Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

Battlefield V DLSS Tests: Trying to Find the Upside, ft. 2060 & 2080 Ti

2019-02-21
all right so let's get this one over with we're going through another game test now it's been several days of these we're looking at battlefield 5d LSS we actually started this several days ago and we push the content back a couple extra days to look deeper and try and find something positive with DL SS we're sure you've seen the criticism online we were working on our content at about the same time as everyone else working on it and after seeing the sort of global response the DL SS basically being Vaseline on the screen we decided to just do anything to try and find a more just something that made it well look better so we tested our TX 2060s 2080 eyes we did a couple of different game graphics settings we did multiplayer single-player stuff like that trying to find the good in DL SS and well we're not completely empty-handed but let's just talk about it in this one and do qualitative and quantitative tests with DL SS in bfv before that this video is brought to you by the gigabyte Z 390 ARS master motherboard which comes equipped with one of the more powerful Z 390 VRMs for heavier overclocks on the new 9th gen intel cpus the ARS master is also one of the few motherboards with a real heatsink this generation featuring a mix of high surface area fins and looks oriented cover blocks oh and it's also got updated our DV illumination learn more at the link below in working with DL SS which requires DX r to be enabled and does not have an off option for ray-tracing we've noticed that there seems to be some sort of gradual memory leak in battlefield 5 with DXR enabled we haven't been able to confirm this with 100% consistency or confidence but it happens enough that we're at least somewhat confident and talking about it as an issue over a period of time related to the GPU memory available the game seems to eventually slow to a bit of a crawl we noticed this most prominently on the maps devastation and Rotterdam the RT X 2060 had even just 1080p with lo DX our settings would bottom out at 0 FPS for long intervals sometimes completely crashing this seems to coincide with memory filling up but it wasn't just that the 2060 was doing it because of a 6 might limit or something our twenty atti with the XR on the higher two options also begins to encounter hard stutters occasionally in testing we have not been able to reproduce this issue with DX R and D LSS both disabled but we have reproduced it on two systems with two different RT X GPUs with some consistency but not 100% consistency the memory seems to be fully allocated on each device when this happens so our speculation and it is pure speculation mind you so we may be wrong on this is that it's a memory leak of some kind within the game code the 2080 TI will sometimes take 20 minutes to start encountering these occasional stutters so we don't see it in our brief test passes and then as a side note with DX R in Unreal Engine we've noticed that there is a similar sort of slow down over time so this may just be a DXR issue within Windows that would be Microsoft's domain and they actually make a DX R and then Nvidia plugs into that so it may be a wider issue with the XR but either way we wanted to point out that there are occasional memory leak issues in battlefield 5 with DX r enabled and that would also go with DL SS being enabled we're going to put some DL SS on and off comparisons on the screen now but we won't reveal which is which and what the resolution settings are until the end you should look at this and personally determine which you think looks best then we'll reveal which is which later will keep the left and the right consistently one setting so that they remain the same as we change footage of screenshots you just pick which side you think is the best looking at DL SS from the perspective of quality our very first impression was that the differences it were difficult to see an actual gameplay like DX r ray tracing when you're actually playing the game battlefield 5 these things sort of get overlooked in favor of while playing the game as we use D LSS more across more maps and began to really scrutinize it sometimes with the face is one inch away from the monitor we noticed that it almost seems to remove a sharpening filter and then as you sort of sit back and look at it wider you don't need to be that close to the monitor to notice the lack of the sharpening because once movement slows down once you stop running around and you just camp somewhere for a second which is real gameplay you'll notice that it looks sort of blurry and this effect this blurry or blurring or lack of sharpening becomes more noticeable as you look for it or as you become aware of it so this is not something that we noticed immediately but of course noticed within well a couple of minutes of actually playing around and looking at it so we're not sure if there's a literal transaction going on where there's a sharpening filter or a post processed sharpening effect that's being applied and then removed with the LSS but in before and after comparisons we have of rotterdam devastation the shooting range and single player we noticed that DL SS off shots always look like they're significantly sharper than the LS s on this is not good or bad really it's sort of subjective but in general most people do tend to like the contrast or the sharpness based on comments online about this issue so you can look at this two ways either there's some sort of sharpening post process that happens with DL SS off and that's removed with TLS is on or that isn't true at all and DL SS just does this and that's probably the worse of the two options given those two let's reveal which side has been using which settings now if you chose one and now disagree with your choice that may be a bias if you couldn't tell the difference between them that's also noteworthy because then you should clearly just use whichever is better and so far as performance benchmarks and that's always going