Gadgetory

just a heads up you won't get +200 on the core well we did the Titan V was actually one of the better overclocking cards we've worked with this year the biggest problem with though is thermals and then to a lesser extent although not that much power so this is where we're focusing on the power consumption of the card the thermals of the MOSFETs the GPU and the thermal performance of this cooler which has some slight modifications over the Titan XP and we're gonna look at how much room we have for improvement when moving to our hybrid mod which is hopefully coming up soon before we get into that this content is brought to you by the thermal take flow RGB closed-loop liquid cooler which is a three hundred sixty millimeter radiator plus three one twenty fans that are RGB illuminated the if then we'll take it rain fans at that this is a four point five done a detect pump which is one of the faster pumps you can learn more at the link in the description below so some quick errata from the teardown video after posting our teardown video where I basically said the vapor chamber in here is pretty much the same as a Titan XP vapor chamber sends a couple of small fin differences and differences in how the rear end of the vapor chamber kind of swoops up on the side the correction received was that Nvidia basically changed the fin stack to be copper as well so it's not just the cold plate of the vapor chamber it's also the fins now that's not that relevant as you're gonna see in a moment because it's still this cooler on a high wattage card and it's power-hungry card it drives a lot of heat so it's kind of nebulous how helpful that change is considering that just moving to a completely different cooler altogether would probably be the better move in terms of peak performance but NVIDIA doesn't tend to want to do that for their Titan Class cards which is potentially some mixture of the audience not carrying or maybe they don't have the noise concerns or they put them in rack mount boxes or something here's what I want to know if you're someone who buys this or the Titan XP or any card like that that is a potentially targeted at workstation deep learning machine learning let me know below if there's you would like to keep this cooler as opposed to having a better one on there that might be dual axial for example if you just have two fans blowing down onto it in a traditional open-faced cooler design let me know if there's a reason for that I suppose exhausting heat out of a box but if you're doing it at the expense of choking or GPU significantly which worried about the show then is it really worth it that's what I want to know so we're looking at this card the cooler and the thermal the sort of results that we got for everything and we're gonna start with some clock versus temperature performance it will give us an understanding of how the clock behaves on the Titan v card including how it down clocks or how it attempts to regulate itself to control temperatures prior to diving into thermals we can start by looking at baseline performance with Auto settings and then our overclock and then we'll look at where the thermal limitations are being encountered this chart shows a frequency over time during an automated run of firestrike ultra extreme and normal followed by time spy at stock auto settings the Titan V is operating in a peak clock of about 1770 megahertz and gradually diminishes throughout each test pass which you can see after the gaps where it falls down to an idle rate if we begin to plot core temperature which is the same benchmark you'll notice that our peaks to 84 degrees drop clocks almost immediately and what appears to be inversely proportional to temperature rise not exactly how it works but you get the idea these tests aren't even that long they're less than a minute each in most cases and we're still slamming against the 84 degree while that Pascal and Volta carry the stock cooler is incapable of keeping up with the power load generated by the card when left to self-regulate let's manually impose a 100% fan speed for the next round so this chart shows the complete stock settings except with a fan boosted to max speed this is primarily to understand performance and is not sustainable in any real environment as noise output for this is at around 60 DBA still looking at frequency we see that between auto and 100% fan curves with no overclocks at all the frequency picks up considerably in a few of these tests we also see differences of up to 100 megahertz and a bit beyond in some instances that's a lot of performance left on the table just because we're using this bad cooler on the Titan V these data points illustrate that we are throttling hard on thermals well before we run into power limits but those are the next limitation as for how hard that impacts performance here are the fire strike ultra scores for the auto card leave 100% fans beat card and the overclocked card the stock card pushed a graphic score of 77 48 points while the stock card with a 100% fan speed gave us a couple percent boost about two percent our fully overclocked card is well beyond both of these numbers and part of that is just because we increase the power target as well so running into thermal limits giving us a two percent reduction and then running into power limits fire strike extreme has the difference between the stock and 100% speed tests at 2.7 percent I've seen here and so the scaling changes based upon what test you're doing this next chart shows our overclocked performance with a two hundred megahertz core and HBM overclock our court now pushes toward two gigahertz at times this is compared to the previous clocks that were 300 megahertz lower in the worst cases this performance disparity is from