Gadgetory

the Titan V may not be a gaming card but it gives us some insights as to how the Volta architecture could react to different games and engines as opposed to Pascal the point isn't to look at raw performance in a hundred different titles but to think about what the performance teaches us for the future potential gaming volte cards this will look at Volta architecture obviously you shouldn't be spending three thousand dollars to use a scientific card on gaming but that's what we're doing for now and we'll add more tests in the next few videos our teardown it should already be online but for now we're focusing on overclocking and framerate and then we'll move on to production power and thermal videos before that this video is brought to you by thermal Grizzly makers of the conductor hot liquid metal that we recently use to drop 20 degrees off of our temperatures thermal Grizzly also makes traditional thermal compounds we use on top of the IHS like cryo not and hydro not pastes learn more at the link below so this is it the card is in pieces we are currently in the process of probing all the different MOSFETs to figure stuff out for build Zoid for his PCB analysis and we're working on other tests as well but for now we're looking at frame rate now this card to catch everyone up to speed you can find the specs table in the article linked below we can put it on the screen as well no point in going through all that when you just read it and it's a bit boosted over-tighten XP in CUDA cores but there's some deficit in the core clock which tends to happen as you increase core count the big thing here though with Volta are with the Titan VIII specifically is that it's not a gaming card that means it's got a whole lot of stuff on the GPU that will never make it to gaming GPUs as they roll out one of those examples would be tensor cores those are utilized for deep learning machine learning type of algorithms and have no purpose in gaming so these are features of a card which will be functionally off when we're playing games and that's part of the cost that would be removed from future gaming cards and it also means that you can put other stuff there it means that you could potentially reduce the physical footprint of the GPU or that even the HBM might change in the future to a different memory type we don't know the exact details of gaming voltage cards but it gives this gives a foundation from which we can build a base understanding and start to look into Nvidia's next big architecture so let's start with overclocking notes we have an OC stepping table that we can put on the screen that shows the complete stock card starting out operating at a peak frequency of sixteen eighty two megahertz and firestrike extremes at looping stress tests with an average of 1507 megahertz after thermal limitations are applied simply increase in the fan speed to 90 percent immediately pushes us to 1605 megahertz average with no other changes and we next increase to the power target and fan speed dragging us up to sixteen seventy two megahertz average from there we incrementally stopped core offset upwards eventually encountering stability issues at around two hundred twenty five megahertz with one reboot in between due to driver crashes our final core offset was two hundred megahertz and the final HBM offset was also two hundred megahertz we left the fan at 100% speeds and were still bound by thermals sitting at around 81 to 84 degrees Celsius will be taking the more in-depth thermal measurements including vrm measurements in our next few content pieces along with power consumption but today we're looking at gaming alone this card has a lot more room for overclocking in it but we need to liquid cool it we might do things like shunt mods in the near future with some help from Build Zoid to really see what we can get out of the card right now it looks like we're running into thermal limitations first so to be very clear on how Pascal and it seems Volta both work the first thing we learned about Volta is boost functions pretty much the same way as on Pascal there may be new things in there we don't know about yet but as far as how it works with thermal and power limitations what's happening is once you exceed the magical 60 degrees Celsius number you do start dropping clocks of it now obviously you tend to always operate above that number for the most part without going liquid or crazy on your fans beads so you're not really thermal throttling and so to speak you're just not clocking as high as the thermal limitation would allow you to once you get to 80 degrees Celsius it really starts dropping clocks and 84 you basically hit a wall so that looks about the same on volt it might be a little bit lower on the be the Titan V that we have here and if it is lower its 81 or 82 degrees instead of 84 but we're basically up against the limit so we'll try and do a hybrid mod but subscribe so you can catch that when it comes up if it does when you see if we can actually adapt a cooler to it without