to be the lower resolution with DL SS off giving you better performance and really in a lot of instances either indistinguishable or better visual quality depending on your preferences and of course YouTube compression might also kill any minor differences so that's of course a consideration as well back to it though the idea is that DL SS should make it easier to run DXR we've already seen that this is objectively true NVIDIA has done well with delivering on that promise but it's still even easier suggest E is the next resolution class down when speaking strictly to performance and the looks as we've shown are either not that different or are maybe a little bit better with that lower resolution although you might not like the sharpness there's some instances where we can see DL SS being better and we'll talk about those in a bit we already showed some charts for that stuff what we need more of is qualitative versions of this discussion so here's your point of consideration the lower resolution and DL SS off tests like 1440p instead of 4k d LSS still show sharper objects than with DL SS enabled and 4k resolution after Nvidia's blog post we also decided to test the r-tx 2064 this with 1080p and 1440p we thought maybe maybe this one would be better would have either better quality or make more sense because it's a lower end car this set of tests also however showed us that the D LSS images were less sharp they were in some instances just plain blurry almost like there's a constant layer of motion blur applied even when the camera's not in motion we saw something similar in the initial execution in Metro where we it looks sort of like with our TX GI on there's that layer of motion blur DLS s also impacts the object render distance where some objects get completely cold with TLS s on this is something we noticed in the single-player French campaign level in the mud distant trees for example we're getting cold with TLS s but remains present with the LS s off there were only a few objects where this happened but it did happen and of course the more you play the game than where you'll find of those in the same vein DLS s impact LOD scaling and object detail and distant objects visible plainly and some of the neighboring trees in those clips earlier we can see that the branches sort of pop in and out you lose some of the height or the depth of the trees depending on the LOD scaling we wanted to wrap our content here originally but figured that there had to be a benefit somewhere so we looked and we looked and we searched on multiple GPUs and after suffering some memory leak issues we eventually found an area of debatable improvement in the multiplayer map on rotterdam shot from the spectator view in the clouds we noticed that the thin geometric detail on ships like the masts and on the train would alias heavily with DL SS off and running 1080p remember this is an RT X 2060 so it's realistic you would run that resolution this was aliasing to the point where there's almost a marching ants effect as you move around it could be considered annoying maybe if you're really looking at that with DL SS on in 1440p those objects either get sampled out and omitted maybe not the best or get smoothed out depending on which object it is and how much data there is for the LSS to work with this is not necessarily good or bad in an objective form but it is something for you to determine the value of in general if you have to look that hard for a positive difference it's probably worth just leaving it off in our previous DL SS benchmark with Final Fantasy 15 we found that 1440p resolution always produced better performance than 1440p with DL SS faking 4k and super sampling which we can show again on the screen now in that video we also talked about the pros and cons of each configuration sometimes like with the license-plate text that we showed in the Final Fantasy benchmark DL SS was arguably better for legibility than just straight 1440p but of course 4k was better than both unsurprisingly it was worse always than real for cable that's because DL SS is meant to sort of fake 4k and improve your framerate at the same time which is what we're looking at today in other instances the LSS was significantly worse than 1440p like where we had flickering and fences and light fixtures whereas 1440p looked normal and without those anomalies we talked about much of this and chalked it up to a semi random seed being generated to alternate where on the screen the LSS was sampling this we assumed was true because even when paused the menu screens would still show flickering and objects that have really thin geometry or are just maybe using alpha textures or something like that both of which can describe the fences where we saw a flickering so the example of flickering when paused does coincide with maybe some sort of semi randomization in the sampling pattern or something like that since it's not fixed but when in movement it makes more sense that it is flickering as it is because you're starting to sample around these really thin pieces of geometry or really thin elements like in the fences in a way that it might just slightly miss the actual piece of metal hit some of the background and then you get a super sampling of that which is no longer accurate so that explains the flickering in that game we didn't see it quite the same as we talked about earlier for Battlefield 5 but for Battlefield 5 our first benchmark sequence will look at one of the more intensive single-player levels and ultra presets for the 2080 ti but with our TX only set too low we'll look at our TX ultra next for this one 1440p with TLS has often our TX at low once again chart tops but that isn't really surprising at all the question as we answered earlier is more about what looks better this is the other half of that equation though with DL SS off we're at 82 FPS average with lows reasonably x at 59 FPS 1% 55 FPS 0.1% using again our 1440p setting where you actually can't enabled ELSS if you wanted it our standard deviation is minimal in these tests as you can see the bars on the