Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

GTX 1080 Ti Founders Edition PCB Analysis & Shunt Mods

2017-03-18
hey guys build loaded here and today were going to be taking a look at the 1080 TI founders Edition PTV now interestingly enough the PCB here is exactly the same as what you would see on a Titan X Pascal so that means the actual copper fiberglass sandwich that you know pushes all these that all the electrons move through that's exactly the same however Nvidia has changed the components that are used on the that are actually attached to it and they have added more components but before getting to build Zoids analysis this coverage is brought to you by EVGA and their upcoming 1080 TI FTW 3 icx card key learn more in the link in the description below or watch our previous icx review there's a lot less stuff missing from this PCB than compared to the Titan axes as and by missing I mean unpopulated MOSFET mounts and that really the only thing that's the big loss compared to the Titan XP TV is this memory chip right here which incidentally you even if you had a BGA rework machine if you have VGA soul during equipment you can't add that gddr5 X chip back as Nvidia disables the 32-bit memory interface for that memory chip in the GPU core itself so you're not getting that back that is disabled completely no chance of getting that working now then with that out of the way let's take a look at the actual vrm on the card and talk a little bit about how about what they're made up of so first things first up here in the top left we find one point a 1.8 volt we are M this powers Nvidia some of the PLL's of the GPU it also powers the BIOS so the EEP ROM of the Nvidia cards actually runs on 1.8 volts so it takes care of that and it's voltage does absolutely nothing for overclocking it's just necessary for the card to function so not really that important now then down here we find another one of these not really that important this is the PE X or also also slash PLL voltage it does take care of some internal circuitry for the GPU as well as some internal PLL's again it's a case of if this voltage is not present on the card the book card will not function this one sits at one point zero two volts at stock and it can help with overclocking to tweak it up or to tweak it up or down under ln2 but on air cooling and water cooling it does absolutely nothing for overclocking capabilities now then moving on we finally get to a vrm that actually does something for overclocking this is the V Corps v RM and unfortunately because this is Pascal it doesn't do much so this applies the bulk of the GPUs power to the GPU core runs most of the compute logic in the core sits at around one point zero eight volts by stock maximum voltage limited by nvidia is one point zero nine three volts it's questionable if it actually can hold that voltage steady but that is the maximum voltage you can set in software now unfortunately so you know you can't really push this voltage that far however this vrm is the one that you'll see most custom cards significantly changing and it does it is rather important for ln2 overclocking but for air cold and water-cooled overclocking it really doesn't matter as long as there's enough power available for 2.1 gigahertz overclocks which is generally what pretty much every Pascal card can hit if you get lucky some of generally the clocks end up between 2 to 2.1 gigahertz from what I've been told if you increase the core voltage of yeeah through physical mods or a certain modded bios which exists for only one card currently if you significantly increase the voltage you do not gain that much core clock that it makes it worth it and the power of power draw increase is quite significant so while this vrm has been very important on other architectures on Pascal it doesn't really do much until you pull the you know out of the cards below zero again unfortunately so that's your V core v RM and above it we find the memory v RM which is another case of doesn't really do much for air cooled or water-cooled overclocking this powers the gddr5 X memory chips they run on 1.3 5 volts and they don't really need a whole lot of power and tweaking this voltage can get you a little bit more frequency out of the gddr5 X memory chips it's mostly useful again on lnto where every couple megahertz helps on air or water cooling you'll just massively increase the power draw for very little actual performance increase so also you can't actually access the crm through software whatsoever so then finally moving on we have another 1.8 volt vrm which I'm not entirely sure where it's located I think it's somewhere in this area of the PCB again I'm working off of photos not the actual cart like I don't have a card in hand so it's rather hard for me to be accurate with a lot of these things and it should be somewhere in this area and that's a 1.8 volts voltage and that's called that's known as the VPP and that's able to do necessary for the gddr5 X chips to run it does absolutely nothing for overclocking so not worth worrying about whatsoever so I this this voltage might be tied to this vrm I'm not entirely sure I started noticing that some and like 10 80s GTX 10 80s had a dedicated one for this just 1.8 volts and it was in this area Theisen X Peas I'm not sure so I might be wrong about that one either way it doesn't really matter that much unless you're trying to revive a dead card in which case I really hope you have a working one because that's the best way to reference what could be wrong now then with that out of the way let's take a look at the V Corps and the memory v RM in more detail even though they don't really do much for overclocking you probably want to know what they are and aren't capable of so the V Corps v RM is a bit of an odd thing as we do have one two three four five six seven inductors and one two three four five six seven eight drivers so this is perfectly fine the driver counts at least eight drivers is perfectly fine as this V RM is run by the U P nine ninety five eleven voltage controller from UPI semiconductor and this is an eight phase voltage controller with up to six hundred kilohertz switching frequency how which is actually on the low side for v RM controllers a lot of other voltage controllers will go up to one megahertz or even two megahertz however that is not that important because really high switching frequency in your V R M means high low power efficient basically ruins your power power efficiency as the V R M will be basically burning heat to turn the MOSFETs on and off more more more times than is necessary it can help with the V RMS ability to regulate voltage under very under transient loads of the