Gadgetory

so we're finally doing the shot mod and we're trying it first on a three thousand dollar card the Titan V the idea of this mod is to help reduce the power restriction that we're facing with the card and Vidya has shunts that are inline and those help determine how much power is being drawn so if we can trick the card the GPU into thinking it is drawing lower power than it actually is that will give us more power Headroom for overclocking and for maintaining higher clocks for longer periods of time this will help us in our endeavors to hold the top slots on time spy extreme or whatever until the real overclockers get their hands on cards and it'll be a good learning experience so we can show how you can find shunts and what the process of shorting them does this video is brought to you by the gamers Nexus anti-static mod mat our mod mat uses a high quality anti-static surface with a rubberized finish we also have a custom paint job on it which includes reference points and cheat sheets for PCIe eps 12 volt and other power cables along with quick reference throne paste application guides a screw sorter for your video card tear downs and it includes a common ground point and a grounding strap to help protect the products you are working on from electrostatic discharge preorder your mat now at the link in the description below so starting at the very basics the shunt resistor exists so that Nvidia can figure out how much power is being supplied to the GPU at any given time which allows them to then determine at what point it's time to start power throttling the GPU we have removed the thermal throttle scenario by modding the card to have a liquid cooler on it there are three primary candidates for clock drops or clock monitoring or mitigation on NVIDIA it's thermals we've resolved that it's voltage and then its power the last you kind of go hand-in-hand so to figure out how much power is going to the GPU and Vidya is putting an inline shunt resistor which has a known voltage going into it 12 volts coming from the PCIe connectors and it has a known resistance in this case five milli ohms if you look at the components it's written right on so what they need to know is the current and they can figure that out from all the other numbers that we've laid out so then if we short the shunts with liquid metal we are basically tricking the GPU into thinking it's pulling less power than it is there are components on the board like the AI na 3 2 2 1 which we can highlight over on this side of the board so this thing right here is the AI na 3 2 2 1 this device is responsible for monitoring the voltage drop across the shunt resistors this is a shunt resistor there are several of them on the card so what happens is when the AI na 3 2:21 detects that we are starting to near the power limit it's going to send the signals required to start power throttling the clock we're going to have lower core clocks as a result and that's what we're trying to fix so point out the shunts that we have we've got this one over here the suspicion is that this links up to this 8 pin connector and I'm going to show you how we can prove that just to be safe there's this one over here this shunt also 5 milli ohms we think talks to the 6 pin by way of proximity there's another shunt this one we know talks to the PCIe slot and then finally there's mystery shunt over here and we're not sure what this goes to and then we're going to short these two exclusively to try and trick the GPU to a point of thinking that it's facing less resistance maybe increase our power budget a little bit by doing so so let's start off by making sure that the shunts are the correct ones if you're ever planning to do a shunt mod on an expensive card and you find a tutorial on the Internet whether it comes from us build Zoid or so on random you should probably double check to make sure the information is accurate you can do this by taking a cheap digital multimeter or voltmeter and what we're gonna do is measure to determine which shunt is in the power line for which power connector so let's just go ahead and start with a scenario that we think will will not work so what we're looking for we're measuring resistance so you switch into ohms so measuring resistance between the the power line which is going to be actually we have our our GN mod mat over here showing its usefulness already so this is a plug it really quickly if you got a store that gamers Nexus dotnet slash mod mat you can pick one of these up we have them on pre-order now they're anti-static as well but it has this handy chart on it so we have an 8 pin and a 6 pin and what we need to do is measure the 12 volt lines from each against the shot you can measure either side of the shunt versus any of these yellow pins on our mat if and I can demonstrate this if you measure this pin the extra cents pin actually right here cents B if you measure this cents pin it's not gonna work it'll give you an invalid value if you measure any of the ground pins same thing so we need only the yellow and those are our 12 volt lines the reason we need these is because this 12 volt line is going to have the shunt in line with it so if we are measuring resistance from the shunt to the correct 12 volt line we should see zero ohms or very close to it zero resistance if we're measuring incorrectly or to the incorrect line or to a ground you'll see an increasing number over time a nonzero number