Gadgetory

this monstrosity is what got us however briefly up to the top 10 on several 3dmark world record score charts on hardware but a couple things there it's a little bit unfair because of Titan V and we have a half one we also have a 79 80 exe so looks like costs of money these days get to the top of the charts but while it is a little unfair from that aspect we were still able to get further than we should have otherwise by slapping a liquid cooler to a Titan V and now we have the thermal the voltage the power and the clock results from doing this modification 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 throw 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 we've done a few of these mods now and the reason they're cool is because they are literally cool they're cooler than the air-cooled card so this thing because it is inefficiently cooled compared to the power that it is capable of putting out is going to perform worse on charts and just to get everyone up to speed you may already know this but Nvidia is Boost 3.0 and the way Vega works with its own boost version they both have a clock dependency on thermals and then on power and voltage we can solve for one of those things exceptionally easily we can remove the cooler and put a better one on it in our initial coverage of the Titan B we actually tested this card stock versus this card stock with a 100% fan speed and saw that it actually did help quite a bit unfortunately at 100% fan speed because the Titan V is quite powerful we were still bumping up against the 83 to 84 Celsius wall before it starts to limit its clock Headroom if you talk to Nvidia they will specifically note that is not called throttling because by their definition of the term it is not throttling and instead it is a limitation of how far boost will allow you to go now for ease of terminology I am going to call that throttling because if it's not performing to its maximum potential it's just sort of easier to call it that they do have a point but I think I have a point too so in order to get past this first wall we put a liquid cooler on it next while it's power and for that one you can't do much other than short the shunts which we are doing in a separate video and you'll see more about that soon and that only helps so far as well so what we can do here today is taps the thermal difference test the frequency stability that's a big one and then the scoring difference is in addition to this we can probe various points on the back of the card which I'll show you later to determine the actual input voltages to V Corps and V HBM as opposed to what a softer reading might tell you which is often at least slightly inaccurate typically on videos locked to around one point zero nine three volts we with this mod have been able to go beyond that because we're not thermal throttling and it has some extra Headroom shunt modding theoretically might help a bit as well but that's more of a current difference than anything so let's go through the numbers the the big thing here is that it does objectively help the liquid cool these cars we've shown that now for several and video devices and and E devices including Vega Frontier Edition where interestingly we showed that you can reduce power leakage by improving the cooling efficiency doing that with this card it's the same thing by reducing the the temperature of the dye and the surrounding components like the HBM dyes we are also reducing power leakage with CPUs as an easy example every 10 degree drop in CPU temperature it gets you about 4 percent less power consumption by way of reducing power leakage similar idea applies here which means that with GPUs you can then take that extra overhead that you've reduced from leakage and put it back into the core as more efficiently utilized power to drive the core clock higher let's start out with a frequency Alyce's of the core for the Titan V for this chart we're looking at how additional thermal Headroom gives us better boost clocks and sustains them for longer under heavier loads this is over a period of roughly 25 minutes for a fire strike ultra burnin where we freeze-frame on an intensive scene we haven't applied any overclocks here these are stock settings aside from the cooler on the hybrid the air-cooled card starts off with the clock of about 1810 megahertz and rapidly drops down to a range of 1600 to 1700 megahertz dictated by thermals you'll notice that the GPU temperature sits steady at 83 to 84 degrees Celsius which is the point at which boost 3.0 encounters a thermal barrier updating charts to show the hybrid you'll notice immediately that we have a higher boost frequency which is what we want is also a direct result of the improved cooling solution we're at around 46 degrees for our GPU core temperature now as opposed to 86 degrees and we're plotting around 1800 megahertz for the core clock the air quotes card average closer to 1700 megahertz so that's a big uplift for Jessie cooler swap we're already at plus 100 megahertz and would be an even bigger gap if we had pushed below 40 C which we could do with a lower ambient temperature to illustrate how this comes into play and briefer synthetic tests this chart shows a frequency over time during a loop scripted 3dmark fire strike and time spy suite both tests have a 200 megahertz core and HBM to offset applied we even gave the air-cooled card some help primarily by setting its blower fans 100% speeds or about 60 DB of noise far more than the liquid cooled device the difference is again that the liquid cooled card manages to consistently sit at or above 2000 megahertz frequently hitting 2032 megahertz peak and the air cooled card meanwhile has harder drop-offs throughout the longer test we see this more toward the last few spikes and it tends to sit about 100 megahertz behind the liquid cooled hybrid card and again that's with its fan at 100% speed