Gadgetory

today we're reviewing the nvidia titan r-tx for overclocking gaming thermals and some acoustic performance looking at the first of the two cards in the lab we have a third card that just arrived to trade off for one of our defective units which was stuck to 1350 megahertz as you saw in another video so we'll be looking at SLI after this the Titan r-tx cost $2,500 outbidding the r-tx 2080 TI by about two times and being about 500 dollars cheaper than the previous Titan V flagship in this class it also only enables an additional for streaming multiprocessors with for more SMS and 256 more lanes there's not much performance to be gained in gaming scenarios from that change so the big gains are in memory bound applications as the Titan RT X has 24 gigabytes of gddr5 from the 11 gigabytes on an RT X 20 atti today though we are going to focus on overclocking power thermals and gaming and then we'll look into production applications later before that this video is brought to you by us and the gamers access store you can go to store dock cameras nexus net to pick up one of our ceramic mugs critically-acclaimed mod mats or educational video card tear down and PCB anatomy posters that teach the names and placements of all the key PCB components learn more at store des cameras XS net or click the link below getting into these really briefly then as noted in the intro the 24 gigabyte frame buffers it's something that benefits Adobe premier for example where when we edit our chart the charts that you see in this video that are lines over time those eat a ton of memory the way that we do them and because it's all cuda accelerated playback it happens to fall on the GPU and so there are times where we'll hit maximum memory allowance on the card 11 gigabytes or whatever it may be and run into issues so this is something that would maybe help in those scenarios it's also something that would help in blender for example where if you're doing blender one of the best reasons to not use CUDA acceleration in blender is if you run out of memory on the GPU if you have a project that is particularly intensive on memory and we've had them because our lead video editor does blender animations for us as well sometimes we have to push to the CPU because you can fit more memory there given the way CPU and memory configurations work so those would be the reasons why you would want something like this as opposed to a 20 80 TI and we're gonna put off production testing until our TX is fully enabled in applications like blender or v-ray or whatever else there may be out there once they're fully enable then our TX actually does something unique and those applications will go through the tasking but today we are going to demonstrate gaming performance which is more or less talking to the users who take the I just want the best approach because you should still know you're getting so we've got sli 2080 TI is in here and b link is sli it's still using AFR and we're testing that versus the titan RTX we have every other device that's come out this year a few notes a few places where things have been retested because far cry for example had some regression on performance but we'll talk about that when we get to it other than that we had one device stock up 1350 megahertz we just had an Nvidia ref visit us locally fortunately one is in the area and we traded him a our 13 50 megahertz unit so they can try and diagnose it get to the bottom of the issue and we took a new one that's working fine and we'll continue with our SLI testing from there so you're really paying for memory here and that memory is best applied towards things that honestly we're not testing today but there's still a lot of people who want to buy these things for gaming and hopefully this will help those people figure out if it actually makes sense for you so let's get into it overclock stepping comes up first you'll need to know this information to go with our gaming benchmarks later on the Titan RT X overclocking process follows the same steps to the xx atti and becomes instantly throttled by its cooler under the out-of-the-box fan configuration built in to be BIOS for the fan curves under complete stock settings we saw an average frequency of about 1800 megahertz and times by extreme with a 19 20 megahertz peak frequency power measured about 280 watts via gpu-z that we have separate power measurements later for some hardware measurements the fans seem is to stick around 1515 rpm to maintain the thermal target of 75 degrees in this configuration maximum power target offset is only 14% extra so 114 percent total leaving us with a limited amount of room to increase performance in fact you'd have more room with some of the high-end over in 2080 TI's and just a sub mall side note here percentage is a percentage of a baseline so if a card is 114 and another one 125 percent they could actually be the same wattage like the same maximum power and watt with different percentages to keep that in mind but anyway we ended up at 18 30 megahertz average up 30 megahertz and that's without any core frequency offsets we were also instantly throttling at 88 degrees causing the clock to bounce around between 1780 megahertz and 1850 mega Hertz once we started actually applying offsets like + 102 core for example we just started throttling a few steps later we blasted the fan speed the 3700 rpm which we can highlight and that's the maximum speed and we saw a performance uplift in a significant fashion right Lisa frequency at