graph so these numbers are a fairly consistent run to run the 82 FPS average puts us at 9 present faster than 4k with DL SS off but with RT X completely disabled if you want it to play at true 4k at a reasonable frame rate clearly it's possible by completely disabling RT X and just using normal in quotes their game graphics settings enabling RT X but keeping DLS s off drops us to 52 FPS average so like-for-like we dropped from 75 to 52 FPS and lose 30% of our performance instantly with RT x set to low this isn't a scene that doesn't have much actual ray-traced reflections there's not really a lot going on for ray-traced reflections it's constantly processing but that doesn't mean it's really showing you anything and that's not to mention some of the visual artifacts that we can show as well like fire shimmering and the camera despite no fire being actually nearby or relevant to our viewport the comeback story is with DL SS sort of where we instantly gain 27% performance to 66 FPS average from 52 FPS average even still this is worse than 1440p with TLS s off and RT x low unsurprisingly and worse than just disabling RT x and playing with 4k instead it all comes down to image quality and we'll put one of those comparisons back on the screen now if 1440p with TLS s off looks worse than 4k with the LS s on and it's noticeable and frequently worse then TLS s is worth it if not it's clearly not worth it we've given you the tools to decide for yourself what you think but we'll loop back on this more on the conclusion let's look at the next test and it's in the same area except it's now with our TX set to ultra from low so all this is the same except our TX 75 FPS average is the high set by 4 KD LSS off and our TX completely disabled this now surpasses 1440p and this particular test as opposed to the previous one and that's because the 1440p tests are now working with our TX set two altra instead of low the end results is that we lose 20% of our performance from the previous charts 82 FPS average high now ranking below 4k with the LSS often our TX completely disabled big important note there if you wanted to keep our TX on you could either take 1440p and our TX ultra or 4k with the LS s on and to some extent our TX ultra which lands us at 54 FAS average this is still worse than 1440p baseline which had us at a 21% higher frame rate if you wanted a more like for like comparison going from 4k with the LSS off at 37 FPS average to 4k with the LSS on did provide a noteworthy improvement to 45% that's a damn good jump it's just that 1440p is probably still better or functionally equivalent for most people so at that point we question whether the gain matters in this particular game that said as a technology clearly there's an op lift and performance here it's not really that new of a technology the implementation is new but the idea isn't so there is a gain it's a big gain - it's just that it's not you're losing some image quality there so maybe this is promising for other applications we just need to see if those ever come to fruition and be convinced from an image quality standpoint the performance standpoint there's a good argument we really need to see that image quality standpoint though finally just for one more charted area we also tested the practice range for the opposite load level of the previous single player level the practice range is probably the lightest weight area of the game it's not really that representative of actual play but it's still useful for just kind of the opposite end of things so this almost never uses RTS reflections we thought we'd look at it as a performance trial 1440p leads once again although FPS scales higher in all tests as a result of the simpler environment we're at 103 FPS average with well time to load versus DL SS off RTX off and 4k at 90 FPS average with 4k DLS s off we are applauding about 56 FPS average still a steep decline from the 78 FPS average with DL SS on let's instead look at frame time performance as always frame time plots are useful for their more accurate representation of frame to frame performance rather than potentially averaging out any important data points lower is better for frame to frame interval shown as frame times on the Left axis but more consistent is better still than lower for this one we can first plot 4k with DL SS off and RT X on showing frame x between 22 milliseconds and about 37 milliseconds the frame to frame variants which is what we actually care about isn't all that bad here there are a few spikes greater than the desirable sub 8 millisecond variants but overall the so-called smoothness is fine it's not high frame rate so that's not great but the consistency is acceptable for reference sixteen point six 67 milliseconds is about 60 fps for the 4k DLS s on test drawing that line now we see performance significantly improve over DL SS off and those a few larger frame time spikes also now vanish with this run reducing the workload on the GPU overall helps to ensure more consistency frame to frame which is also significantly advantageous vs. 4k with the LS s off so it appears that DL SS is working it's doing well for performance it's just that there's those other problems so again this option does not exist in a vacuum and 1440p posts better frame rates with similar frame to frame intervals resulting in the best overall experience in fact the frames of frame intervals are better on average by at least a couple of milliseconds the longer the line the higher the FPS here and of course the shorter the line is on the Left axis the lower the frame time also a good thing so we see that the 1440p line ends up with the most frames drawn it has the best frame rate it's also the most consistent so really it's just objectively the best and so far as performance so going back to the image quality scenario is the only fallback for DL SS and this illustrates well why for 40p remains the best option in this use case from at least an objective standpoint that's what we have for you then so we had I guess memory issues notwithstanding which do seem like they