crushing that we're cracking the HBM first because $3,000 so that'll come at the end of the process but that looks like about where we're limited is 200 megahertz core and memory it's actually a very good overclocker so far we haven't really had problems with it other than running into thermal and power limits to things that we are capable of somewhat resolving if not mostly resolving so that's something we'll be looking at shortly there will be no partner models of this particular card but presumably Volta will come to consumer and if it does we would hope to see similar overclocking behavior as we're seeing here but the clock behaves basically the same as Pascal as far as it pertains to the various boost limitations and interactions with thermal power and voltage restrictions we have no voltage control right now to speak of so short of doing shunt mods that's about where we're stuck let's look at some of the other benchmarks there so we can start with time spying synthetics the point of this is to establish a baseline for how this particular architecture behaves with very repeatable reliable synthetic benchmarks that have specific elements in their benchmarks that we can pinpoint and say the difference between these two numbers is tessellation or is ray tracing or stuff like that so we'll start with time spy as always before the test methodology and the platform we used you can check the article linked in the description below times pi gave us more trouble than fire strike or any of the games for that matter and refused to launch the second graphics benchmark at all without breeding memory back down to a 150 megahertz offset from 200 and core down to 175 everything else was fine at 200 for each times pi measured our graphics core at twelve thousand three hundred eight points for the stock Titan V plays in at twenty two percent ahead of the stock Titan XP tested again today that one was at ten thousand ninety two points overclocking the msi 1080 TI gaming x got us the closest viii at ten thousand seven hundred ninety eight points and this marks the stock Titan V as fourteen percent I had a v10 a DTI gaming X overclocked card comparing the frame rates from each of the two graphics tests will help us understand where Volta is doing better Volta holds approximately a 23% lead in graphics test one which week in the show and about a twenty two twenty one percent lead in graphics test two graphics test two has about three times as much tessellation as graphics test one and relies more heavily on volume ray casting which dynamically traces raised through the volume for each pixel graphics test two tends to be significantly more memory sensitive and will crash faster under borderline stable vm clocks this gives us a starting point for where Pascal and Volta diverge particularly when knowing the tessellation versus a ray casting differences of the two benchmarks well yeah fire strike let's start with ultra and check the graphic scores for baseline this one shows that less of a lead than x pi which uses newer lower level programming techniques than fire strike with fire strike ultra we're actually seeing our Titan XP overclocked card outperform the Titan V by about five point three percent overclocking the Titan V of course gets it ahead of the overclocked Titan XP leapfrogging it by eight point three percent this isn't all that impressive particularly considering how much better it did in times by relatively versus an overclocked 1080i gaming X the overclocked Titan VIII is about 16% ahead again for the price difference not that impressive to understand why this behavior occurs we can look at graphics 1 and graphics 2 scores once again graphics test 1 loads the GPU with Polly's and heavy tessellation it does not however apply much of a compute load graphics test 2 increases compute workloads and can stress the memory more similar to times by graphics s2 but with less tessellation focused that in times by looking at the numbers we see somewhat considerable gains in the Titan v with graphics s1 ranking at 48 fps average stock as opposed to the Titan XP is 40 FPS average stock the Titan V manages a significant 20% lead here but also falls behind in graphics test 2 the V is at 26 FPS versus 26 FPS from the stock Titan XP and graphics s2 and overclocking each card gets them both 229 FPS for graphics s2 despite a considerable 22% lead in GT 1 for the Titan VIII the 55 versus 45 FPS numbers for example it is tied or slightly behind in graphics test 2 in the worst case scenarios and a quick shout out to the crossfire Vega 64 is there showing that times by fire strike and 3d mark in general still care a lot about multi-gpu even if a lot of the games out there nest don't necessarily show the same level of near doubling and performance but looking at all this data it appears that the Titan VIII has stronger potential in tessellation and geometry heavy scenes and even with a great amount of geometric complexity and it might also be somewhat memory bound in gt2 but we're not 100% positive