Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

NVIDIA PCBs Slowly Displacing Partners | 2080 FE Analysis

2018-10-26
hey guys build Zoid here from actually hardcore overclocking and today we're gonna be taking a look at the party X 2080 front founders edition PCB so there's not really much else to say let's get right into it with identifying all of our major voltage regulators before that this video is brought to you by us and the limited edition foil graft logo shirt this four color foil shirt is the iconic GN graft logo with average one percent and point one percent bar colors it's printed on a soft high-quality and custom made 100% cotton shirt and is available on store dye Karen's Nexus net until stock runs out once it's gone we will not be making more of these shirts we sold out within two weeks of our previous limited-edition shirts so click on the link below to preorder now first up we have of course the V Corps vrm which is the largest and most important because it powers the GPU core and that obviously uses the bulk of the power on a GPU above that we find the memory the memory of erm so that's V mem over here and that of course powers the GDD r6 memory chips and below that we had the V core right there over in this corner over the card we welcome corner yeah basically a corner we have three smaller voltage regulators these would be for the 1.8 volts and I'm not actually labeling the specific regulator's here because I don't know which ones which but there will be a 1.8 volts rail up X rail and a USBC rail so the 1.8 volt rail is for a supporting as a supporting voltage for the GDD r6 memory chips it also it also powers and videos BIOS chips those run off of 1.8 volts as well Peck's rail is for the PCIe interface as well as some internal PLL's on the GPU core and then the USBC rail is of course for the USB C port on the back of the i/o since you can't actually well you need a voltage regulator for that rail and then over here we also have what I assume is 5 volts because these power stages run off of five volts so you need power you need a vrm for the bigger vrn funny how that works out now then with all of those you know identify let's get into some of the details of the two main ones which are the V Corps and the memory of erm so the V Corps here is a one two three four five six seven eight phase and this is actually completely reasonable because we do actually have an eight phase voltage controller here that's actually not that new for high end and video cards I mean the 1080 T I already have used a u p95 11 which was an 8 phase voltage controller here we have the successor of the 95 11 the 95 12 which is also an eight phase-- but adds on a few more features it goes up to two megahertz switching frequency which is not useful because at that point your V RMS efficiency is going to be terrible but the idea is that basically if you are using a doubling scheme or a quadrupling scheme you can run that really really high switching frequency into your doublers or quadruple errs and on the other end of those it'll either end up as 1 megahertz or 500 kilohertz at which point that's actually a reasonable frequency to run into various MOSFETs and power stages so that's some of the reasoning behind why some voltage controllers can go up to ridiculous switching frequencies when you wouldn't ever want to run anything that high into any of the the power stages that are being used with them so it goes up to 2 megahertz the other major addition with the U P 9512 is that it has an SM bus interface which previously there was no no digital interface whatsoever on the U P 95 11 that that thing basically the only interface it used which the U P 95 12 still uses was a PWM vid and basically PWM vid is a is a PWM PWM signal that comes from the GPU core into the voltage controller and that is used for setting the core voltage you can't actually ask the u p95 like in in the past with the U P 95 11 it wasn't actually possible to ask the 95 11 what voltage it's actually outputting like it was not possible to do that and so basically your software voltage reading was basically whatever the driver was telling the chip to produce not what the chip was actually producing which is just kind of an interesting side note about how the ten series cards worked but the 9512 can actually report things like the RM current output you know voltage output voltage you can change the switching frequency you can change the load line calibration you can change all of this over the digital the the SM bus that it has only issue is as far as I know actually writing to it might be disabled by and by default and there's no public register maps for the chip so good luck with accessing that though elmore is there is going to be an Elmore EVC which is basically a little I Square C controller which actually hooks up to the to the SM bus on this and that'll give you those kinds of settings and the reason why that works is elmore works at asus so he has access to the button and to the NDA data sheets which include things like you know register maps so yeah there will be like there will be ways to actually get control of all those features it's just not very easy because again you'll have to solder onto the GPU in order to use an elm or EVC anyways this does mean that we have a real eight phase vrm here and it's not doing any of the weirdness that say the 13 phase monstrosity of the twenty atti is doing because i like that is just odd matter phase account is literally but eight phases is a pretty normal phase account that's completely reasonable to you know get out of a you p95 12 because it does actually support this phase account now for the