we'll call it okay so we already know where we think each one goes to I think this goes to eight this goes to six I'm gonna measure them to the opposite of what I think just to make a point so we can measure either side of the shunt resistor again so let's just choose the sense pin so again referencing our mat this far right pin is a green pin it's sense it is not voltage so we're going to measure that verse is the shunt resistor and we should get an increasing value over time a non zero value as you can see that number is increasing right now if we measure it versus the ground or the black on the mod mat you can see that is an increasing nonzero value so then if we measure it against a 12 volt line we're gonna see either a zero value or nonzero let's measure first against the 8-pin that should be nonzero and it is so you can see that numbers going up so it this is not this shunt resistor is not in mine with that power connector let's measure against the six pin all three of the bottom row are 12 volt referencing our mod mat so we can point out one of those I mean you get a zero value now it will occasionally increase a little bit because it is one a cheap meter and two it's refreshing its measurements regularly but whenever we kind of back off and go back on to check we're gonna get a zero value or extremely close to it so this shunt then is in line with that six pin because we're getting basically zero values every time we check next thing to do is check this shunt so we've now has determined that this is goes to here this goes to here so we can short these two so here's the safety disclaimer liquid metal contains gallium gallium could potentially react poorly with ten solder contains ten these shunt resistors are soldered to the PCB with a ten solder if we put a gallium liquid metal on the top of it it is possible that if we don't protect the solder you could over time have a reaction that causes the liquid metal to eat the solder and your shunt will fall off the card so that's your disclaimer we are only doing this mod for benchmarking and it will be very brief we're not leaving this for months at a time I'm not recommending that you do it it's at your own risk so we should be fine but I wanted to point that out so to prevent any issues there you can use a nail polish around the perimeter you can use a liquid tape you can use electrical tape there's all kinds of options we're gonna put a very thin layer on here and just make sure we're protecting any components that are neighboring to prevent the liquid metal from shorting those while we are also shorting the shunt so it's just kind of paint over those now part of this process is building like I said a wall or a moat or a well around the resistor per take you below it we're gonna have this vertical and a bench that means if liquid metal drips off it's gonna drip down so this we're gonna use thermal grisly conduct a knot and I have actually some liquid metal that is still still on the q-tip from originally using it like a month ago which goes to show that this stuff doesn't actually just evaporate over a period of a month like some people say online so we're gonna use whatever's on there and if I need more I'll add it and trying to be very precise here this can be removed with rubbing alcohol pretty easily cotton ball also is very helpful in removal and we just need to bridge each side of the shunt to the other side as always with liquid metal the key is to use as little as possible especially when we're concerned about solder joints and solder health of the ten underneath it so we don't want any excess that'll drip down and cause damage and this you'll notice it doesn't have any liquidy look to it actually looks like it's going on more like a paint than anything right now like a paint pen almost so if we get kind of a zoom on that painted over shunt you can kind of look at the other ones around for comparison there's no pool of liquid metal here obviously you're gonna want to watch multiple sources if you're really doing this rather than just take it only from me but I think you'll be able to get an idea of what it looks like from a first-time perspective while also hopefully understanding why we're doing it so that should be okay at this point you can still see the five million text on it for the most part so it's not too thick of a layer if it doesn't work I'll put more on there and at this point what you need to know is what happens next so next we're going to clamp the twelve volt on each so we've got six cables to clamp and they're gonna be the lines closest to the back of the card where the backplate would be so that would be this bottom row down here is what we are clamping so we've got one two three four five six and this is a sense pen we don't want it so we're clamping these six connectors and that will tell us power consumption the next part is to run a test that has a known power consumption for us we can use x bi or firestrike and then we compare those numbers to the numbers with the shunt mod not applied with the cardstock and if in fact we are seeing higher current going into the card through PCIe that means that it worked because that's all of that happens before it ever hits the PCB so the shunts aren't doing anything yet and we're measuring only input current what's happening over here is the GPU is being tricked into thinking that it has it's actually able to pull that extra current through