on that right now the advantage in times by suggest that there's improved asynchronous compute performance with the Titan VIII over the previous Pascal cards and this is something that we can look into pretty easily by using lower level API games like doom Sniper Elite 4 and potentially ashes of the singularity all of which use either DirectX 12 or Vulcan and let's start that off with doom with doom at 4k and with asynchronous compute enabled the stock Titan V places at 132 FPS average for average frame rate alone we're about 41% ahead of the Titan XP stock GPU overclocking the Titan XP actually let's give it some help and defer to the overclocks hybrid XP gets it to 113 FPS average this allows the Titan via stock card to hold a 17 percent lead and we did actually retest doom and saw pretty much the same numbers as we've seen for the last couple of rounds of tasks for the Titan XP so no improvement there to speak of over the last few months the stock Titan XP is largely choking on its cooling and for starters it's also limited in power for a restricted performance overall either way once we account for a Titan V overclocked we're at 157 FPS average versus 113 FPS average or back to 8 39.5% lead for the Titan XP scallion is pretty linear between stock and overclocked tests and the Titan V does lag behind in one key area that's frame times we think part of this has to do with drivers in fact if you look at our older Titan XP data from around May you'll notice that it's low frame time performance was actually a bit better as was also the case for the older 1080i data looking at modern data sets with the Titan V and the Titan XP that were both tested again today we see that they're closer in lows than they are to their older test data like the Titan XP from previously the big takeaway here as indicated by times by is that the volta card seems to have improved performance specifically an asynchronous compute titles at least this one doom with Vulcan and we can dig into that theory further with sniper and then reinforce it with D 3 D 11 titles as for the frame time differences that looks like it's more of a driver thing overall or potentially a change in the dooms offer between the original tests and the more recent ones Sniper Elite 4 is our next title run at 4k with high settings DirectX 12 and asynchronous compute enabled stock the Titan VIII operates at 115 FPS average with lows at 97 and 91 for 1% and 0.1% lowers respectively this place is at about 27% I had an average FPS of the Titan XP stock GPU posting massive leads over the predecessor this is after retesting the Titan XP with the latest drivers as well so what we're seeing is a legitimate performance uplift in the Titan v despite its clock deficit before digging into the rest of that chart some more let's analyze some of the data we have now what we've learned in the past was Sniper Elite is that it's particularly shader intensive on both Andy and NVIDIA Sniper Elite is one of the few titles where when overclocked equally the vega 56 and vega 64 cards show a bit more of a difference than we see elsewhere it's also one of the few titles where when you clock them similarly ignoring the memory differences in latency the GTX 1080 and 1070 Ti achieved pretty similar performance in basically every title except for sniper and doom it to some extent because of that difference in shader reliance so sniper actually likes those extra shaders from the 1080 over the 1070 ti ignore the memory latency aspect of that conversation and it also likes the extra shaders from the Vegas 64 over the Vega 56 now this would then follow to the Titan V which we're seeing significant uplift on today and that seems to be a shader difference potentially there's a whole lot of other differences there we can't account for all of them yet it's too early but that gives us a baseline so it makes sense the game is built to asynchronously cue render jobs into different shaders so having more shaders means that you can have more simultaneous render jobs in flight actively so that would follow that more shaders would help with your framerate in this particular instance now sniper is also pretty sensitive to clocks but the difference between what it cares more about is kind of nebulous it depends on the architecture it depends on if it's AMD or Nvidia there's a lot of variables there but generally this game unlike a lot of the others is extremely sensitive to pretty much any change in a GPU except perhaps gddr5 memory speeds back to the Sniper Elite 4 chart for a moment overclocking Titan v2 it's 200 megahertz offset gets us 41% ahead of our overclocked Titan XP hybrid card this is a tremendous gain and shows that the card becomes more constrained by its stock clocks in this particular title and a takeaway is that the extra shaders help but they need to be fed with higher frequencies to really engage Foley for reference we'll highlight a 1080 TI card they're not too far behind all things considered and obviously make far more sense as an affordable