phases themselves you're looking at just one inductor and one power stage in each phase and the power stages and videos favorite 70 amps smart power stage from Fairchild Semiconductor the FDM f31 70 so that's 70 amps and it's a smaller power stage the reason why it's considered a smart power stage and not just a power stage is because it integrates current monitoring and temperature monitoring functionality directly into the chip and that also includes over current protection and over temperature protection the over temperature protection is not exactly a hard protection as it basically just raises a flag but the overcurrent protection will actually just straight up shut the ERM down if you overload it so it's pretty neat you can't actually well you'd have a hell of a time trying to bring this vrm out because it should shut down before you manage to do that they also have some inbuilt protections for various mental functions so in theory which also the you p95 the the new new vault some of the new voltage controllers also have some functionalities that work in and with that well basically if one of the phases on the vrm goes down in theory the vrm could keep operating as long as the malfunction isn't one of those where basically the 12 volts on this side of the vrm ends up shorting through the power stage to this side of the VRM at which point you're screwed because everything on this side of the vrm which is primarily the GPU core is going to be very very dead though that is a well it's not exactly a rare malfunction for a vrm but it's it's it doesn't quite have like it's not exactly something you should be particularly concerned about because it's like it like the arms don't tend to malfunction that often they don't get tend to get overloaded that often especially not when they look like this but it is a realistic like it is a real failure mode that can occur and does cause major issues because well then you end up with cards where even if you try to repair them well this is dead anyway with the FD MF 31 70s here you know that's those kinds of sera scenarios are actually much less likely because of all the built-in protections that they have now the next thing that is kind of interested and so the end result of using the FD MF 31 70s here is that the VR M here gets a pretty good efficiency at least within the current ranges where would normally operate there are you know of course custom PCB RT X 20 80s which have more powerful VRMs but the thing is you won't really notice that difference until you're pushing way more current than in RTX 2080 should ever really need so let's get into those of yaaram sort of the heat output figures now unfortunately the FDM f31 70 is entirely spectat 1.8 volts I did scale it down to 1.2 volts using a datasheet for a 60 amp international rectifier 3575 power stage just because that's like a nice modern 60 amp power stage it should be relatively good for trying to get the like should be relatively close to the real-world efficiency scaling of this thing compared to say some older power stages which would have actually have a well worse worse time handling the higher voltages because I am pretty sure that these are very much designed for the Intel skylake Zeon's more so than they are designed for NVIDIA GPUs even though currently these are especially popular on NVIDIA GPUs and I've not really seen a bunch of Intel motherboards with B's on them yet so let's thought to get into those efficient into those heat output figures so starting at 200 amps and also worth noting is the switching frequency in the datasheet is spectat 500 kilohertz they also spec like 750 megahertz and no 750 kilohertz 1 megahertz and I do believe 1.5 megahertz so you know they do go up to higher frequencies but realistically this is going to be running 500 kilohertz now 500 kilohertz switching frequency 1.8 volts output you're going to be looking at 200 amps with a heat output of about 23 watts at 1.2 volts you'd be looking at about 18 watts now for comparison the FT w3 from EVGA with its 12 phase VR I'm using the exact same power stage would actually be doing about 17 watts at one point at the 1.2 volts right at the 1.2 volts output and about 21 watts at the 1.8 volts output so very small difference like we're talking single-digit you know improvement in vrm a single-digit lower heat output for the vrm on the F w3 and that's after it adds a whole extra four phases but the reason for that is rather simple at low current outputs the majority of your power loss is not gonna be due to the current it's gonna be you to just running the power stages and well you know basically you can't have basically the issue is that if as your current out like your V RMS heat output is not going to be zero when the current output of the V R M is zero it's gonna burn power just to run even if it's not actually pushing any current and at low currents that actually means you're burning a ton of power doing a whole lot of nothing at higher currents your primary power loss is going to be just all over the current going through the power stage of producing a bunch of heat and then having a lot more phases actually starts being more useful which we will soon see as for 300 amps output this vrm would produce about 42 watts of heat and about 34 watts scaled down but aft w3 at that same output would be producing about 28 watts of heat for 1.2 volts so you'd be looking at about seven watt reduction no six watt reduction at 1.2 volts in the RM heat output by having the twelve phase instead of this eight phase now going up further 400 amps output and we're gonna get to where these current ratings go 400 amps output you'd be looking