so it is both obnoxious and maxed out if we look at some results of the card with its fans and clocks restored to complete auto including the fan speed that's a bit too low by default we see even bigger performance wins the frequency fall-off as testing goes on is more noticeable and we start to lose clocks to the tune of a few hundred megahertz and worst case scenarios liquid coolant is removing that thermal barrier completely and instead of putting us up against power barriers speaking of power let's look at a voltage change this is measured by probing the capacitors on the back of the card for vcore and the probe points for VH BM this tells us the actual input voltage for the GPU rather than some inaccurate software measurement we saw a pretty big swing and voltage range after applying the hybrid mod removing the thermal cap meant that the card would permit itself to draw more power and use more voltage which we see reflected in this table of minimum and maximum measured voltages stock and with air we saw a range of 0.75 volts to 1.09 3 volts the long known max for nvidia cards default and under stock conditions HBM voltage sat at around 1.4 3 4 volts overclocking and over powering the air-cooled card got it to 0.98 volts average and that's lower than the previous one point zero nine three maximum this is again because we were bumping hard into the thermal limitations causing the card to throttle back HBM voltage was a bit higher by nature of giving more power to the HP m and also because the HBM wasn't under as much duress as the core HB m2 has low heat flux and it can deal with the heat better than the GPU core can our hybrid mod with stock clocks sat at one point zero five to one point one 6v core higher than the non overclocked air card and it also ran a VHB M of one point four four six volts overclocking pushed us as high as one point one seven with one point zero seven volts as the bottom-line v HB M ran at one point for five to one point four six in this scenario the extra voltage is a major help and explains why clock stability is so much better when removing the thermal throttle as for scores we saw improvements in fire strike ultra as listed here highlighting first the full stock air-cooled card we scored 77 48 points for graphics and with the stock clocked hybrid card we scored 81 49 that's already an increase of 5.2 percent which is remarkable considering we have only changed the cooler thus far overclocked on the air-cooled card we were at 88 35 points with 100% fan speeds or 92 65 on the hybrid mod that brings us another 4.9 percent ahead of the air-cooled overclocked fire strike Xtreme is similar and shows nearly 5% gains from the stock Titan v2 the stock hybrid card with 4% gains from the overclocked air card and overclocked hybrid card Sniper Elite 4 gives us a real-world example and shows the stock hybrid at 125 FPS average versus 115 FPS average on the air-cooled card that's a gain of eight point one percent somewhat remarkably and is a testament to how sensitive Sniper Elite is to frequency stability the overclocked card push is only marginally past where the air-cooled overclock sat and that's partially because we began bumping into CP limitations and just as a quick demo if you want to check the voltage on your own card for Nvidia at leat well for any of them but what we do is we take a multimeter and then you can stick the ground into the ground on a molex connector so this is our GN mod mat to plug that here which you can get it stored I Kara's axis Donette slash mod mat with a real molex can actually you'd ground there so you just stick it into the molex cable and let it sit and then you take the other one and you can check any of the vcore capacitors on the back of the card so we can really we can just check let's see what's a good one here we just check this one if we wanted to and you just tap that metal contact point on the side that'll give you a V core reading you could also check the HBM voltage and that would be done by using the HBM contact points over here so you just tap this one right here and then you keep this in the molex cable on your power supply and that would give you V HBM so that's how you read those in case you're curious basically contact point on one of these and then VHB em you would get over here so that's the hybrid mod once again we've learned that improving the coolant solution has a big impact on how well invidious cards do they have the clock Headroom there they can even tap into it it's just a matter of whether or not the card permits itself to under the guidance of its restrictions of thermals power and voltage so with a liquid core we've resolved part of that problem we're going to try a shunt mod next and see if shorting the shunts with liquid metal will allow us to trick the card into thinking it's drawing less current than it is because if you short the shunts theoretically it should throw the thing off a bit because the card assumes that those shunts are always 5 milliamps why would it be anything less we short them and change that it's still assuming 5 milli ohms but we might be drawing more current as a result of that assumption so hopefully we can get a bit further with it maybe not much but you're not really looking at increased overclocks here you're looking at better clock stability which is what it's all about when you're trying to improve your low frame time 99 99.9% title performance what we call 1% in point one percent lows or when you're trying to include or improve rather your fire strike scores for things like Hardware BOTS so that's the mod if you are interested in learning more hit the link in the description below for the article you can also buy our new mod map that we did all this work on at store gamers Nexus not net slash mod mad or help us out directly on patreon.com slash gamers Nexus and subscribe for more I'll see you all next time