Gadgetory

some are recent low and GQ benchmarks have included overwatch but we have loaded means to benchmark the game in a standardized environment due to its multiplayer nature this sparked interest internally and GM's Patrick Laden set forth to study over watches performance during real matches versus humans bot matches and the practice range our testing then expanded as we furthered research ultimately turning into a complete test of the graphic settings in the game and so we bring to you our complete overwatch graphics optimization guide for 2017 before getting to that this coverage is brought to you by EVGA and their EVGA gtx 1080 TI FTW 3 which should be available very shortly at this point within a few weeks click the link in the description below for more information this is a revisit to our highly acclaimed graphics optimization series we ran in 2015 starting with GTA 5 and The Witcher those guys are some of our highest performing content to date on the website and so it seems about time to try and revisit that we're starting off with overwatch but have just established a new patreon goal to make this an ongoing effort for future it launches with games and for today's look the big thing here is that overwatch is a competitive FPS as you all know that means it has different needs than something like GTA so for this guide we will be looking at testing individual graphics settings to see which of them has the biggest impact on performance and then tuning those settings jointly so looking at multiple of them tuning them together and producing somewhat optimized settings for our hardware we're testing on this also provides a way to hopefully offset the graphics so they don't look terrible because if you ran the game just at the lowest settings which of course includes the best framerate maybe that's all you care about the game really doesn't look that great but there are ways to get it to retain some of the visual quality and still have your high frame rate too for this testing we're using our standardized case testing bench with three different GPUs the GD X 1080 gaming X to establish a baseline the 1052 gigabyte card and the our X 480 gaming ax 8 gigabyte card our KS test bench is defined in the article links in the description below along with all the other methodology if you want to know what you memory and all that we're in it and it normally is the 1080 gaming X so we started there to establish the best possible scenario some matches were conducted against real players at all epic settings on the Oasis map and that yielded an in-game framerate of roughly 180 FPS average on the GTX 1080 the other two cards with 1050 and 40 gaming X provided a wide range of frame rates that establish our thesis of tackling a potential GP bottleneck in overwatch by looking at optimizing for something like again the 1050 the 1050 runs at about 60 FPS average with lows in the 40s before 80 was about 99 with lows in the 60s and 80s and again this is just a really quick ad hoc test to establish a premise and that's not bad but we can definitely improve it so any improvements we see here will scale to other resolutions as well which is beneficial for high end GPU users with higher res displays so all we've done so far is established that there's room for improvement in the GPU category with this test bench that's good because that's what we're testing for and trying to optimize for today is a lower-end GPU or a GPU constraint for example when you're looking at higher resolution and trying to tank some settings in favor of the high res like 1440 for full testing methodology as always check the link in the description below for Patrick's write-up he talks about how we tested all this stuff and if you need to know exactly how we tested it's all there as for the settings in the game we have some definitions of the graphics settings for overwatch and they're actually straight from Blizzard so they're not speculation they're correct we noticed that a lot of the guides including our own beta guide from years and years ago had either outdated information on the settings or incorrect information on the settings so this should settle a lot of that because it's from the devs that will be in the article as well because it frankly isn't the focus of the video content would take a long time to read through all of it that's not really a chart of all thing so we're going to move on straight to the research and analysis and synthetic testing then get into BOTS matches versus human matches and see if human matches are more intensive and if so how do we deal with that in benchmarking starting with initial research and synthetic testing that we use the practice range to define an initial test procedure and outline our X patience for each of the settings then formed at botton human matches as stated to sort things out further using the GTX 1050 and starting at maximum settings we turn to each setting down in the practice map and the chart on the screen now shows the results for all of that tuning this establishes our baseline for what we should focus on the most during the more intensive at competitive matches and further testing where we're fine-tuning the settings other than just toggling them and these numbers highlight that dynamic reflections shadows and local fog detail are all worthy of attention which we'll highlight in the charts we're happy to see that texture and model detail don't seem to impact framerate as heavily as some other options as they're the ones that have the most visible impact on the game from a graphics quality perspective as a user texture quality is most dependent on a VM as always it given that the higher quality textures of high resolutions will require more video card memory but we never really saw a VM request size exceed two gigabytes on the gtx 950 at 1080p of course higher resolution will change this just quickly noting here's a frame time chart showing the biggest change dynamic reflections and that's with it toggled at vs everything that maxed so you can get a better idea of what's happening in the 1% and lo percent metrics we saw about a 43% improvement from baseline by toggling dynamic reflections off and about a 14% improvement over baseline with shadow toggling which you can see if we switch back to our FPS chart we had originally on the screen and about the same for fog disabling ambient occlusion gave us about 8% performs back and we're also seeing a theoretical maximum throughput of 207 FPS average with the 1050 OC card and with the bench used this means that we're bumping into constraints at that point at a hardware level as the GTX 980 is capable of achieving a higher framerate so we're not at any kind of engine constraint we've got some screenshots here to show the highest graphics quality then medium then lowest in the practice range these are embedded in the article if you'd like more time to look them over and study for differences without YouTube's compression but it gives you a foundation of what the difference looks like just from the presets time to get to the real tests the ones that weren't done on the practice range we're