gaming purchase that still permits max settings ashes of the singularity is our final non DirectX 11 title this one poets a problem for us we didn't plan to ever test cards of this caliber and so our standard 4k test with high graphics immediately demonstrated a CPU bottleneck as you can see here all the cards basically equalized around 97 to 99 FPS average so we cannot draw any conclusions as we're now bumping against external limits and the GPS aren't fully engaged for this reason we reran a few tests with 4k crazy settings and applied an 8x MSAA option to force load on the GPU so with 8 tab MSAA we get down to a point where the numbers are now comparable we don't have as many numbers for this as it was but you get the idea the Titan V operates are at around 82 FPS average when overclocked a lead of 9.5% over its stock configuration against a Titan XP the stock Titan V has a lead of about 10.2 percent put a seventy four point eight five to sixty seven point nine FPS average the low one percent and 0.1% low values represent frame time consistency and are also close by this number sounds a lot more down to earth than previously and demonstrates the extent to which game development impacts scoring of this class of card we can extrapolate wide-reaching results from the other titles across all titles in gaming but ashes is DirectX 12 and it does not leverage the Titan V to the same extent as one of the other DirectX 12 titles Sniper Elite for either that or it doesn't become choked on the Titan XP pipeline enough to demonstrate those Delta's moving on to another chart of limited tests we have hell-blade which is our Unreal Engine representative hell-blade has our Titan via stock art at 67 FPS average while our overclocked variant operates at 74 FPS average for an improvement of 10.5% over stock versus the Titan XP we're at 49 FPS average stock and 54 FPS overclocked putting us at 37% improved when comparing only the Titan V and a Titan XP numbers our improvement is 37% stocks of stock these gains are larger than we'd anticipated considering that this is a DirectX 11 title and several key functions of the Titan VIII are unutilized in games let's look at Ghost Recon for another one at 4k and very high settings the Titan vgpu operates at 62 FPS average with lows at 54 and 50 this is a stark difference from the last few games the 1080i GPUs have no trouble keeping up with the Titan B stock GPU at all and the overclocks 1080 T is e2 somewhat proves that point by achieving parity the $750 card shows the importance of GPUs targeted purely at gaming particularly when matched against its $3,000 scientific targeted counterpart against the Titan XP we're still looking at a difference of about 1 to 2 FPS between the Titan v and XP so nothing exciting there overclocking the Titan V gets us a bit beyond the XP overclocked about 6% ahead of the overclocked hybrid XP performance and demonstrates what's lacking Ghost Recon seems to want higher clocks and the additional shaders just aren't helping this is why it's important to have a wide sweep of games rather than just all games and then talk about what drives those differences or at least what we think drives those differences there's a lot to know here and we haven't figured it all out yet Ghost Recon isn't more old-school in its DirectX 11 implementation if for Honor's shows us similar results we can start building a case about certain DirectX 11 titles and Titan performance at least as it pertains to Ubisoft 1440p and 1080p results will be in the article below and here's for honor also made by Ubisoft at 4k the stock Titan VIII has us at 83 FPS average but with lows ranging somewhat sporadically from 29 0.1% to 40 FPS 0.1% the 1% lows range from 57 to 74 this range is significantly wider than we see on other cards and potentially indicates driver level frame pacing issues with the drivers the Titan VIII still manages to chart top but it's only a couple FPS ahead of overclocked 1080 TI cards versus the stock Titan XP the Titan VIII is about 14% ahead so it's a lot closer than in some of the other titles though the lows are somewhat difficult to compare given their seeming randomness moving on to destiny - for another DirectX 11 title and starting with 4k high settings we get another scenario where the Titan V stock GPU only marginally outpaces the Titan XP GPU the difference is about 4.1 percent with the XP showing significantly stronger frames on consistency than the Titan V we're not presently sure if this is a driver optimization layer problem or a fundamental behavior when deploying a scientific card in a gaming scenario here's a look at a frame time plot between the Titan XP and a Titan V just to give an idea of what those frame times look like when realized over time as they should be the lower numbers here are better but you also want more consistency so it's not just strictly lower is better it's more consistent and lower is better moving on we look at the overclocking version of the Titan V which got an FPS of 125 average in