as 1.2 volt and then for 1.2 volts output which is still above what Nvidia is gonna be running these cards on stock voltage is going to be around 1 volt well for 1.2 volts output I'm gonna be scaling them down based on the International rectifier 3575 power stage just because most power stages scale them very similarly with lower voltages in terms of efficiency so the it like the the lower voltage efficiency figures are like they're very rough estimates so yeah keep that in mind anyway so power efficiency for the GPU core 1.8 volts 1.2 volts and worth noting is that the vrm like these power strategies also aren't specs for switching frequency lower than 500 kilohertz they are very efficient actually even at 500 kilohertz so these are these are really awesome power stages and I hope we get to see them used in more and more applications and there's already some motherboards using these hopefully they start gaining popularity especially since Nvidia is like probably requiring the use of 7tm power stages for all of their for some of their twenty series cards it might be possible that any board part like any adding board partners for Nvidia that also make motherboards might start using these on motherboards just because they're already gonna be buying them just to produce NVIDIA GPUs so then let's start at what I assume stock current is gonna be stock we're gonna probably be looking at around something around 200 amps at 1.8 volts that would translate to about 20 watt so he output at 1.2 volts you'd be looking at more like 16 watts of e heat output so that is very efficient yeah that is yeah I mean what did you expect it's 13 phases worth of 70 M power stages assuming NVIDIA actually has the efficiency shaping stuff set up correctly these things are going to be absolutely great thermal results across basically the entire voltage range then for overclocking I'm not sure that end video is gonna give you enough power limit to go this high but 300 amps you're gonna be looking at about 34 watts at 1.8 volts and probably about 27 watts at 1.2 volts so this is where I think air cooled and probably water-cooled overclocking will also top out then 400 amps this is ice I'd say starting to get into a sort of dry ice overclocking territory for these cards which would also go up to say 500 amps water cooling may actually overlap somewhat into the 400 amp range I'm not sure really depends on how much power limit Nvidia gives these cards so 400 amps you'd be looking at about 50 watts of heat at 1.8 volts output and about 50 wait no reading the wrong page 40 watts of heat at 1.2 volts output so that's only like you know that's only like 18 watts of heat on like this VR I'm at 1.2 volts and 22 watts on this VR M so that should be like more than kula bewitch just already tells you how ridiculous this VR M is like 400 amps and you know 40 watts and spread across this much surface area this thing should run ice-cold even if you didn't have heat like as long as it gets airflow this thing should run an ice cold it wouldn't really even need heat sinks and it's stock settings this is massive overkill so yeah very very impressive but it in theory this should go higher so for 500 amps which would I think be around we're dry ice overclocking would top out you'd be looking at about 68 watts of heat on 1.8 volts and 55 watts of heat on 1.2 volts now then for ln2 overclocking I must I might the the current estimates here are guesswork based on how 1080 tea eyes behave and then how much larger the 2080 TI is than a 1080 TI and the reason why I'm estimating like this off of the 1080 Ti is because the 12 nano meter man factoring process from TSMC doesn't really make much difference compared to the 60 nanometers manufacturing process from TSM see it's the same density it's slightly more efficient it might have more leakage I not really you know it might have more leakage it might have slightly less leakage it's probably very very similar to the the 60 nanometer process because it doesn't have a really really doesn't have a density increase it does seem to have slightly better slightly better voltage frequency curves but it's not a huge jump so I'm assuming that an estimate based off of a 10 80 TI should be pretty accurate so for ln2 you'd be looking at maybe like 600 amps and I'm assuming around 1.5 volts it might not scale that high at 600 amps this vrm would produce about 93 watts of heat so at this point we're really kind of getting out of the optimum range for the efficiency curve as you can see the current draw jumps are getting bigger and bigger and 1.2 volts you'd be looking more at like 75 watts of heat so you know still should be very hand manageable if you have enough air flow over the VRMs and then 700 amps again on ln2 and this should actually be possible to hit if this behaves like a 1080 TI you'd be looking at about a hundred 25 watts at 1.8 volts and about a hundred watts at 1.2 volts so since on ln2 you'd actually be sort of around 1.5 volts maybe one point four volts it would actually be somewhere in between the these two figures for the vrm heat output and that would get very very hot but if you had like a you know high enough RPM delta fan and maybe some kind of heatsink sitting on top of the vrm it on top of the MOSFETs it should still be manageable so especially since you're just gonna be like it's just gonna need to take that amount of a current for the duration of a benchmark it's not like you're gonna be running back current into it for hours and hours and hours which would