maximum frequency was 20 85 when around 62 degrees Celsius and finally the resting point at 20 40 megahertz peak under sixty six degrees was where we sat for the core the trouble is that just like the previous r-tx cards we've become bound by an overprotective power target and voltage limitations and then some of the recall or limitations if you want better noise levels we might try and mod V bios to see if we can surpass this or might just rip one of the shunt resistors off the board and do build toy's mod but stay tuned for that anyway that's our overclock settings for the tests overclocking is pretty limited on this one and that's entirely because of the power limit and gbz was reading power draw at 330 Watts when overclocked but the vrm can handle far more than this as our twenty atti reference PCB analysis demonstrates and this is the same PCB as that it's the same card build joy to analyze for the 2080 TI fu launch moving on to game benchmarking next our game benchmarks will start with Sniper Elite 4 which is one of the best-built games of the modern API using DirectX 12 async compute and 4k high settings that Nvidia Titan RT X ends up at 112 FPS average with lows at 90 fps and 87 fps 1% pause your point one percent low respectively for a comparison 1080p is an SLI do about 170 FPS average or SLI 2080 TI is do with 210 FPS average we also observed the 20 80 TI at about 180 FPS average establishing a difference of about 4 FPS or if you want to do it this way eight point nine six milliseconds our training times for the Titan versus nine point two milliseconds for the 2080 ti we can't think of many humans if any who can identify a 240 microsecond difference in frames a frame interval overclocked in a Titan r-tx to about 2040 megahertz core gets its 126 FPS average outperforming the overclocked 20 atti by about by about 3% the next thing we need to test is mblink Titan RT X's which will work on immediately after this though see our other video for why we got delayed because one was stuck at 1350 megahertz here's a look at frame x between the stock Titan RT ax and stock 28 ET i cards as a reminder frame times of the most accurate representation of friend frame pacing or the interval of time from one frame to the next it's also the best way to objectively illustrate the raw experience without averaging both cards perform roughly the same lower is better and more consistent is best but these cards are both low in frame time and consistent in frame time and NVIDIA has done well with frame time consistency and pacing for this generation for the most part and neither of these cards experience a noteworthy frame time spike or hey it would be difficult to tell the two apart for reference sixteen point six six seven milliseconds is 60 FPS with a two milliseconds at about 120 fps what we care about here is that there's never more than an 8 millisecond of 12 millisecond deviation from the mean so the user is unlikely to detect any hitching or interruption of fluidity some people call this smoothness overclocking the Titan RT X introduces more frame time variants as you'll see in this line plot but nothing severe regularly hitting three millisecond frame time increases over baseline from seven to ten in some cases but this is still relatively fluid and basically unnoticeable to the user F 1 2018 give us a look at the ego engine with DirectX 11 moving back to the more widespread API and away from the shining example of DX 12 that is Sniper Elite 4 for f1 2018 at 4k and ultra high were clearly GPU bound with high ceilings set by the CPU and our CPU reviews you'll see FPS as high as 300 with the right CPU so this is a good means that really tests the limits of the GPU and avoid limiting influence from the CPU this further is illustrated by the sli 2080 TI is at 169 FPS average clearly pulling away from the rest of the pack after the Titan RT X that ends about 110 FPS average stock placement just ahead of 2080 TI f e is 99 FPS average or the 28 e TI XC ultras 105 FPS average maximally the gap is no greater than 10% and that reduces to about 5% with a heavy partner model 20 atti overclocking the Titan gets it to 118 FPS average just ahead of the overclock 2080 TI fe at 114 FPS average at 1440p the overclock Titan RT X runs at 185 FPS average and with lows at 87 fps 1% putting it just above the 2080 TI fe hybrid overclocked card by about 3% the Titan RT x card hits 175 PS average one stock sandwich a net between the overclocked 2080 TI and stock 2080 TI x ii ultra at 170 FPS average or stock for the 2080 TI fe @ 1 60 FPS average differences are minimal which is probably expected when considering that the Titan RT acts only ads for SMS the GPU moving to 72 SMS from 68 SMS so the lacking performance makes sense the VM doesn't really help in games which is something we all pretty much knew at this point anyway at 1080 P we clearly slam into a bottleneck at about 239 PS average this is becoming CPU bound as a scenario and so the results here are relatively meaningless all we learn is that the CPU can't keep up with these cards at 1080p no big surprise there shadow of the Tomb Raider is a DirectX 12 title that's still relatively recent for this one note that the sli tests were conducted without anti-aliasing as TAA causes issues with AFR or alternate frame rendering as a result of the lack of previous frame data between the cards the Titan RT x ends up at 72 FPS average for stock performance which isn't much better than the 2080 ti of e to 67 FPS average the 7.6% uplift