vrm up basically the voltage controller has more chances per second to tweak the current going through each phase however it again like we've seen that custom cards on the 10 custom ten series cards do not clock any better than the reference cards so far so again this doesn't really matter unless you're doing ln2 overclocking where the power draw is significantly higher and the v RM actually has to deal with significantly bigger load swings so the 600 kilohertz limit here really isn't an issue what is an issue is that we have eight drivers and seven inductors how does that work also we have one two three four five six seven eight nine 10 11 12 13 14 15 Dean mosfets so we have enough MOSFETs and drivers to build an eight phase-- vrm but we only have seven inductors so what on earth is going on here well NVIDIA has decided that they're going to share this inductor between these two MOSFETs and the you two MOSFETs and this is done so that they can balance the 12-volt power supplies the present in the GPU so your six pin and your eight pin are not actually sharing their 12 volt layer of copper each of those goes into a different light a copper power plane so one part of this vrm will be wired to the eight pin and the other part of this vrm will be wired to the six pin and some of the VRMs on this card will be running off of your PCIe fault connector and possibly this phase might be to I'm not entirely sure because from the photo I can't tell you what these MOSFETs are actually wired to however this is entirely done by Nvidia to basically let them balance where the card pulls its power from I'm not even sure that that you actively use it well actually no they would be using it since they did put components on the PCB there so otherwise that would be a horrific waste they could just leave them out so yeah so this is basically done by Nvidia to balance the power draw between the different 12-volt sources it doesn't really do much for the vrm current capability because all the other phases are still limited to only two moths that's where as this phase alternatingly uses for so this phase is the least likely to fail in case you push are really if you you know have cooling issues on the VR MN Porsche very the high amount of current through it now then let's talk about current capabilities Nvidia has opted for SD PC eight zero one 6s MOSFETs from Fairchild dual dual MOSFETs from Fairchild Semiconductor and so dual MOSFETs means that this chip right here has two MOSFETs in it there's a high side fat and a low side fat so the basically the end result is that we now have two dual sets per phase so you actually have four MOSFETs in each phase the reason for this is that the two ICS share the current load which leads to them sharing the thermal load which means lower vrm operating temperatures as well as higher efficiency because a big part of vrm efficiency is the heat lost on the moss on the resistance of the MOSFET so if you double up the number of MOSFETs you basically get half the electrical resistance through the MOSFET so it gives you a nice efficiency boost as well as a improvement in thermals now the Fairchild Semiconductor fdpc eight zero one six or a small sets are pretty high-end well are actually really like they're not but Fairchild does make better MOSFETs than this but among all the MOSFETs available these are really really nice they offer extremely low switch switch on and off timings and they also have very low RDS on which means that within videos claim of this vrm being capable to do so in video says that this erm is capable of delivering about 250 amps if you actually push that much current through this vrm you're only going to be dumping 16 watts of heat which is really not that much so if you actually upgrade the vrm cooling to handle a lot more you can push as much as even 400 amps at which point you'll be hitting about 40 watts heat output on the vrm assuming 600 kilohertz switching frequency which is the limit of the voltage controller and not sure that in video would actually run the vrm at 600 kilohertz I can't tell you what they run the vrm app from a photo unfortunately so 400 amps 40 watts of heat output but where is the absolute hard limit assuming that the arm is at 125 degrees and you have infinitely capable cooling as long as that cooling keeps it within 125 degrees well theoretically this erm could go up to 8 great amps we do have a lot of MOSFETs here do keep that in mind that that number does work out these MOSFETs are very very capable assuming you could actually cool 800 amps of going through this going through the CRM because that would generate 156 watts of heat which to give you an idea of what you need to cool that with that's like a hyper 212 Evo literally just sitting on top of the vrm not really practical not really doable your vrm is going to fail if you try to push 800 amps because it's going to overheat and basically destroy itself now that moving on to the memory vrm which again not that important but important enough that I'm going to talk about it uses different MOSFETs from a different company these are Alpha and Omega semiconductor EC 930 s these are worse than the fairchild semiconductor MOSFETs however they are ridiculous overkill as this memory the RM is two phases we have two inductors that you can't see because they're under thermal pads and we do have four of these MOSFETs and these are again dual sets so there are actually two MOSFETs in each of those I sees these are the exact same offsets and video used on the Titan X Pascal except they only use one of them in each phase and they also use them for the V Corps and they didn't do two of them per phase so like this card got a significant upgrade the V Corps vrm has new like better MOSFETs and more of them and the memory vrm has the same MOSFETs and more of them so this card is really really nice compared to like the Titan X Pascal's PCD now one thing to keep in mind is that the memory vrm will never really exceed 20 will be between 20 and 30 amps of power draw depending on how much load is on the gddr5 X memory chips and the memory system and that's really not that much and this set of MOSFETs if you actually assume that kind of current going through them would have very low switching and conduction losses and I'm sort of suspecting that my calculation doesn't work when you put too little current through them like when you run the calculation with too little current through it because it will spits out about one wall now if you actually try to max out this VR I'm at 240 amps 44 amps it should put out 50 watts of heat which is again completely uncool about now at a half that it would be actually reasonable now you know gddr5 