might be just related to DXR and Windows in general because we're having those issues in Unreal Engine 2 so those things notwithstanding DL SS it's just at a point where it's like even editing back the video Andrew was saying he could barely tell the difference and when we were playing the game we could barely tell the difference until you have to stop and look and then yes you can see it but like when when we first started playing the game and you're actually playing it like for real not just walking around aimlessly trying to find differences you're actually playing the game the differences are very difficult to spot because a lot of the time it amounts to some blurring of the textures some loss of detail some of the sharpening is gone but a lot of the time those elements aren't things you're really paying attention to when you're trying to shoot people in the game so so the difference is overall in action are not that noticeable and that's not really a good thing it might sound like a good thing in some ways because we're not saying it's worse or well noticeably worse when you're really playing but it's really not a good thing because again you can drop the resolution you can get better performance uplift as a result of it because it's just it's straight 1440 instead of 1440 super sampled which is what DL SS is and so because it's straight 1440 performances up and then it's a bit sharper things like that now sharpness is sort of up to debate to some extent whether you'd like it or not we showed footage earlier but for the most part you really do have to just like kind of pause and look really close to the monitor to try and see those differences when you're first trying to spot them but when do you see them and it is really obvious like once you start noticing the differences suddenly you can't not see them so definitely DL SS is overall lower image quality we would say to some extent objectively on the subjective side there are some debatable areas of improvement like maybe for example in the masts in the ships in some of the train prefab elements stuff like that small mesh details things where it's it's really thin geometry there's not much geometric complexity there or maybe an alpha texture or something like that you start seeing some debatable improvement and that's mostly and just either either it misses when it samples and it doesn't hit the small piece of geometry it samples everywhere around it and then you lose that contact so it doesn't get drawn in the super sampled version or you hit it and it gets almost like emboldened in which at which point the the aliasing issues go away with the marching ants effect on those masts and other thin geometric elements or alpha textures or whatever so that's kind of what we're dealing with it's it comes back to the same philosophical thing that some of the other stuff we've looked at with our TX where if there is some sort of improvement but you have to look really hard to fight you have to do what we're doing for a job to find it then does it matter and if there's a performance uplift and there's an arguable maybe sometimes improvement sure turn it on but if there is not a performance out there's actually detriment and the benefit is exceedingly rare or completely subjective that's where it gets either easy to say turn it off and just drop the resolution or if it's subjective then I'll have easily look at it figure it out if you like one more now overall reading the comments online it looks like the verdict is more or less that people don't like DL SS in battlefield 5 we don't really have a extremely firm stance on this one it's not it is not impressive it is not necessarily adding meaning it's well let's rephrase that it is not adding meaningfully to performance that's very firm so then it just comes down to does it add to the look and really overall the answer's no not really because the only way by and large for most scenarios in the game that it benefits the visuals is if you don't like the sharpness if you maybe you feel like the sharpness and battlefield is unrealistic and maybe you feel like it should be a bit blurrier or at least less defined and that's potentially a valid criticism so in that case sure turned it on but it's kind of the wrong reason to turn it on so that's really where the the problem is is it's that's not that's not what's supposed to that's how the benefit that's intended so so if there's benefit there absolutely feel free to use it we're not telling you not to but it's just it does not seem to be quite what we were expecting and NVIDIA interestingly has painted itself into a corner because it promised the world when it brought this technology on to display in Germany and people are now expecting Nvidia to do what they said you know would happen and it's just not it's not there yet so maybe the OSS will get there some of this stuff is just it's a matter of like not losing some of the detail this is a technology that's supposed to improve over time it has fed more ground truth images and improved in that way but it is still it remains to be seen how good the LSS can be if we can get to really where it needs to be final fantasy we showed previously trophy clips in this video and it just was not on average a very good experience to enable because you start getting flickering and dissipating objects and thin mesh materials disappearing LOD scaling changes sometimes unfavorably when you could just drop resolution call it a day so not particularly impressed it's not the worst thing in the world but it's just either there's you don't notice the difference or you don't like it and neither of those is a great option because performance is down regardless so that's it for this one thank you for watching subscribe for more go to patreon.com/scishow sexist to support us directly in what we do or you can go to store documents access net to pick up a GPU artifact and shirt like the one I'm wearing right now it's on the store for 20 bucks so thank you for watching subscribe for more i'll see you all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.