three different factors we've increased the power budget eliminating that concern and have increased the fan speeds 100% eliminating the thermal concern we've also manually overclocked the card and all three of these produced the chart-topping performance numbers that we showed in our previous gaming benchmark video for the Titan V looking at thermals the fan speed increase helps prolong the time window before reaching the clock limiters at 84 degrees and Beyond still towards the end of the longer test we were getting up to around 87 degrees resulting in clock drops over the duration of that test at 120 percent power target though the temperature target changes to 89 degrees if you just allow the slider to match with the power target just to demonstrate the previous generation this chart shows stock Titan XP vs. stock Titan V scores the Titan XP holds a higher clock when both are left to self regulate but it's still beaten in most tests by the Titan V you can learn more about that in our again previous Titan be gaming benchmarks video this helps illustrate that the core count increase negatively impacts maximum stock clocks that's not to anyone but is made up for in benchmarks that can actually leverage those cores times pi is a good example and one which leverages lower level programming to distribute load more evenly across additional cores to keep more simultaneous in-flight instructions going to all of these shaders let's move on to component temperatures the chart shows the GPU temperature and two MOSFET case temperatures measured by thermocouples that we mounted to various locations on the card the left side Center MOSFET runs warmest at sixty seven point seven degrees with the right side middle MOSFET at fifty one point nine degrees Celsius both of these values are well within spec these parts can take 125 degrees plus without issue or without much of an issue other than D rating and some life span hid and this follows the trend of nvidia founders cards typically having more than adequate cooling for VRMs despite these somewhat awful cooling for the GPU itself they do actually do a decent job at keeping the prm's cooled part of this is the selection of the V RMS and the power that the power stages can handle we've seen this previously on reference 10 series GPUs and the MOSFET temperatures here are completely controlled the GPU however isn't we're bumping against 84 degrees frequently which means clock regulation over time this frequency chart from our 30 minute fire strike burnin shows rapid clock degradation upon hitting the 84 degree wall where the cards at stock configuration automatically regulates its clock speeds this brings us down from 1837 megahertz to 1702 megahertz and is another demonstration of why Titan V could be so much more powerful if they just put a better cooler on it this is sort of a repeat or a broken record of what we said about Titan XP when we did the hybrid model on that we looking into that more shortly well you're not noise normalize temperatures is almost pointless as the Titan V just won't be able to compete with AIB partner models of lower end hardware even though lower end in this instance is a relative reference to the 1080 Ti still if you wanted to keep a 40 decibel operating noise level the guard would throttle down heavily and operate with a GPU core temperature of 90 degrees plus with MOSFET hot spot temperatures of 71 degrees plus and that's in open-air with a case you'd be in worse shape for almost all instances the MOSFETs are still fully within reason but the core is throttling us hard this card runs hot and also loud with its 61 decibel max output at 100% which is really barely enough to sustain the overclock as we had it configured with the 40 DBA tests it helps to know our noise levels the Titan V operates similar noise levels to other Nvidia reference GPUs we're measuring about 31 DBA Idol with the average fan speed under Otto conditions placing us at around 48 DBA and going to 100% speed has that 61 overall it's quieter than a reference Vega cooler but it's still ultimately inadequate as a cooling solution 44% puts us around 40 DBA just for reference to the previous charts this set of charts will show a total system power consumption went under gaming workloads well start with 3dmark fire strike in this test the Titan V stock card is drawing 350 watts from the wall for the entire system comparing this to its neighbors with the same system and power supply we're at 345 watts on the Titan XP 381 on the Vega frontier edition air card and 347 on the stock EVGA tenet ET ISC to the overclocked Titan V starts really pulling down power here pushing up to 440 to watch total systems drawn fire strike this puts us on par with our power play table modded Vega 56 with a liquid cooler which consumes 447 watts for the system though we've later pushed that card up to more than 400 watts on its own measured with a clamp on the 12-volt rails the difference is that we had a power target of 200% offset on the modern Vega card and part of this inefficiency on the Titan V likely comes down to Volta not being a gaming targeted architecture with all these components on the die that go unused when gaming or benchmarking fire strike moving on to Ghost Recon wildlands the Titan VIII system pulls 388 watch one stock with a Titan XP at about 375 watt stock remember this is total system power draw not clamped draw for a neighbor comparison the 1080 Ti demonstrates its performance efficiency at 370 watts for the EVGA sc2 and overclocking our Titan VIII without any mods to it at all gets it to around 420 Watts right around where our overclocked Titan XP landed the power modded Vega 56 card is the most power hungry here at 476 watts for the system as compared to its stock 