Gadgetory

everyone I'm Steve from gamers Nexus dotnet and we're here to talk about our overwatch GPU benchmark that we ran over the weekend there was another overwatch beta weekend so big disclaimer here it is in beta where it was at least in beta and that means nothing's finalized the optimizations aren't done and drivers aren't done from Nvidia or AMD so at this point all of the performance metrics are or should be temporary and that because the drivers sports not there and the game is not optimized fully so what you see here is not necessarily applicable to the game as it launches if you're watching this in 2016 then maybe check and see we've done an updated benchmark to make sure that the optimizations haven't changed in substantially but still the game is fairly playable across most GPUs in our test suite and that's what you'll see here today so it is a it's a it's a decent title in terms of scalability across GPUs and the main thing here is that blizzards got a fairly useful settings menu where you have good control over the options and in our initial settings tests where we look at Delta between different setting options there is actually a pretty big disparity between some of the less useful settings which means that you can get a good amount of FPS out of something without necessarily losing a noticeable amount of fidelity if you are on a lower MGP you so let's dive into it here we already released the benchmark course on the channel it's less than two minutes the video the course itself is 30 seconds and you can see that if you want to replicate our results or try and see where we got our data the course was done in single-player to eliminate a lot of variables a ton of variables we always a single player were possible and in limited testing for internal use at gamers Nexus we found that the Delta for single player versus player was generally pretty low so even with the added characters we weren't seeing a huge amount of impact to framerate and that's a good thing that means our tests are easier to scale and reproduce the only time there was a big impact was when sort of everyone converged on one spot and there was a lot of action a lot of effects going on at the same time in those cases there was an FPS packed with a couple percent but overall this is a pretty reliable test for multiplayer and singleplayer the items in the settings look out for include primarily render scale surrender scale just super samples and down samples the rendered resolutions of screen that's all it really does and render scale comes to it comes a different couple different flavors so there's a low which is a 50% scale that means you're down sampling your resolution quite a bit down 50% from what it would natively be which is a hundred percent there is medium at seventy-five percent high at 100% that's what we ran these on and what you should probably run them on unless you're on a setting where you want to super sample ultra 150 percent and epic at 200% what that means is if you run ultra and you're on a 1080 display you're effectively rendering of 1440 output so that's similar to DSR and VSR from Nvidia and AMD and epic at 200% is effectively a 4k output if you're on a 1080 monitor and this has the impact of just running natively at 1440 or 4k so if you do the render scale ultra or epic there will be a performance hit to the same tune of a resolution increase that's it's about the same the next item anti-aliasing is listed as low medium high and ultra which is a little confusing so what these mean low is FX AAA which is kind of the blurrier but very performance friendly setting medium is SMA a low quality not sure how many taps that is I'm guessing it's probably 2x high is medium quality probably 4x and ultra is is high quality which is going to be probably 8x or 16x we haven't confirmed those numbers yet the initial settings comparisons we did look at a couple of the different settings in the game and compare them across multiple resolutions so the the biggest impact to a setting performance wise is going to be changing the render scale and if you run epic it's basically four times the resolution that you're rendering if you're at 1080 so I were actually at any resolution because a 200% increase so render scales the first thing to look at for improve performance ambient inclusion has about an 8.5% FPS impact and disabling this reduces the apparent depth of things like foliage and trees and buildings it reduces the shading and depth of our appearance of depth on objects as surfaces connect we have a wall like that or something that's kind of sticking out more than other then the ambient occlusion value will affect that and it's an 8.5% performance it refractions effectively zero FPS impacts we need to further research that local reflections again negligible impact effectively zero FPS that's things like mirrors and mirror like surfaces metallic surfaces texture filter quality is normally something like anisotropic filtering which generally has about a zero performance impact but in this case we saw a 3.7 percent Delta between max and off and that helps with the application of textures on non oblique angled surfaces dynamic reflections was a huge impact on FPS so at the max versus high range we saw a hit of about 29% at the Hybris medium range we saw 13% and at medium versus low we saw almost 15% performances so that's the second place to go after render scale a quality couple FPS between settings so that's really not a big hit right now that may change will research