Gadgetory

hey guys build Zoid here from actually hardcore overclocking and today we're gonna be taking a look at yet another r-tx 20 80 TI custom PCB though admittedly I've done I think I've done most of them on my own channel so the like not that fitting but I've done a lot of these at this point so I think this is the only custom card I haven't done yet so this is of course the EVGA FTW 3 and personally for me the highlight feature here is that this has a BIOS which before that this video is brought to you by Corsair Vengeance RGB memory coursers vengeance RGB Ram uses pre-screened ICS for better overclocking Headroom and tighter timings courses Ram also has the benefit of wide region motherboard support by landing on qualified vendor list for motherboard makers and turned wide compatibility between boards learn more at the link in the description below ultimately I'm not sure how useful it's going to be because you can't mulder the BIOS on Nvidia cards but you can you know experiment with like using other cart like trying to use a the the BIOS off of a different card to try get the power limit or something but ultimately all of the biases are going to be more or less the same because all of the biases need to be approved by Nvidia and from what I've heard Nvidia is pretty awful about approving biases that do cool stuff so yeah but it is a neat feature and yeah like it is nice that you do have the option even though unfortunately is probably not going to be all that useful to most people now then let's get onto the probably the more interesting part the RMS V curve erm is the pretty standard layout for RTX 28 ET i where you have one group of phases on this side so that's that's one group of v core and then you have the other group of ecore over here so that's also V core and above the larger of the two v core groups we of course have the memory power so V mm v mem is for powering all of the I don't know why I can't write tun M is for powering all of the GDD r6 memory chips around the GPU core then we have a bunch of minor voltage regulator that I'm not even gonna try to point out on this card because they're they're kind of all over the place and they were really high hard to identify but uh the card does ultimately need sort of the following voltages it needs a pecs rail and that's quite possibly generated by this or this or it could be some other thing that I'm not even noticing on the card but that powers the PCIe interface as well as some internal PLL's of the GPU core then there's a 1.8 volts rail which powers the BIOS chips that nvidia uses as well as the VPP rail of the GDD r6 memory chips so that's a supporting voltage for those to function it doesn't really do a whole lot if overclocking wise it doesn't actually do anything then we have also a USBC rail because all of the 20 ATT eyes have a USB C port so you need a power supply for that because you don't get 5 volts for your USB port from the PCIe slot and because you really don't want to have the vrm powering the VRM yeah there's a vrm for the VR I'm probably like since you don't want that one going down when somebody messes with the USB port there's also a separate 5 volts voltage regulator located somewhere on the card probably in this area though this is like that's an input filter that's probably an info input filter that's an input filter this could be 5 volts this also could be 5 volts I assume this is more filtering so yeah really hard to sell like really hard to identify these minor rails with just pictures of the card so that's the best I can do there but they do exist they're scattered around the card they're not really all that important now then I guess let's get into the important V RMS starting with the V Corps so the V Corps here of course looks like a 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 phase the issue is that for some reason and I assume it has to do with transient response requirements but basically all of the r-tx 28 ET eyes ultimately don't use any doublers and instead what they're doing is what I called ASIS out on for the on the Maxima Silla Maximus 11 hero where what they're doing is essentially they take one PWM signal from the voltage controller and shove it into tube phases at the same time which is why you have so many power stages and inductors when the you p95 12 voltage control are located on the back of the card we're not actually gonna go there because they're well we can go on the back of the card it's just a bit blurry oops I've put too many divisions on that side but the you p95 12 that this card uses is ultimately an 8 phase voltage controller so that's this chip right over here and in fact because you uh because EVGA has a nice and tidy PCB here we can actually very clearly see the current monitoring circuits which you'll notice are one two three four five six seven eight I mean clearly we can blur Alesi them but yeah these are the current monitoring circuits and so basically as far as this chip is concerned the vrm is an eight phase-- and as far as the pwm signals go it's also an eight phase-- because there's no doublers to shift the pwm signals out of phase so ultimately you have to power stages turning on at the same time and therefore they're in phase with each other there's no there's no sense calling them separate phases when they turn on at the same time and this is actually what's going on with act most of the RT x 20 ATT eyes now there is a couple cards which do have ten phase voltage controllers with ten current monitoring circuits and you can actually see that on them so those definitely have ten phases but those are also wired to look like sixteen I like those also look like they're sixteen phase or some higher phase count but ultimately all of the the twenty ATT eyes are running either ten phases and the reason that these cards don't use doublers because we can kind of see that you know like on this side of the card we have our six six inductors and power stages here