332 while total system seongjun the titan be overclocked system is Pauline 8.3% more power than the stock card idle power consumption has our complete system at around 80 watts with the card just draw in enough power for the fan and some signaling that's pushing less than a couple of watts through the 12-volt rails going as the PCIe connectors so the takeaway here is a repeat of what we said about the Titan XP takeaway basically ignoring the cards actual performance capabilities and they were impressive in some instances specifically asynchronous compute workloads DirectX 12 Vulcan low-level API is ignoring those and looking just at the thermals it's severely limited with the current cooler so it's possible that a company like EVGA could put out some kind of aftermarket hybrid solution for example which would solve those problems you'd have to buy it and assemble it yourself though because Nvidia is the only supplier for this card so there will be no AIB partner models unless you're buying aftermarket coolers and adding them and if you're interested in that let them know because they're probably not gonna make it without significant interest any of these companies there's a three thousand dollar card and how many people buy that and then mod it is the question we're gonna be one of those groups of people though we are modding it hopefully to be a hybrid card which means we're gonna add a liquid cooler to it now the big challenge here is that the mounting hole spacing is different from what we're used to I don't think it's even the same mounting hole spacing as the Vega Frontier addition card and so we're gonna have to get out a drill and make a plate ourselves to solve that problem you also have issues where the HBM is very fragile and so with the wrong mounting pressure it's pretty easy to crack those dies you have to start with just enough pressure to get contact and make them stop moving around on the plate and then go from there but the cooler itself stock is not impressive it's actually pretty bad these blower coolers including the one that it's on top of they're all pretty bad compared to anything that you'd want for actual performance out of the car because it's dropping clocks it's like a hundred megahertz off the top just out of the box basically it's got all this performance on the table you don't even have to overclock to get it you don't need to go into precision and set at 200 yards offset it's just not necessary to get that extra speed because the core the GPU itself under boost 3.0 has the capability to push up to a limit there are three limits with booster you point out there's power which we're not hitting there's voltage which we're not hitting there's thermals which we are heading at 84 degrees so with a better cooler just let's just play pretend and say there's a water cooler on here out of the box you put it in the system it's gonna run about a hundred to 100 70 megahertz faster depending on what kind of scenario you're in with a stock card how bad your case is how bad your room ambient temperature is all that stuff that's a lot of performance you can get back for basically nothing and then once you go beyond that it gets the overclocking and overclocking this card it's actually going to the positives for a second it's a very good overclocker we were pretty impressed with it and after some mods shunt mods or liquid cooling mods we expect it'll do a whole lot better but with a better cooler on there to start man it be it just be a lot more impressive of card cuz just in the gaming results alone ignoring anything that is bills for in the compute world the gaming results alone with these bigger overclocks on the stock cooler we were seeing gains that were sometimes 20 22 percent higher than the stock card which is kind of insane in a good way so it's got a whole lot of room here it's really easy to push the card pretty far 200 megahertz core and HBM and it's just being choked by this cooler which is unfortunate and it comes back to the question at the beginning of the video when do you want this cooler as a legitimate user of this card because I can think of a few core reasons for cards that aren't this like let's say the 10 series the 10 series you might want to cool it like this or Vega 56 cooler if you are an SI and you want something that you can roll out easily repair and service easily and you know will work in poorly maintained environments which would be most si customers probably that would be one reason another reason would be if you're using a small box and you need to exhaust heat out of the entire box like if you have a radiator that's front mounted and pulling in hot air you need to get that out of the case you can suck it through the video card video cards not gonna be happy with you but it be great way to get air out of the case that's already been warmed so those are some legitimate reasons do either of those or does the latter one apply to an actual user of this card though is what I want to know and I don't know that we'll get many answers because I would doubt that many people are buying one of these but if you were a Titan XP purchaser the same question applies to you as enthusiasts though and this card isn't really meant for you either but as enthusiasts looking at this type of cooler in general yeah they tend to be pretty disappointing fortunately there are aftermarket mods we can do is just gonna take some work and we'll have a video series on that so subscribe for that as always you can go to store that gamers nexus net slash mod mat to preorder our brand-new mod mat that we've been working on developing or you can go to patreon.com/scishow since I was out for smaller contributions subscribe for more thank you for watching I'll see you all next time