at about reading the long set of notes 68 watts of heat and about 54 watts of heat at 1.2 volts output the FT w3 for comparison would be doing about 40 on that 1.2 volts output at this point like we're talking about a pretty major difference that's 14 watts you know less power consumption that's almost that's probably close to half the amount of power that all of the GDD r6 on this card uses so yeah that is that you know that starts being a significant difference in power consumption right there and then the highest current rating that the datasheet really makes easy to calculate because while these are technically 70m powers they only speck the datasheet up to 60 amps which is kind of normal 60 amp power stages normally end their data sheets at 55 amps and then the lower I am stuff that actually ends where the current rating ends but for these 70 M power stages they stop at 60 so 480 amps for you know 60 times 8 that's just kind of easy to calculate here so 1.8 volts you'd be looking at about 96 watts of heat which is like a loft and then at one point two volts you'd be looking at about 77 watts of heat for comparison the FT w3 would be doing 500 amps because that's a 12 phase so I actually had an easy time calculating the 500 amp figure and didn't have to go to 480 the the 12 phase FTW 3 would be doing 500 amps at about 72 watts of heat output for 1.8 volts and for 1.2 volts it would have been doing a it would be doing 58 so basically at 500 amps output the FT w3 the RM would be producing about as much heat as this produces at 400 amps output of course at the reasonable current ranges like this is realistically I don't think like on water cooling you really shouldn't be able to break that 300 M figure so you know so 200 amps is around stock and somewhere between 200 amps and 300 amps is gonna be like your usual overclocking range so stock to sort of h2o and I'd say 300 amps you to already be a store or starting to look into like dry ice range cooling you know so stock well I should have also included air cooling because there's obviously custom air coolers and those are pretty good there and so you know really you'd already be pretty much like this vrm is pretty good for everything except maybe ln2 and really even with the liquid nitrogen you know you'd probably not hit that 480 am figured you'd still be probably somewhere in the 400 amp range maybe even lower 1080 tea eyes would actually go into the sort of 400 amp range and this is technically a bigger die but it may not scale to as high voltages as a 1080 tea I did you could push like 1.5 5 volts in a 1080i or even 1.6 if you had them really well cooled with this thing you're really not gonna like it if this ends up running on less voltage then you know you're not gonna hit that amount of current pull so even on liquid nitrogen you'd still not be you know in that really uncomfortable for Hut like that this is a lot of heat and even this is already a loss of heat though obviously you'd be somewhere in between these two figures on the liquid nitrogen but uh with these you know high higher heat outputs it's not like you'd really be limited even on liquid nitrogen with the with the eight phase here you could maybe marginally have a slightly cooler running vrm with the twelve phase on like the FTW three well practically speaking you wouldn't really see much of a difference this wouldn't really this vrm shouldn't really stop you even on liquid nitrogen so yeah this is this is a nice PCB it's it's a solid VRM you know it's it's not massive overkill well for me maybe for stock air cooling in h2o you could say yeah it's pretty overkill for those but I wouldn't really consider this massive overkill we we get into massive overkill territory when we're talking about well say the FT w3 right that has way too many bloody phases this is more like reasonable amounts of overkill or what I would consider a solid vrm because you do get that really nice efficiency right there at where it would normally run and then sort of 300 amp range yeah you could probably do better if you had a couple more phases but uh you still wouldn't really notice the difference very much right like we're still talking single-digit different single-digit reductions in vrm heat output and that's with going up to a card with an hole extra you know four phases which is basically a 50% larger vrm so because this is an 8 so going to 12 yeah this is a 50% increase in phase account and you'd still not see much of a much of a difference so this is a really really solid V curve erm as just not you know ridiculously so and above that we of course find the memory of erm which is a two-phase using yet more of the FTO on earth is going on give me a second there fix that so the the two-phase memory vrm is of course using yet more of the FDM of 31 70s so this is massive overkill the memory of erm you're looking at sort of current outputs of say 20 amps and 40 amps which is like like you shouldn't even hit 40 amps output at all but 20 amps you'd be looking at about 2 months of heat for wealth of 40 amps you'd be looking at about 4 watts of heat using the FTMs 31 70s for the memory is massive overkill there is absolute like the GDD r6 should never pull so much like really should never pull more than 20 more than 20 amps because this does have even less memory chips than like a 20 atti because this is on that 256 bit memory bus instead of the 384 so that kind of that that's kind of them so the memory of erm you know absolutely great no problems there the voltage controller is yet another you p90 512 which is kind of an odd decision because for example the 10 series those would all actually use different