we have the other ten located on this side well we can see on the back of the card we can actually see footprints for what are probably doubler chips because there's three of those one more down here and another one down here they're conveniently within the sort of distance of the power stages and then on the other side of the card we have this one right here and this one and this one so we can clearly see that somewhere along the line somebody wanted to use doublers and then they went like okay no we can't use them for some design reason I'm assuming it's gonna be due to transient response because that's really the only thing that doublers affect heavily well relatively heavily you can try design around it and ultimately GPUs are probably more much more sensitive to this than CPUs which is why it seems that end video is probably requiring that people just don't use doublers otherwise I'm pretty sure a bunch of a board partners would because if you look at high-end 1080 T eyes there's doublers freaking everywhere but with the 20 with the the 28 et eyes nobody's using any doublers because basically when you use a doubler there's a couple nano seconds of delay added to your pwm signals and if you're just trying to match the output the output voltage of the vrm relative to the current pull of the GPU core and you have a GPU core as large as a what is it tu 102 pulling you know limited to 200 and like the thing is like nvidia gets most of their power efficiency not from the actual manufacturing process itself but from really aggressive power gating so the end result is that yes this GPU has an average power consumption of something below 300 watts I can't remember what exactly it was but the peak and minimum power consumption that the card will pull are going to be massive relative to like way off from the average power consumption which means the vrm has a lot of work to do in terms of transient response because your current draw is anything but consistent and for that reason you know eliminating the doublers makes kind of a lot of sense from a design perspective for the r-tx 28 for the r-tx 20 series of GPUs because that is kind of like everybody's doing it even the galaxy all of Fame card which was $1,800 doesn't have doublers so that's the only reason I can think of that is like yeah you don't want to use doublers now doublers do have some benefits they can improve your vrm efficiency because they can give you the ability to properly current balance all of your phases instead of just trying to rely on the fact that as power as power MOSFETs get hotter they go higher resistance so like if this chip gets you know if this chip is pushing more current it's gonna be producing more heat over time it's gonna get hot enough that the internal like the resistance of this chip is going to be high enough that it's gonna push more current into the chip that it's in parallel with because these two will probably be sharing a PWM signal so it'll kind of self current balance but it's not gonna be optimal and also your with doublers you would have slightly improved well slightly improved you'd have improved input and output ripple the thing is once you have eight phases trying to go even higher on the phase account and really like there's diminishing returns so going from eight phases to sixteen the ripple and the output ripple probably wouldn't change a whole lot so there's nothing really wrong with this being an eight phase but it's definitely not so sixteen and my statement about the founders edition in the past being a thirteen phase is just straight-up wrong that's also an eight phase and in fact most of the twenty series is either on eight or ten phases now then so that's kind of the that's part of the control scheme phase wise the you p95 12 also introduces a two megahertz maximum switching frequency which we cannot access well actually we could if we add a full datasheet for the chip because I have recently gotten my hands on the pin out for this thing and it is actually very easy to vault mod and it should be to actually physically set your modify the load line with the resistor on the GPU so that's pretty cool that should be modifiable modifiable with a hard mode as well switching frequency is actually also set with a resistor however that resistor also sets the address of the chip on the I square on the SM bus interface which is also you could all which is also like it's similar to I square C but you can actually use I Square C as well for that but uh yeah so ultimately you can't like while you know if you have the datasheet it would just be a matter of replacing one resistor on the card and you could change the switching frequency to whatever you want which would potentially like well the reason why you'd want to do that is if you crank up your vrm switching frequency you can actually reduce your vrm output ripple at the cost of more power powered more heat dissipation because you're switching your MOSFETs on and off more often in the power stages so the power stages produce more heat the vrm gets less efficient but you get better output regulation so it's kind of kind of how that works is it's it's a fun fact vrm efficiency and vrm output regulation generally don't go hand-in-hand with each other like they're there they're not necessarily linked to each other but anyway this does go up to two megahertz and it is on the still on the PWM vid interface which is the reason why it is so very easy to hard malt this thing so yeah that's the you p90 512 and it's basically a bunch of bolt on upgrades over AUP 95 11 in fact the pin out for this thing is super similar to the u p95 11 which is also why it's so easy to why it should be relatively easy to modify now then for the actual power stages on this card we are looking at the ever-so-popular FDM f31 seventies these are pretty much the default power stage for an RT x 20 series GPU everybody uses them well not everybody Asus likes to use they're more expensive