obviously no one ever this game anyway so for the real test we fought BOTS with the same team and character composition same exact character models same characters per team so that's important and then we also use the same bot difficulty which I believe Patrick had set to medium and the same map which was named Bonnie for all tests the objective was set to attack and we played offense for every single round because we're playing against bots so it's easier to control how the round progresses and attacked and that means we can just control the task easier because we can define when things progress or end or whatever for human player matches same thing same map same objectives all that stuff the only difference is humans and obviously they have full control over which characters they choose so that impacts FPS potentially but that's what our FPS benchmarks set out to determine let's first determine if human matches are significantly more demanding than bot matches of an equal player account these recordings happened over about a five to six minute period and were averaged over the duration which is really the only way to validate a potential test method as there's so much variation between matches that a short test pass does not contain all the information required to make a sufficient analysis this chart shows some of these slight differences between boughten human matches on NIM body and a Hollywood but nothing significant we saw a two FPS swing from bot two human matches on embody and could easily account for that and what is experienced in each match that would be the usual variation between what players are doing in the game considering those changes match to match these numbers are actually remarkably close and they illustrate that we can safely test using foe bot matches without concern of misrepresenting real gameplay they are effectively identical and bot matches of course are much easier to control for in testing so that is the preferable mode of test this also gives us more control over the testings and again medium BOTS can be easily controlled in a fashion that doesn't conflict with our ability to perform specific actions in the game for example not dying and these results being effectively equal established our baseline for validating a new test method going forward here's the next chart we're now playing full bot matches on ambani with otherwise epic set and 1080p on the GTX 1050 OC from MSI then were manually tuning individual settings as mentioned above dynamic reflections at shadows and local fog were the top candidates for optimization we also chose to test effect detail lighting and refraction since we felt that these were options that didn't have a real chance to affect things in the practice range and could potentially contribute to 1% and 0.1% load dips in game the trend scene and the practice range continued in game although the actual frame rates changed drastically as you might expect the maximum improvement again resulted from disabling dynamic reflections but the 0.1% low value remained at 45 FPS turning down effects detail lighting and refractions produced frame rates indistinguishable from the maximum settings even in the 1% in 0.1 percent global categories this confirmed our choice of three top candidates for optimization and so getting rid of dynamic reflections now in our bat match for real-world representation we see that this frees up 30% increase Headroom in average FPS for shadows we see a 17% Headroom improvement while setting everything to low post a maximum possible change of 210 percent and that is from the highest settings let's narrow things down here's a chart focusing on three settings offset from epic all on low and then one setting with reflections shadows and fog sets a medium and a final config with them disabled these are the three most impactful to framerate so this begins our final optimization attempt on the 1050 simply turning these settings down to medium keeps frame rates well above 60fps on average with about a 50% improvement jointly disabling them completely gave 117 FPS average or 95 percent improvement jointly but again they're disabled completely and that's not great or about 69 to 70 FPS lower than turning every single setting to minimum we've got a frame time plot that demonstrates the low values better as well where we can see that the improvement from completely toggling these settings is noteworthy over the medium configuration but we find the frame times with medium to be accessible given the visual trade-off shadows can have an impact as well on competitive ability and as one setting we'd prefer not to completely disable here's a look at the same configuration that the RX 480 the RX 4 ad often bumped up against the 300 fps cap under these configurations and that means the 267 FPS average that you can see on the chart is something that would be higher without said cap at maximum settings the 0.1% low was still above 60 fps perfectly playable however the three relevant settings disabled yielded a 0.1% low rise to nearly 120 fps good news for owners of 120 Hertz displays with a strict zero-tolerance policy scaling here shows medium settings at boosting from our 99 FPS baseline by 44% comparable to the GTX 1050 scaling as you might expect we see an 81 percent improvement by disabling reflections and shadows completely and that sums it up so for people who have lower end GPUs that are causing a bottleneck in the system for overwatch meaning your GPU is the constraint because it is not powerful enough to keep up with the CPU which in this game is completely reasonable CP is not that high demand the order of settings to experiment would be this you should change dynamic reflections first and then shadow detail and then local fog detail at this point you should be able to push those other settings to higher values without a huge performance hit while clawing back some of your FPS from the more drastic changes in the categories listed here ambient occlusion is another option for anyone who is truly desperate but we saw maximally something like an 8 percent difference there so that is something we try to leave on though if you really need something else you could play with ambient occlusion as well other settings don't significantly affect the GPU and decrease quality with no real appreciable benefit in terms of framerate throughput and turning them off completely in terms of things like shadows is unnecessary and undesirable because shadows can actually provide some competitive advantage depending on how well you play and that type of thing so as always thanks for watching you can find the full guide link in the article below and subscribe for more patreon.com slash gamers and access to help us produce these things specifically because they do require a lot of effort for game specific guides and they aren't always quite as interesting as Hardware benchmarks for our audience so just as a side note using his own guide Patrick was able to get his FPS above 70 with his aging r9 285 XFX GPU so not bad at all definitely beneficial hopefully this helps you let us know I'll see you all next time you