the same tests provided a substantial 21% lead over stocke excitin b-but the card still struggles with frame time consistency performance is comparable to the Titan XP and 1080 Iowan stock minimally and illustrates there are low-level API gains that stop with a title like destiny - at 4k and highest settings we only have a few cards present as the new destiny - Nvidia drivers wiped out our charts because we needed to retest everything the Titan be tested at 88 FPS average with the overclocked granting an 18% uplift still significant it would appear that destiny cares at least a little bit about clocks whereas some of the other titles we looked at don't necessarily care as much and might like shaders more the Titan VIII also carries a 13.6% lead over the Titan XP stock GPU went under the highest settings and what this illustrates for us is that as we move towards certain types of titles like Ghost Recon and for honor we start seeing performance that's a lot closer the bigger gaps for the Titan V and the Titan XP at this point were around 14 15 percent somewhere in that range so being that that's the widest we see obviously that's not extremely exciting for $3,000 card to be fair it's not really meant for that type of task but the idea is less of what it's meant for and more of what we're seeing in the other games which is things that have a sink compute things that use lower level api's and some of that could be just coincidence that the game developers focus more on optimization in general we know for example that in software put a tremendous amount of effort into building Doom to work extremely well with just about everything so the fact that we're seeing big performance gains there it's kind of hard to pull from that just how much does that apply to other Vulcan titles and there aren't a lot of Vulcan titles really test there's Talos but no one plays that and it's pretty old at this point anyway so we're left wondering at that point Sniper Elite 4 shows an asynchronous compute direct x12 implementation that benefits the Titan VIII pretty heavily but then again we look at the Ubisoft titles and you encounter questions of but how does it look with DirectX 11 ashes of the singularity which is a long-standing DirectX 12 benchmark that's had plenty of time to be optimized on all drivers at this point so is pretty close performance sometimes 10 to 13 percent between the V and the XP which again brings us back down to earth of these aren't so exciting gains but then again you look at the gains that are huge and they are kind of exciting at least from a numeric value different standpoint so that's what we have for the Titan V so far basically if we're to try and learn from this data given that it's not something probably any of you or if any then not many of you will buy a given that parameter the takeaway really is just that it looks like it's worth keeping in the ion Volta specifically as it pertains to things like async compute implementations and low-level API performance this is a key area where NVIDIA has not increased its gains as significantly as it's been able to with DirectX 11 optimizations driver releases and a long-standing code base of DirectX 11 support so given that Titan V so far is looking pretty good in those low-level API element ations with the V which is admittedly a brute force solution there might be a change there for volta versus Pascal and Maxwell which although not necessarily bad in those implementations were certainly not that impressive as compared to their relative DirectX 11 performance where NVIDIA has spent a lot of time and money building their codebase drivers and architecture to work well so this might be a point where Nvidia starts tipping the scales towards the lower level api's if that's true and we're not saying that necessarily as we need to see gaming cards if that's true it'll come down to developers of course actually doing more with these low-level API is but be careful how I've already take that because it's basically speculation based upon some testing of a scientific class card that has an insane amounts of cuda cores and is not built for this particular task so we can only take that so far but take it for what it is it's an economic experiment basically the high-end card you could gain with it if you wanted to it's just not meant for that so 1080 Ti is still the clear winner if you're trying to pick it for a high-end gaming device but we'll see what both it does when it comes to gaming so that's all for now check back soon for our production test which is more what the cards meant for and we're going to be doing a lot with power thermals overclocking modding as well so you'll want to subscribe for all of that you can go to store gamers Nexus not in that slash mod mat if you like the mat we have on the table here and I would like to pick one up it's on pre-order right now and is brand new and if you would rather support us in patreon you go to patreon.com subscribe for more I'll see you all next time this mattes pretty cool