here is undesirable in the face of a 100% price increase but note again that the bigger reason to buy this card is for its vram increase not for any other reason that all most likely benefit non gaming applications overclocked and Titan r-tx puts it at 70 FPS average which is about the same as overclocking at 28 e TI card although our 28 ET i was able to push higher and core clock this is part silicon lottery and likely part thermal density as well 1440p the Titan RT X runs at 119 FPS average stock which is functionally tied with a 28 e TI f e unconstrained thermally by way of hybrid mod the 28 e TI XE ultra sits at 116 FPS average with a fe at 114 FPS average overclocked inputs the 28 e TI in the lead due to higher clocks again with the Titan RT acts just behind it 128 to 130 FPS average not far from margin of error at 1080p the cards that still show scaling and haven't run into a hard limit yet but we do see the Titan RT acts in 2080 TI XE ultra and 2080 TI fe for that matter are all about the same performance level there's functionally no difference between these devices GTA 5 gives us an interesting spin on results for this one test adverse at 4k the Titan RT X runs at 93 FPS average outperforming both the stock and overclocks 2080 TI fe card the 2080 ti XE ultra ends up at 88 FPS average and would surpass the Titan RT X with an overclock but overclocking the Titan RT X gets it to 98 FPS average still outperformed by the dual GPU classes above it although the Titan card looks better here than in some games it is still clearly not worthwhile for purely gaming use cases they can do them just fine the best in fact but if you're only gaming there's not a lot of value for you at 1440p the Titan RT ax ends up near the top at 156 FPS average we're beginning to become bound by the game engine this engine frame caps at 180 7.5 FPS making it impossible to see how much a more Headroom there is for the sli cards as an example still the Titan RT x card stock ranks about even with an RT X 2080 tion overclocked establishing no meaningful difference we have 1080p results we'll put them on the screen but at CPU bound to a point that everything becomes equal there's nothing to learn here so let's move on some recent game updates made it apparent that it was worth retesting in Far Cry 5 so we reran the 2080 TI fe numbers for this one the Titan RT X ends up at 70 FPS average with the overclocked variants gaining 5 percent performance 273 FPS average the 20 ATT i retested lower than its original results from a few months ago placing instead at 66 FPS average with the XE ultra also tested freshly at 67 FPS average the Titan RT X ends up ahead at about 6 to 7 percent of a lead when all are retested on these drivers and with this newest game update something here seemed to affect the performance of the 20 atti so Titan RT ax does end up in the lead although original testing would have put the 20 atti in the lead it's just that's no longer the case with current game version and other updates when both are stock 1440p scaling places the Titan RT acts as 4.9% ahead of the 2080 CIF II scaling that reduces as resolution reduces here our original 2080 ti results would have us at around 126 FPS average although the new drivers and game update have impacted scoring and that's why we reran it we also rear and the 2080 CI for all the other tests that you saw in here but those results didn't really seem to change so far cry 5 for whatever reason got the biggest change and that's probably because of game updates 1080p equalizes results we are becoming CPU bound here so all the top results are within reasonable margin of error and are capped by the CPU for power consumption testing we're testing between the wall and the system for total system draw the test platform is 100% controlled including control over all of minor voltage rails on the motherboard the fan quantity the speed the type and the keyboard and mouse even and every single other part in the system power supply of course is also controlled failure to control even some of the voltages in the motherboard and leaving them auto like for the author smaller voltages that you typically would overlook in a normal system can throw off these readings and leave them inaccurate in some times significant waste so we control all of that stuff very heavily we're also logging over time so that you can see a proper look that includes peaks and lows rather than an average sum for the test this chart is with ashes of the singularity escalation under a 4k crazy workload giving us a real gaming workload to look at power consumption this pushes the GPU pretty hard just about to the limits running the Titan RT X card our total system power consumption Peaks at 480 watts out drawing the 20 ATT IFE system by an additional 24 watts the only thing that outdrew the Titan RT X card was our heavily modded Vega 56 contraption which used a 250% power target to blast power way pass to the stock allowance putting that total system power draw at around 640 Watts peak Vega 56 is typically closer to 350 watts for total system draw but the Titan RT X card as far as stock operating cards is heavier in terms of power requirements than the 28 e TI fe although not by that much but an additional 24 watts is something to keep in mind and a lot of that is probably from the additional memory too because it is driving a little over two times the amount of memory as on the twenty-eighth ETI thermal testing puts the titan r-tx under load with either fur mark