Texas Instruments parts but these are relatively cheap they are very powerful they do push 70 amps they are so I forgot to mention these are smart power stages not just any power stages and the reason why these are called smart power stages is because they integrate current monitoring and temperature monitoring and actually a really like they also integrate protections like 140 degrees Celsius and I'm doing that wrong again that just now this looks better to me I'm just gonna do it that way I don't care which way is the correct way at this point but 140 degrees Celsius over temperature protection and 80 amps over current protection so they won't allow you to actually like overload them now the cool thing is since there's so many power stages here also this is one of those things where if that 80 amp OCP kicks in it'll force the current onto the other Aldens of the other power stage in the pair of two so it'll kinda like if you were maxing out your your current capability then they would actually current balance themselves kind of just bouncing off of the OCP there though if they spend enough if they bounced off that enough it'll eventually raise a flag with the controller and the controller will probably just shut down the vrm at that point so it's not like you can overload this forever but ultimately since that trip point is so very high and there are so many phases you're never actually going to hit that like there's no way you're actually gonna hit that under normal applications unless you try to like like well I guess if you wanted to make like a low voltage room heater with this VRM yeah you could probably hit the OCP if you were working on doing something like that but with while powering the TU 102 core that cores gonna die before the voltage is high enough to push that amount of current now then the end result of all of these lovely power stages and having this many of them in the vrm is that the efficiency on this thing is pretty epic so it's not the most efficient though the the Texas Instruments parts are slightly more efficient I think Infineon also has power stages that are slightly more efficient they're all rated as 7tm parts but yeah some of the competition has slightly better slightly better parts they're also way more expensive but ultimately efficiency wise what we're looking at here for a sweet VR I'm switching frequency of 500 kilohertz and a output voltages of 1.8 volts and 1.2 volts which I've finally gotten my hands on a datasheet for like a proper datasheet for a 70 amp power stage so instead of scaling this off of like a previous generation 60 amp power stage which is what I did in the past this is actually scaled off of a comparable kind of part so this is now more accurate than in the past and I think in the past it was actually a bit low compared to what what it's gonna be now so anyway this is what the the datasheet of the FD MF 31 70 itself is perfect for I can't tell you what I'm scaling this according to because that data sheet is nd aid now then for oh yeah and well if runs off of 5 volts only so I'm not gonna mention the drive voltage here so then for what for the stock current level of around like we're not like it's not actually gonna run stock at 200 amps it's gonna run probably a little bit more than that but around 200 amps output 1.8 volts or 1.2 volts you're gonna be looking at 1.2 volts about 20 watts of heat and 1.2 volts at about 17 watts of heat honestly with this many phases you could probably like unfortunately the thing about all pad like smart power stages and and really any kind of power stage is that everything has an efficiency curve that looks like this and when you have 16 of them you tend to be in this area at like the lower currents now as you go higher current the efficiency starts getting better for the vrm but your overall efficiency of the GPU is going to be lower because it's gonna be pulling a ton more power and not producing a whole lot more fps so just kind of the Nate well it's gonna be pulling disproportionately more power than it's going to be producing fps right like you could get like a 20 well I don't I don't remember how much these overclocked by but let's say you gotta get a 10% performance increase well you're probably going to see more like 20 or 40 and power consumption increase for that kind of performance increase but ultimately the vrm is actually going to be running a bit more efficiently at the higher current outputs now for overclocking on sort of water cooling you're probably going to max out somewhere around 300 amp range maybe a little bit above that at that point the vrm at 1.8 volts would be producing about 30 watts of heat and at 1.2 volts it would be producing more like 26 now as we keep going up in current 4 at this point I'm gonna say like sub ambient so probably around dry ice levels you might be looking at something like 400 amps of current pole and at that point 1.8 volts would be producing about 45 watts of heat and 1.2 volts will be producing about 39 watts of heat and the great thing is even at this point you probably would just get you you would probably get away with just a gentle breeze over the vrm without even needing a heatsink on it to keep it from overheating because that 39 watts is spread across a lot of parts right like that is spread across all of these power stages and they're not even like in one big group right next to each other so ultimately the thermals on this thing are just like this VR I'm if you're on like water cooling you could totally you know if you want one to water cooled the card and you can't get like a full cover block then it would be totally a valid thing to do to get one of those GPU core only water blocks you'd be fine because the arm is gonna be completely fine though the vram on the other hand that might have a bit of an issue but the the vrm itself is not gonna have a problem with a sort of almost zero airflow and fire almost zero airflow in environment because it's