voltage controllers for the memory of URM because that's an 8 phase voltage controller running in two phase mode and that is really kind of silly to do it's not you know like I get why that like it does keep the Bill of Materials simple and it may you know ultimately it might not cost that much to just keep this chip anyway but in the past they would have actually opted for a different chip like they'd have a you p90 511 and then they have like AUP 1666 for the for the memory of erm but here they're just like now whatever just run the 8 phase for it and right run the 8 phase chip for the two phase memory of erm so that's kind of an interesting decision there that Nvidia went for and then of course in typical RT X 20 series fashion we have two current balancing circuits these don't actually current balance the phases the cart balancing for the phases is of course done by this lot of this chip right here that's its job no these extra circuits here are four current balancing you're very like your oddball six pin here and this is actually just a 6 pin like the the this part of the connector is just like a solid block of plastic I really don't know why Nvidia didn't leave that area open you know because if you have one of those you know if you have one of those 8 pins that looks like you know if you have one of those 8 pins that looks like this where you have the you know the the ones with the separate 8 pin part like the plot of the two extra pins separate like that and they add up like this well you could own a lot of cards you could just sort of take this part and if this wasn't right here that could just sort of sit you know rest there it's not there's not anything here that it would short-circuit out by being there you know short out against by being there so I really don't get why Nvidia decided to block that off because honestly it just makes as far as I'm concerned it just makes a cable management potentially a bit more messy and it doesn't really add anything to the card right and initially I was really confused why on earth they didn't just use a straight up a pin which I still think is just like why why even bother with this like yeah I I don't know I don't get why what this is supposed to achieve it doesn't look good like like nobody's gonna see this once it's plugged in nobody like no idea absolutely no idea I'd honestly argue they could have just kept it as an 8 pin and just made that 8 pin optional and then just like because that's actually supported by the PCIe standard that's why the the extra part of an 8 pin like there's no actual power like no real power transfer like all the X's well all the extra two pins on an 8 pin add is this extra ground connector they don't actually add an extra power line right they add a ground connector and they add a sense line and it's like they could've and the whole idea behind that is is that extra sense line tells the connector that hey you're running in 8 pin mode not 6 pin mode so in theory you can have an 8 pin which will work as a 6 pin right that's that's why the spec is done the way it is like it's designed so that yes you could plug in a 6 pin into an 8 pin the card can then identify that it doesn't actually have an 8 pin plug it in it's just on a 6 pin and but then it can decide to run at like reduced power limit or something right you could you can do those kinds of things that's why the freakin sense pins are built in but now Nvidia is just like let's just put a block of plastic I don't know for such an expensive card I feel like they could have you know use that as an extra feature but hey then again you could also just cram double 8 pins on just about everything with my logic because it's like well if you don't need the xray then you don't have to plug it in and it's part of the specification that you don't need to do that it but I'd say it's like this is weird it just looks weird they should have just used a normal 6 pin but anyway they do have these current balancing circuits and those don't current balance the vrm no no no see that would be far too useful now they current balance the power the power connectors because Nvidia with this basically power connector system they'll have like a 75 watt power limit on the 6 pin they'll have like a hundred and 50 watt power limit on the 8 pin and they'll have another 75 power 75 watt power limit on the well actually that one on the 12 volt rail your so only supposed to go to like 66 watts and so because of this and the fact that Nvidia actually monitors all three current inputs on their cards with you know these shunts you can run into funny situations where say you set your you know you set your power limit on the card to and I'm just for convenience sake gonna set that to 75 Watts I'm too lazy to deal with that let's say you set the power limit to 300 Watts right so the whole card is allowed to run a whole 300 Watts well you can run into a situation where you're gonna be pushing say on this right you're gonna be pushing 50 watts into that and then you're gonna be pulling 150 on the 8 pin and you're gonna be pulling 50 on the on the PCIe slot and the thing is yeah you're not hitting the 300 watt limit but you are hitting the limit for this power connector right here so the card starts to throttle because that's what you want with your card to do so in videos solution to this problem that they created because an AMD cards can't actually do this they they don't know where the power comes from like if you wanted to well you could do some very odd things to AMD cards with and in terms of their power systems like there's our x5 ATS that have a six pin that literally isn't detected at all like you don't have