it more our first set of charts here shows the preset scalability testing so this is an internal test we always conduct but I decided publish at this time and this shows relative performance across a single configuration just the 980ti and it shows deltas between frame rates at low medium high ultra and epic and that informs our decision on which settings to test for the final benchmark so we can't do them all realistically the first chart shows the Delta loss gain Delta loss and gain from setting to setting and this is only one charge so you're only looking at the average FPS you don't see one percent and 0.1% lows here and this shows ultra as our established go-to setting that's why you see it at a hundred percent that's effectively what we're considering normal for this test what we think most people will be running and if you look at the other the scaling here Ultra is about 26 percent faster than epic which scales up a few of the settings and it's about 30 percent slower than high with low allowing almost anything to play overwatch but note here that we were hitting 300 FPS which is the games cap right now so it might actually be an even bigger gap between low and medium depending on where it's capping out on FPS 4k for the first benchmark we're looking at the fork HRT's this is obviously a pretty heavy hitter on the GPU sweet reducing the render scale to medium which is a 75% render pushes FPS about 20 to 30 20 to 30 FPS higher across these cards on the 980ti especially but at that point you're basically reducing your resolution so it's worth considering dropping to 1440p natively there's currently no SLI profile from we've gathered I'm getting some crossfire tests done in the next beta so we'll figure out if there's any scaling there but SLI performance shows zero scaling across all resolutions and presets testing right now tested right now which means that you're seeing these results where it's kind of unplayable at these settings with 4k you have to you would drop the settings substantially or just reduce the resolution of play overwatch is a game where you really want close to 60 F guesses it does start to feel jarring and weird at 50 even 55 FPS so I would try to target 60 here running epic with a high render scale 1440p resolution is a lot less brutal to the GPS than the 4k resolution test which is 4.6 million pixels at more than 1440 so it makes sense the game runs reliably with high frame rates from the 390 X and up including the 970 the 390 X of course the 980 and the 980 TI and these are 9 to 90 X runs close enough that some minor tweaks would push it over the top but low 0.1% frame times hammer and these cards across all of the overwatch tests so this is an optimization level issue either on blizzards end or Andes driver's side and is something that should be seriously considered I would not base potentially a GPU purchase on these results right now because as drivers roll out this 0.1% hit that exists on Andy and not and Vidya potentially be reduced at 1080 epic that's probably what a lot of you will be running if not 1080 ultra this shows how well blizzards overwatch plays on most hardware at 1080p and Andy again suffers from poor 1% low and point one percent low frame delivery but averages are up across the board for all cards the drops do hit harder and the hardware though optimization should resolve this is going forward by looking at the other stuff everything from the GTX 960 and up including the 380 X is brand new 970 the 290 X and so forth kind of run overwatch reliably at 60fps with 1080p epic and there are 9 285 is close enough that we'd consider it within striking range if you made some tweaks and this is a look at the 1080p Ultra chart which is a lot more forgiving on the cards you can see that the lower end hardware including even the 950 starts to enter decent playability range so what's the best video card for overwatch it depends on your settings of course right now at 1080p it's fairly playable on the GTX 960 which is 170 at 2 gigabytes 200 plus at 4 gigabytes the r9 380 another $280 card where the r9 380 X was brand new and is a couple percent higher average performance than the Nvidia cards at the competing price range what has worse low frame times at 1440 we shift our recommendations toward the r9 380 390 x excuse me and the GTX 970 which is been on pretty crazy sales lately about $290 with the 970 performing significantly higher in 0.1% low so that is definitely something to consider but again it's a beta so it might change 4k is effectively unplayable right now as the SLI and crossfire profiles come out as the games optimize that might change if you want to play 4k right now you've got to do it on a high-end hardware so that'd be a 980ti or fury ax or similar which we do not have a fury X for testing right now and you would still have to drop your settings from ultra or epic down to something like medium or mix of medium and high which is not really something that feels great to do but it would allow 4k to be more playable on a mix of high and medium and maybe you wanted to alter settings so you have to do a lot of tuning there to get that working properly right now but that is all for this benchmark get check the channel for all the other stuff and of course the benchmark course that we already ran to see the max graphics output of overwatch here's what it looks like and check the patreon link if you want to help us produce more of this content that'll be down here in the post roll video I will see you all next time