the headers because it thinks that it's running at a lower wattage than it is which is again just P equals VI calculation so that's what we're testing okay excuse the set layout changed it's been a few days we did some more testing on it so for the card we had up making a few changes from what was done in the first half of this video the main ones are that the third mystery shunt which builds I'd had initially said until we can figure out what it does leave it alone we end up shorting that one anyway we needed to so the first round of shorted testing done with the modded version that you saw leading up to this point did not really work it didn't do anything bad but we weren't when we tested the frequency and the current it wasn't any different at all really than a reference card we're talking tenths of an amp difference within error well with an error so it didn't work so the next step was to apply some liquid metal to the third shunt and did a quick multimeter check of that one to checking the resistance and found that it aligned to both eight pins the empty eight pin and the wired existing eight pin that's on the card so the third Gault's is two shots go to that eight pin connector so we shorted that one as well and started to see a bit of a difference it was maybe one to two amps getting to be enough of a current difference where your 1020 watts more than you used to be pulling through the cables that's still not a lot it doesn't really change anything the frequency looks about the same so after some consultation with build Zoid and others we found that the amount of liquid metal you apply to the shunt directly impacts how much you are bypassing that 5,000,000 resistance per shunt resistor so the reason for that is well there's so there's two things consider one is the more liquid metal you apply you start encountering two issues one is that it can drip and cause damage or short things on the card the other concern is that it can start eating through the solder around the edges of the shunt resistor if there's a lot of it that's kind of dripping over the edges of the shunt or just you leave it there for long enough that it will start eating through what's what's on there for the solder for the 10 so that's a concern we apply just a little bit more what we'll call a medium amount of liquid metal it's basically it's not quite a pool of liquid metal but it's a it's another layer to the surface that started showing real impact and then we did a bit more than that what we called heavy but it still wasn't a pool of liquid metal and that allowed us to go from something like 18 amps to at the high 32 amps it's a big difference there a major difference nearly double getting towards it so it works we finally were able to get a shunt mods work we're gonna go through the data now on some very quick fire strike scores using our game bench this is different from our X 299 bench that we used for the the top 10 record attempt we'll talk about that more later but we're starting with the game bench to establish a baseline of what kind of frequency difference we saw and what kind of current difference / power consumption difference we saw let's start with some fire strikes scoring just to establish the gains were seeing and we'll move on to frequency and power next with fire strike ultra our first shunt mod attempt had us going roughly the same as our normal hybrid card without the shunt mod the differences were within variance and error so non-existent and again after consulting build Zoid and several dare bower videos on this we decided to short the third shunt which got roughly the same results initially but after we added more liquid metal we finally started seeing change with a medium application of liquid metal a bit more than what we showed in the earlier half of this video but not yet a giant pool of it we were able to achieve a constant score uplift of about 0.7 percent it's not much but it was consistent repeatable and was enough to show that we were on to something adding up more liquid metal helped we boosted up to 95 18 points equating a 3 FPS gain if you were to convert it over the non shortage shunts and the result is a boost of about 2.7 percent that's enough to start budging our score is higher for firestrike competitive overclock and a testing or Hardware bought scores things like that if you're shooting for higher ranks and overclocking is worth considering it's not really worth much for actual gaming but that's just something you can see in our sniper elite scores which we'll throw on the screen now as you can see what something like Sniper Elite an actual video game that even leverages the extra clock stability more than most games would the games still aren't gigantic it's something it's enough to prove that it works but not really enough to do as a normal gaming upgrade you really want this for more competitive workloads we saw less scaling with fire strike extreme expectedly it starts becoming a CPU bench at that point but we'll look to our ex 299 bench momentarily for bigger gains what's more interesting is that the shunt mod provided enough power to our GPU to complete the times by benchmark previously without these shunts shorted we had issues with times by crashing under this plus 200 megahertz offset overclocked for the core and the HBM we could not complete the benchmark without reducing our clock offset or increasing liquid metal on the shunt resistors