to plug it in because it's just in parallel with the existing eight pin so that leads to some funny situations like if you plug in an eight pin and or you plug in an eight pin anis expand the card actually doesn't see a difference because the six pins just kind of there it's no I actually properly hooked up and so those cards don't care where the power comes from they're just gonna grab it and use it but in video cards because of all the lovely power monitoring they're they're gonna complain if you start overloading one of their power inputs so you need these power balancing circuits to basically you know it's like say say this fate like this would be hooked up to say this phase and this phase is pushing say 20 Watts right and normally this phase would be on that overload at eight pin well it decides that okay so this this is overloaded so the power balancing circuitry routes it to the six pin and that frees up the 8 pin and the 6 pin goes up to 70 watts output instead of just 50 right and this goes back down to like 130 and then you're back you just backed off the power limit and the throttling stops so that's why NVIDIA has this power balancing circuitry which I quite frankly feel like you know it would have just been easier if you eased up on the power limits you know you could have just done that that would have done this that would have achieved it a similar effect though admittedly some power supplies might not like that a lot but quite frankly it's just nobody's like nobody in nobody except well there's only two companies that make GPUs so that's kind of hard to argue about but generally speaking in the past even Nvidia didn't bother with this so I don't see why they're bothering now and part and the things making up the sort of power management system on this card are of course the two you p75 know seven six five one cues which control the power balancing circuitry and then for current monitoring we have these two chips right here which are NCP forty five four nine ones and those are used across all Nvidia of twenty series cards so far both these and these so that's how in video sort of has their their power management like those are the chips taking care of the power management the NCP's monitor the various shunts around the card like these guys also this right here should be a shunt resistor that right there should be a shunt resistor we can also see some little MOSFETs right here so those would be used for power balancing as well then we can have where else do we have it there's more of them around the card it's kind of hard to keep track of like oh yeah there's one up here we have some more random little MOSFETs you know so basically Nvidia adds a whole bunch of circuitry to their cards to better manage the the power the input power can set like the where the power is coming from into them to be to reduce the amount of power throttling the cards do because of Nvidia's or very low power limits so yeah but I mean hey nice way of solving a problem you created I mean you know instead of just setting the power limit to something reasonable or you know just ignoring the fact that you're overloading one of the inputs it's not like it's not like pushing 200 watts through an 8-pin is actually going to cause a problem because if that did cause a problem vega's would be setting people's people's systems on fire which they don't do so also skylake chips would be doing that at like sky like X CPUs would also be doing that and actually a lot of like past GPUs as well but yeah so that is it for the RT x 2080 solid enough PCB you know has has the usual 20 series NVIDIA sort of features 70 amps smart power stages a whole bunch of card balancing and power monitoring circuitry that quite frankly I think is unnecessary very solid you know you have an extremely overkill memory vrm and a solid vcore vrm to go with it and yeah there's not really much else to it there's that's it that's that's the RT X 2080 founders Edition PCB it's uh it's solid enough you know there's definitely more more powerful PCBs out there I really don't think you'll notice a difference if you actually go to those though those cards may ship with much much higher power limits and quite frankly you know the sort of the leading so far the leading in like the leading features for RTX 2004 clocking or basically having a very high power limit having a really big cooling system or a good cooling system generally speaking with heat sinks the bigger it is the better it is at dissipating heat so big and good kind of go hand-in-hand in that sense but you can also design big things that suck at cooling that's just like but if you know what you're doing then you know if you make it just bigger than generally it'll also be better if you don't ridiculously screw up the design in the process so now it that is it for the video so thank you for watching like share subscribe leave comments questions suggestions down in the comment section below if you'd like to support gamers Nexus there is a store gamers Nexus dotnet where you can pick up things like the gamers Nexus mod pack which you can see in the background of the video and also there's the gamers Nexus patreon if you'd like to you know support gamers Nexus through that and I run a channel called actually hardcore overclocking which is why I introduced myself as build Zoid from actually hardcore overclocking where I do a whole bunch of other overclocking related content like more PCB breakdowns and as well as sort of other stuff that's overclocking related it's kind of hard to describe so that is it for the video thank you for watching and good bye
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.