just so massive overkill for the lower current outputs here but it is worth noting that at this point it is very much like in the peak part of the efficiency curve so that's kind of why the like the founders edition it has a 13 phase it nails the peak of the efficiency curve at stock settings so yeah that's kind of kind of the reasoning behind why with a lot of the 28 ET Iser so it was so weird in the vrm department now then moving on from dry ice to more like liquid nitrogen we sort of 500 amps and possibly even 600 amps and you know off the deep just in case we're also going to look at what would happen if you tried to push 700 amps though I don't think that you'd ever actually hit this a completely maxed out 1080 TI on ln2 would be somewhere between the 400 and 500 amp range I'm assuming a completely maxed out 20 atti will be somewhere between and I just screwed up my paintbrush which is great will be somewhere in between that 500 to 600 amp range so yeah but this is just kind of like a theoretical consideration and really even this is mostly theoretical because how many people are ever gonna run liquid nitrogen on one of these now for that 500 amp range at 1.8 volts you'd be probably looking at about 60 watts of heat of the arm vrm heat output so at that point you might need more than a gentle breeze over the power over the phases and at 1.2 volts it'll be producing about 52 watts and yeah like at that point air flow would be a good idea 600 amps you're gonna be looking at more like 82 watts of heat which yeah you're you're gonna need airflow maybe even heat sinks but probably just air flow assuming you have a high enough rpm fan ultimately it is worth considering that in extreme overclocking the vrm is fine as long as the benchmark finishes before the vrm overheats so you don't even need to have good enough cooling like you don't need to be able to run the vrm at like 600 amps for an hour it needs to last for like five minutes before it overheats if it can do that you're good so even here you could probably get away without heat sinks and it's also worth noting that the liquid nitrogen has a tendency to freeze everything around the GPU core so if the vrm is sitting I like if you're just sitting the idle on desktop for long enough over time the i/o section of the GPU will like freeze over and so will the like everything really all the way down to the PCIe slot and through the motherboard and yeah the longer you run for and the more time you spend sitting idle on desktop the more frozen everything's gonna get overtime so yeah like the VR I'm cooling considerations on ln2 are kind of are very different from like normal use case scenarios now 600 amps 1.2 volts you'd be looking more like 70 watts of heat and the the airflows thing still applies like this is a lot of heat at this point and 700 amps output you're gonna be looking at about about a hundred and six watts for our 1.8 volts and about 92 watts for 1.2 volts so yeah at that point is gonna be pretty like that's a lot of heat but it's also worth noting that you know at 1.8 volts output 700 amps you're gonna be looking at almost 2 you're gonna be looking at close to like 1.4 kilowatts of power right like 1,400 watts coming out of the vrm and the vrm is only going to be producing about a hundred and six watts of heat which is still pretty good efficiency right ah so yeah that's worth considering but at that point yeah you will need the arm cooling though of course this is even on ln2 I don't think you're necessarily gonna hit that so the vcore erm and sort of standard RT x 28 ET i custom PCB design fashion is complete and ridiculous overkill for what most people are probably going to use it on which is going to be somewhere in this range but yeah it's nice to know that you know who you're there's no chance of you ever overloading than the RM on this thing and these do come with a lot of safety features that include functionality to like save the the GPU core if the high side MOSFET or something fails so even if one of one of these breaks you're probably gonna be fine with the VR like the VR I'm still gonna be relatively functional which is a really neat feature of these these smart power stages now then for the memory vrm we're looking at similar levels of ridiculous overkill and actually the memory vrm is weird a little like very much kind of you know it's in that part of the the efficiency curve where it's like you could probably get away with less phases and not have any issues but the memory of erm is a actual three phase you can't you know do anything weird about that the voltage controller for that is also another you p90 512 except this time it's running in that is not 95 12 that's another you p90 512 right there that's running in three-phase mode and same as what I said for the one on the back this one on the front is going to be very easy to mod just because it's the same chip so if you want to model your memory voltage that's great news isn't it though admittedly like that's probably a great way to kill all of your GD dr6 just because gddr5 X also already wasn't exactly tolerant of high voltages either so yeah also the the voltage applied to the memory chips that is also slightly applied to the GPU core so you could also break the memory controller so again not the best idea to mess with that for daily usage and then since this is the sort of bog standard three-phase memory power that we've seen on every single RT X xx atti using the exact same power stages as every other RT X xx atti this also has the same exact efficiency where it does 20 amps and we're again talking about these operating parameters 20 amps output well except for the part that we're only doing 1.8 volts cuz scaling down to 1.2 volts doesn't work at this kind of low current output level 20 amps output you're gonna be looking at about two outs of heat which is this is below what I think the G what as far as I'm aware the GDD are six memory chips should be pulling or