x doesn't mean that much power so this memory vrm is ridiculous overkill and it is too true phases as it is controlled by an UPI while you pee this thing right here is a u p1 6 v 8 which is a 2 phase voltage controller with integrated gate drive however I'm pretty sure this is a dedicated MOSFET driver right there I'm not entirely sure as you can clearly see that part of the photo is very blurry and I do have other photos but without like a mem without the card in hand it's really hard to say for sure so actually covers the build quality part of this video and to be completely honest with you I'm impressed I'm really impressed with on what NVIDIA has done with this PCB compared to the Titan ax Pascal it is a significant upgrade I as I said before better MOSFETs in the vcore vrm more MOSFETs in the vcore vrm the memory of the RM is the same MOSFETs but there's more of them so that's also a nice upgrade and like really there's nothing to complain about for air cooling or water cooling like I don't see a great reason to go and buy a more expensive car just for the PCB because as I said well at least from what I've heard on ambient cooling you're really not going to be able to benefit from a significantly more powerful vrm or a significantly higher phase count v RM or a just of erm that has a better voltage regulation won't really get you anything as we've seen on past 10 series cards like 10 80s all pretty much hate to - 2.1 gigahertz regardless of who makes them so assuming that you have enough cooling of course which is not part of what I'm covering here so I'm actually impressed because this PCB is significantly over built like nvidia could have gone with less MOSFETs cheaper MOSFETs and and they didn't so I'm impressed here and you know really don't see like a reason why you would if something that you're going to be water cooling I really don't see a reason why you would go and wait for a different PCB - kind of like for for custom cards to come out because getting water blocks for those is going to be harder and everything so if you're just water cooling this PCB is great if your air cooling then you have to put up with the founders edition cooler which is always going to be worse than the custom coolers that are available and the other cards which is really what you want to go after with 1080s so then that covers the build quality let's do some modding on this card because evidently it's strong enough let's try blow it up if you want to mod the power limit of n video cards beyond what the software allows there's a very simple solution for you the Nvidia cards monitor their current draw with shunt resistors located here here and here now there's one issue with actually modding these shunt resistors measure khara power draw by basically measuring the voltage drop across them and the problem here is that if well the mod works by shorting the current resistors which lowers their electrical resistance which lowers the voltage drop across them when you push current through them the problem with that is is if you short them out too well so like let's say you take solder and just solder right across one of those you're going to end up in a situation where the Nvidia driver will notice that the GPUs reported power draw is way too low for any kind of realistic scenario and it'll put the car in safety mode and safety mode foreign video card means that the GPU core is stuck at 139 megahertz pretty much making the card completely useless so what you end up having to do if you want to actually mod the power limit on in video cards is you either have to buy other shunt resistors and put them in parallel with the ones already present on the card if you're going to be doing that you don't really want to drop the resistance below one-third of what it currently is so this right here is a five milli ohm so you would want to drop it to say you wouldn't want to drop below two milli ohms resistance and you can do a calculation for parallel resistors now the simpler less mathematical option is to just say take some liquid metal thermal paste so say thermal Grizzly conduction or collaboratory liquid ultra and just apply that to the shunt of course that does leave exposed liquid metal sitting on a shunt inside your system so you know some people aren't comfortable with that solution but don't apply too much apply a you know just enough to actually short the shunts but not so much that you have a great big blob of it on there otherwise that will drip off and cause all kinds of issues so you can actually do it with the liquid metal and the benefit to doing it with the liquid metals instead of soldiering is that you can actually remove the rip liquid metals with with ethanol acetone or isopropanol you can remove it and it looks like you've never actually modified the card whatsoever which is really useful if you want to send the card back in for our ma after it died and that's really up to you to make this side if that's morally wrong or not I'm just saying that's an option the other thing is if you overdo the mod it's really easy to clean it up so that's the other thing now one word of warning about the KU Laboratory liquid ultra I've had one person tell me that they use KU laboratory liquid ultra and it ate the solder so if you're going to be doing I recommend using I recommend using thermal grizzly conductin aw as that one I know for sure does not eat solder cool laboratory liquid ultra for most people it's worked fine from what I've heard except this one exception which I which I got a message about where it actually straight-up ate the solder so and the shunt fell off and that breaks the card because the shunts literally connect to the six pin to the vrm this is like that shunt is the only connection between the 610 and the vrm so if that falls off the card is not going to work you're going to have to resold er that back on there and then your warranty is toast so that's less than ideal so that's really all there is to say about video cards any more advanced models would are in my opinions like I'm not going to cover any more advanced modes here thank you for watching like share subscribe if you haven't already and if you have subscribed there's the notifications button in case you want to be notified immediately about every video put out by the people here at gamers Nexus if you would like to support us we do have a patreon you could go and donate to that if you would like to see more content like this you can head over to my channel which is called actually hardcore overclocking where I do PCB breakdowns and a lot of overclocking thank you for watching and see you next time you
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.