or 3d marks depending on the test thermocouples are attached to a hot spot GDD our six module and a hot spot MOSFET both are in the centre of the rest of the SM DS in the area gpu-z is used for logging the rest of performance counters at the aida64 for additional validation for our power virus workload we monitored a maximum GPU temperature infirm of about eighty degrees Celsius with the hottest GTR six module at 76 degrees this is on a functional card so it wasn't locked to 1350 the MOSFET was running at around 80 degrees and GPU temperature here is it's warm definitely and it will impact frequency we already know that frequency steps incrementally with every couple degrees on NVIDIA architectures so to be in the 80s for the stock cooler is unimpressive but also expected this becomes more of a problem for anyone stacking multiple of these cards in an array because you no longer have the argument of being a blower device that can get rid of all that heat you're gonna have one fan at least that a sandwich in a very hot area with its fan up against the back plate of another card so that's a point of concern for you as for the vram temperature it's completely fine for one device in an open bench 90 to 95 degrees is spec and the mosfet can take 125 to 150 degrees so both are with an operating spec and even if you put it in a case it'll be fine it's just once you start stacking multiple devices you have to keep an eye out with fire strike endurance test and we saw a frequency start at about nineteen thirty five megahertz when at 40 degrees then drop to about eighteen thirty megahertz 1845 megahertz and fluctuate based upon core temperature the temperature seems to hit about 84 degrees before it lightly throttles down the clock and hits 80 degrees and tries to stay there overall this is the final chart for the review noise for the Titan r-tx runs a 33.9 DBA level when set to 41% speed which is the slowest possible that's much louder than most other partner models under idle conditions or low speed conditions at the average fan speed of 60% we're at forty two point nine DBA going to 100% pushes us to 58 DBA expect on average to be at around 43 DBA for most you cases this card is acoustically outperformed by board partners but unfortunately no board partner tighten our TX cards will exist that said you can stick a water block on this or a reference 20 atti card cooler on it for instance because the same PCB so that's the review for the Titan r-tx as it stands for gaming and overclocking and we're gonna take this further for overclocking so you'll want to check back soon for that make sure you subscribe because it will start getting crazy with these devices the cool thing with these they have our TX 2080 TI fe PCBs that means that really as I have 99% certainty that 20 80 TI water blocks will work on these unmodified so it's the same PCB it's the same GPU there's really no reason it shouldn't work as far as we're aware there is not a single component that's different so you could take a water block for a 20 DTI that already exists throw it on a Titan RT X and it shouldn't make for some really fun content which we'll do separately now that's performance oriented it's more of a drag race at that point then a functional or practical look at the device but it can teach us a few things about overclocking and how the architecture and touring scales so you'll want to check back for that but as for whether you should buy it well it's pretty straightforward really if you're just playing games then the answer is no because you can get to 20 80 I is for the price of one of these and we don't necessarily recommend to that either but if you're spending that much it's just there's really no performance to be had now we didn't see performance regression we did originally in Far Cry 5 until we reran the 2080 Ti and then we reran it at every other game just for certainty results in that really game pretty much it's the same but far cry 5 we saw a performance regression where FPS fell down a bit as like 8 FP s or something off the top so that impacts the performance but the Titan r-tx remain in the lead by a bit it's like three to seven percent on average behind the game and that's just not exciting but to be fair to Nvidia they haven't marketed this as an extreme gamer card so this should really just goes out to people who are thinking of buying Titan RT X for gaming because it cost the most and is therefore the best if you see this video then save yourself the money get a step down put that money towards anything else like literally anything else and this or just keep it there's nothing wrong with that either so we're really really hoping to look at these from a production standpoint you recently did a video by Rob Williams contributor here at workstation contributor on workstation GPU performance you can check that out it's about a month ago we're hoping to revamp that kind of approach with some of these higher-end r-tx devices so you'll get a production look at it just that's gonna take some time because a different pipeline and we're waiting on software update so check back for that once it's it's time otherwise think you're watching as always subscribe for more we go to store that gamers access net to pick up one of our products like the shirt I'm wearing today or a video card component tear down poster which is behind the dissapointment shirt back there you go to patreon.com/scishow gamers next stops out there as well that you're watching I'll see you all next time