and here's why we're seeing those gains this chart is a frequency versus time chart you'll notice that most tests like the benchmarks without shunt shorting have some fluctuation and dithering between clock frequencies although we can hit peaks of two thousand twenty thirty two megahertz the clock won't hold those and we'll spend a lot of time ramping up and down based on boost for amateurs we've mostly solved for thermals and are operating at a steady state temperature at this point so our limiter is power the heavily shorted shunts show a nearly flatlined frequency which is exactly what you want notice that the frequency isn't any higher than the previous Peaks but is holding completely steady at nearly a perfect twenty seventeen megahertz this studying is similar to what existed during Boost 2.0 era is what contributes to our boosted scores the clock is holding rather than bouncing between 2032 and 1900 and so we get higher scores as a result out of fire strike and here's a look at wattage down the PCIe cables as a reminder power is equal to current times voltage so we're pushing up to 33 amps in some situations where we previously were stuck to about eighteen to nineteen amps with a twelve point three volt rail that puts us at 406 Watts peak down the PCIe cables versus the previous maximum of about 234 Watts that is a massive difference and is starting to exit what well is way out of range or what the card was designed for but it can handle it as long as you have cooling on the BRM so this does not account for PCIe slot power so there's a bit more power being drawn there but we didn't short that shunt so it remains baseline from before the chunk mod and for anyone worried about the V around temperatures we had that covered as in our live stream of taking the top ten slot one of them we had a 120 millimeter Corsair maglev fan blastema left half of the BRM a be quiet fan blasting from the top and a knock to a fan from the right side on the back of the PCB where we were able to cool the right half of the vrm plus the controller not only were the FETs operating within range they were operating cooler than the stock card was because of direct air flow it matters more than just a heatsink and finally here's a couple graphic scores from the X 299 bench with one of our fire strike runs we've improved a bit here as well it would help pump us higher on the hardware bot score charts if we want to play around with those again in the future at least for now until the other overclockers get their hands on the Titan V's and give us us an idea of how the improvements scale across a different platform so that's it for this one it's pretty cool the thing here to take away is that this would work on any pass gal card so this applies to a 1080 tea I just as much as it does to it 1062 a Titan V same idea the difference of course is when is it worth it and probably not very often for most of you that the risks can potentially outweigh pretty heavily the gains you're looking at you can get better gains than we did I think from what I've heard from their Bower and from Build Zoid there's potential for better gains but this is not a mod that increases your overclock offset in any meaningful way for the most part it's a mod that stabilizes the clock so that it does not choke on the power consumption to running the same high clocks it's just it flatlines instead of jumping around that's what improves your performance so it doesn't give you a ton of headroom that by nature of the mod but it gives you enough that if you're competing for overclocked scores or something like that it would be worth doing I wouldn't recommend this for a 24/7 gaming machine certainly people do it their Bower himself has talked about having 1080s setup for half a year at a time at least with the liquid metal shunt mod but he's not like every user so generally speaking I don't know that I could in good faith recommend the mod just because it's if you're not experienced with liquid metal which we've got experience for all the D lids it's easy to screw up and depend on the orientation of the card if you do a vertical GPU mount for example you're fighting gravity now so depending on how strong your fans are if you move the system around if it gets bumped or just time it's possible that you could have some drip that would short other components it's also possible that you could apply too much or too little and either it does damages the card it does nothing it might just short it and stop it from booting at which point you clean it off and you're fine there's a lot of different things here and the point isn't the scare you away from doing it it's just to say that make sure you understand why you want to do the mod and then be careful and you should be fine if you're careful about it don't rush it and be patient with the application and all that stuff make sure you test thoroughly and it should be fine it's just I no that is worth it for gaming but it's worth it to get higher scores if that's what you're going for so there's nothing wrong with that so that's all for this one as always if you wanna helps out directly you can pick up the mod mat that we used in the first half of the video this one on store like Aaron's Nexus dotnet slash mod mat it's up for pre-order now or you can go to store it at gamers access not net to have a shirt like this one or one of our other products subscribe for more I'll see you all next time