mean you'd need a much more substantial cooling system so yeah this VRM is incredibly it's really impressive that is for sure and it is definitely in fact this is the most powerful vrm nvidia has ever put on any consumer GPU this vrm here is more powerful than the VRA than the 2v core VRMs on a GTX 590 this is more powerful than the the two VR at V curvy arms of a GTX Titan Z the only the arm that Nvidia makes that is more powerful than this is the one used on there like volta 100 series cards and I think the Pascal 100 cards as well because those are like compute accelerators and those are absolutely massive and those are the only cards that I can think of that I think how well no the Titan V is the own load that I know for sure hasn't more powerful er the RM and the Tesla's based the Tesla's use the same PCB as the Titan V but as far as consumer cards go NVIDIA hasn't made one with a more powerful vrm than this and honestly like I welcome this because this is a founders Edition where it's like you you know this would actually be the bit this is one of those cards where like the number of power connectors is actually a bigger problem than the the vrm itself so yeah this this is a very very impressive vrm and in fact if I was taking this on Alan to I'd definitely be hooking up more power connectors because this in this is a huge die if it scales to the same voltages and clock speeds that you know the 10 series did this thing is gonna pull an insane amount of power but at least this time around like the the the vrm is definitely ready for that so yeah very very impressive PCB from Nvidia and this is this is one of the few I don't have anything to complain about on well actually already with the 10 series and video really stepped up their PCV quality and this is just like at this point it's like the board partners are gonna have a hard time designing anything significantly better than the reference cards and honestly if you're like considering water cooling or something then I would just get a founders Edition I wouldn't worried about getting a custom card whatsoever because I the there's nothing I can really see here that it be like yeah there's gonna be a significant upgrade if you got a non founders edition that's just that's just not gonna happen the this is already so freakin up there in terms of erm build quality that you know we might like we will be seeing a custom board point like custom cards from board partners with 16 phase of ecore VRMs but it's like about it's not really gonna make a difference for your regular overclocking and it might have some improvement on ln2 but even there it really might not because this this this is a 10 phase in terms of interleaving and it is crazy powerful crazy powerful so yeah I'm very impressed with this very very nice vrm from Nvidia right here now then moving on to the memory of erm this is just more of the same same power stages again the FDM f31 70s from Fairchild Semiconductor and so these are massive overkill and this is a three-phase and it is controlled by this chip over here which is yet another and monolithic power systems MP 2888 which in my opinion this is well that's an older choice because NVIDIA has other voltage controllers that they could have used right there but yeah it's you know there's nothing wrong with using the most high end voltage controller compatible with your cards for basically everything so the memory BRM is absolutely massive overkill and from reading some micron GD dr six product briefs and just information from micron on genie dr six the memory system on a twenty eight ET i should hold around well the the memory chips themselves should pull around 30 watts of power which really isn't that much especially considering that you have like you know 210 amps of peak output maximum output capability on the memory of erm so the memory of Yoram is ridiculous overkill really this this vrm should stay in the single digit heat output because at 20 amps output you're gonna be looking at around 2 watts of heat and at 40 amps output you're gonna be looking at about 3.5 and the gddr5 and you know the GDD r6 is not actually going to pull 40 amps though it might scale with voltage if it behaves like gddr5 x like gddr5 x scaled all the way up to like 1.5 volts on if you had it on liquid nitrogen on air cooling you could reasonably push it to like 1.4 to ish without too much risk in theory and it would scale to that but is still memory system memory chips don't tend to massively increase in power consumption even as you raise the voltage and operating frequency which is kind of weird about them but it's like I've checked it and it really doesn't like you you can crank up the memory frequency on a lot of say gddr5 by say 50 percent and it only makes like a 20 percent difference in power consumption so memory is kind of weird like that and the end result is that the memory of your arm here is ridiculous overkill and at stock these chips will be running at one point three five volts so yeah that's you know but more nice vrm right there and now we get to the sort of the cool stuff while interesting things that NVIDIA has done around the card to try help with overclocking so there's these two buck converter looking circuits they're not actual buck converters they're actually power balancing systems and these are these are what I think is Nvidia fixing a problem that Nvidia very much created so the problem that well I'm not sure how much of a problem it is I never really tested attended like a stock 10 Series