it might be yeah that that might be a bit low I think they're more pulling more like 30 amps and at that point the vrm will be producing about 3 watts of heat and at 40 amps output the vrm will be producing about 3.5 watts of heat so yeah this absolutely does not need any cooling whatsoever this is ridiculous overkill but that's standard for these RT x 2020 ATT eyes anything else I'm missing oh yeah we have a whole bunch of circuitry on here for car monitoring and current balancing kind of an interesting difference between this card and a lot of the other cards is that EVGA actually has an AI na 3 2 2 1 somewhere on here which is kind of a and that's different because normally all we see on 28 et eyes and really any twenty series cards is the on semiconductor NCP four five four nine one current monitors and those basically take care of all your shunt resistors so that's what's monitoring those there's also I think there's the one of the shunts is on the back of the card and we have two of those chips because there's also other shunt resistors that need to be monitored like I'm pretty sure that they're like those are shunt resistors as well that are also being monitored and the reason why we have all of this new monitoring circuitry on these cards is because NVIDIA has per power connector based power limits and that means you need to actually balance your power connectors if you don't want to but bump bounce off the power limit all the time which is something they've sort of improved from the 10 series now if you're wondering if AMD has the circuitry MD doesn't have the circuitry because AMD doesn't bother to actually monitor how much current is going into the card they just care how much current is going through the vrm and that's that's really the only thing they monitor and more interestingly they only monitor the power consumption for like v core on most cards and on some cards they'll also include the memory power but the memory controller because AMD has a separate memory controller rail they don't bother with that one that one's static that one's not monitored so yeah that's kind of a the kind of an interesting thing to consider altum Utley Nvidia you know has all this extra current monitoring and current balancing circuitry just kind of because they're really focused on power efficiency I guess but it does really up the complexity of the cards which is why we have to NCP 445 for nine ones like each of these can monitor for shunts so yeah there's there there's a lot of current monitoring going on on these cards the other thing we have all over the place is a whole bunch of like these random while they seem relatively randomly placed MOSFETs and those are used for current balancing along with like these inductors right here so we have another one of those circuits down here and that is taken care of by UPI semiconductor U P seven six five six five one chips and those just basically choose which which power connector to hook up to which which well like they're they're gonna change with the power connector that certain phases are pulling their power from based on how much power is being pulled through each power connector so it's gonna optimize that you like never exceed the 75 watt limit of the PCIe slaw never exceed the 150 or like whatever power limit it is that the eight pins each have so yeah they can current balance all of those so that you never act well so that you max out your power allowance because ultimately you're still get empowered throttle at some point but it's gonna reduce the amount of power throttling happening otherwise happening compared to what like the 10 series cards would do so that's uh that's pretty pretty neat and I've already talked about the the sort of interesting phases on either side of the GPU core layout in past videos so if you want to you know find out about that you can watch those and for the oh yeah and for the shunt resistors I've actually done a video about how to mold these like really advanced where you can completely disable the power limit I'm not I'm not talking like change the power limit by a couple percent I mean no power limit whatsoever so I've done another video on that as well and I'm not gonna include that because that takes a while to explain so yeah that is the RT X xx atti FTW three from EVGA it's a pretty I hate to say it but it's basically a bog-standard custom RT X xx atti as far as I'm concerned the inclusion of the bios which is nice I do really like I do appreciate that quite a bit but ultimately it's really not that different from all of the other RT X xx atti is out there the V RM is basically the same efficiency has everybody else the the face count is the same as everybody else the power balancing is the same as everyone except MSI who somehow seems to have gotten away with eliminating a bunch of it but I guess that's because they do have the extra six pin so yeah that is the the FTW FTW 3 its it's really good it's just not anything special because they this is kind of standard for RTX 28 et eyes but yeah so that is it for the video thank you for watching like share subscribe if you'd like to support gamers Nexus there's store gamers store dock gamers Nexus dotnet or you can pick up things like the mod map that you can see in the background of the video here and also shirts and mugs and other kinds of merch and then what else is there we also have a patreon so if you'd like to support us through that you can there's gonna be there's gonna be links to all of that somewhere below with a video description or comments as well and if you'd like to see more overclocking related content I have a channel called actually hardcore overclocking where I do things like explain how to do proper shunt mods not just liquid metal or something like that there's a better way to completely disable the power limit and also mods for the things like the you p90 512 so if you'd like to check that out that would be pretty neat and that is it so goodbye