card that much but Nvidia runs a per power connector power limit system and if we think about this the 1020 atti has a TDP from Nvidia of 250 watts and in videos TDP very much means that because of how Nvidia monitors their GPUs power consumption when NVIDIA says TDP is 250 watts that means the card will not pull more than 250 watts and the reason for that is they have shunt resistors monitoring power going into the card on every single input and so I assume this one right here is for one of the eight pins I think if I had the card in hand I could verify that but from the pictures it just kind of looks like that one probably is an 8 pin power connector a pin monitoring shunt and this one I'm gonna assume is also for an 8 pin and this one I think is for the PCIe slot so they basically monitor the power power going from you know every single 12 volt source on the card so the 2 8 pins as well as the PCIe slot and the thing is in video when they impose their power limits they might do something like you know 40 Watts on the slot and then a hundred and twenty now that's too much 40 okay you know what let's just simplify this they might go like slot 50 Watts and then the eight pins would be a hundred watts each and then when you raise the power limit which in theory like I'm not sure that they'll even let you set it to 375 Watts that would be like a lot normally the Nvidia power slider goes to like what plus ten percent maybe but let's say they did let you go to 375 Watts so then you'd be looking at like 75 watts on the slot and then 150 watts for each of the power connectors and the thing is since you have this power limit Kerr connector it is actually possible for the GPU to be pulling 375 watts in a scenario where it's pulling 375 watts and it's trying to pull say 200 Watts from one of the power connectors and on a 10 series card that would actually be something that could happen because of the more basic power balancing system that they had and basically when that would happen the card were throttled because it would bounce off the power limit for the eight pin that it's over drawing so to negate some of that Nvidia has per you has a new power 12 volt balancing system which looks like that and that and there's some other MOSFETs scattered around the card that I assume also have something to do with it because there's a lot of these 12 volt balancing control chips actually on the card so the idea behind the 12 volt balance is basically if they if the card starts going out of spec on one of the power connectors the circuitry here can actually switch which power connector is providing power switch some of the phases from one of the eight pins to the other eight pin in order to get them back into balance and that would prevent the card from bouncing off of the per connector power limit before it hits the total GPU power limit so you basically kind of get a more and basically get more power more usable power limit than on a 10 series card the thing is if Nvidia just at least from my perspective is an extreme overclocker see this 8 pin can perfectly fine handle 300 watts and this one can too and so as far as I'm concerned the power limit for this GPU should be more like 600 Watts when overclocking and the per connector power limit it should be like you know well let's say 650 none of the phases should be hacked like the vcore phrases should just not be hang off hanging off of the PCIe slot and so 50 watts would be for the PCIe slot and then 300 Watts for each of the power connectors and NVIDIA as far as I know has never done this just because they technically you're violating the PCIe spec but if you actually look up the specification for these 8 pin power connectors they can take 300 Watts just fine that's not a problem the only issue that would you could run into is if somebody's running like a daisy chain power connector yeah they might overlook they would overload that daisy chain because you'd be pulling like 600 Watts through it but I think honestly Nvidia could just have a disclaimer like hey if you want to raise the power limit this much you need to rewire your GPU or you run the risk of melting your power connect melting your actual cables which would then potentially lead to a fire if you have a crappy power supply without working over current protection and they do make those so yeah it's just like it's cool that Nvidia you know is freeing up more of the power limit by making sure that you don't bounce off of the per connector power limit as much as you used to be as much as it used to happen on the 10 series I'm still not like I still think they should have just given you a higher power limit anyway for the 12 volt balancing they're actually using a UPC I semiconductors you p7 6 v 1q and the other complaint I have for this and chip is that it doesn't have a public data sheet but the description very clearly says that this thing has over current protection so in theory you might run into a situation where like you say you modify the the shunt the shunt resistors by either stacking of more shunts on top of them or just completely replacing them and then you'd still hit the power limit of the 12 volt balancing circuits instead of the the shunts so yeah and also speaking of the shunts Nvidia has changed the chips it uses for the monitoring those they are now instead of using the Texas Instruments ia32 to one they have the current monitoring done by NCP 45 for nine ones and these are actually for shunt for shunt monitoring like they can monitor for shunts each and these don't actually output a nice square so these don't have a nice Gracey interface or anything like that they spit out a differential signal that then goes into a analog digital controller of some kind that you know Nvidia basically chooses it might be fed right back into the GPU core I'm not sure but basically this converts the high voltage the high voltages it measures across the shunts into low voltages that can be read back by you know low voltage analog digital converters like what you would have say on the GPU core or on some other chip sitting somewhere on the PCB and these are much faster than the eye on a 3-2 to 1 and those are also part of the you know rebalance like the upgrade to the balancing circuitry and the power monitoring circuitry from Nvidia because on the previous cards you know you would just basically have like a % TDP indicator now you're gonna have an actual power draw in indication because of the the changes they've made to the power monitoring system and also the 12 volt balancing is now faster thanks to these the problem I have with these is that these have a bunch of tiny little SMDs so modifying these is a bigger pain than modifying the ia32 2 ones for higher power limits so on these cards I would honestly recommend that if you want to lift the power limit and you're gonna be using a soldering iron to do it you're gonna want to put like a 1 if you want to really blow up the power limit you're gonna want to like you put a 5 milli ohm shunt on here so that's 0.005 ohms on top of the existing tion you're gonna get at Wyatt you're gonna get twice as much power limit so instead of like the 150 watts you would get 300 watts on the 8 pin assuming that you can even set it to 150 to start with I'm not sure and if you wanted to really blow out the power limit like say if you were taking the card on in liquid nitrogen or dry ice you would actually like I would go all the way down to like a 0.015 ohm shunt and the reason for that is is if you actually put that in parallel with a 5 milli ohm shunt you're gonna get like it was a resistance of like 0.01 1 the and so then you're gonna get a roughly 5 times increase in power limit at which point even if you were at stalk you know which is 250 watts while 250 watts times 5 is what 12 hundred and 50 watts power limit total for the card so then you shouldn't have a power limit though you would have to probably add some extra power connectors because the 8 pins they're good to 300 watts they can even if you have as a power supply with 16 gauge cabling you can even actually push as much as 468 Watts through each of these but the thing is you know a thousand watts is actually even more than that so you would really want to add some extra power connectors and it you were doing that you'd have to add two of them because I'm not sure how the 12-volt power planes are actually internally wired on the card and like you know in theory you could get away with three powers three power connectors total but since there's only two to start with it's gonna be hard to wire up a third one so you'd have to wire up to two just keep it even yeah so that is the twenty atti founders edition you know really solid PCB on Elland like vrm is in my like similar to the 1080 TI this vrm is actually adequate even for like ln2 overclocking so if you're on water cooling or air cooling you have nothing to worry about except how high is the software power limit and if you want to modify that you can just solder you know shunts in parallel on top of the existing shunts and that solves that issue it's like the 12 volt balancing stuff like I know I complained about it but it's still from an engineering perspective that's a really cool thing to do because it does mean that you know they close or follow the specifications for the eight pin power connectors though I think you know you could just ignore those up to a certain point you can kind of ignore that but yeah very very solid you know PCB from Nvidia and as I said you know it's like it's gonna be really hard like in my opinion pieces like the there's not much room for the Adhan board partners to make a much better PCB for extreme overclocking I would actually value the addition of like a third eight pin power connector which I think I've seen some cards that might be doing that getting us extra eight pin power connector for ln2 that would make sense but for like daily usage you can just buy a Founders edition knock yourself out like that there's not much to improve on here right like you've got a really really solid v core vrm massive overkill on the memory VR I'm the the new 12 volt balancing stuff which I don't really care about and yeah there's I mean what's not to like this is a really really really sold card so yeah that's it for the video and thank you for watching like share subscribe if you have any comments questions suggestions you can leave them down in the comment section below and if you'd like to support gamers Nexus there's a gamers Nexus patron patreon and you can find that down in the description below there's also store dog gamers Nick store gamers Nexus dotnet if you'd like to buy shirts or mugs or any other kind of merchandise and then if you'd like to watch more overclock like really overclocking focused content I have a channel called actually hardcore overclocking where I do overclocking stuff and also some electrical sort of engineering things here and there but mostly just overclocking so yeah that's it for the video thank you for watching and good bye