Game Benchmarks stub

Overwatch Video Card Benchmark – A Scalable Title Tested at 1080, 1440, 4K

Posted on November 23, 2015

Forthcoming team shooter Overwatch is Blizzard's first new IP in years, fusing familiar FPS and team-based elements with MOBA-like playable characters. That, at its core, is what we'd call a “team shooter,” a genre that's been popularized most recently by Team Fortress 2.

The game is still going through closed beta testing, with select Battle.net accounts receiving invites to play-test the game over a few weekends. This weekend's test was, according to Overwatch PR Manager Steven Khoo, an attempt at learning “how Overwatch runs on your system” and a reach-out for “technical feedback.” We figured we'd throw ten video cards at the game and see how it does.

Overwatch isn't particularly GPU intensive, but it does make use of some advanced shadow and reflection techniques that can impact FPS. We performed some initial settings analysis – shown further down – to determine top-level performance impact on a per-setting basis. This is the basis of our eventual graphics optimization guide (see: Black Ops equivalent), something we'll finalize at the game's launch. For now, the goal was to provide a foundation upon which to base our GPU test methodology with Overwatch. This graphics card benchmark looks at the best GPUs for Overwatch (beta), testing 1080p, 1440p, and 4K resolutions across “Epic” and “Ultra” settings.

Explaining Some of Overwatch's Settings

Remember that Overwatch is in beta, so this test is not necessarily representative of the final product. We do not have finalized drivers from either major GPU manufacturer, meaning there is presently no driver-side optimization for Overwatch. We are also limited on settings knowledge to our own experience and understanding of graphics technologies; some of these settings, until we can speak to Blizzard's tech graphics folks, have left us a little limited on detail. We have provided our best, informed answers below.

overwatch-settings

Advanced Settings which don't require a restart to apply: Render scale, refractions, ambient occlusion, local reflections, texture filter quality, dynamic reflection, and anti-aliasing.

Field of View: This can be thought of as a scaling of a “wide angle lens.” Field of View adjusts how wide the player's camera is. FOV is usually measured in degrees. Overwatch uses a slider instead of a numeric measurement, a decision with which we disagree, with the right-most position being the widest FOV. Wider FOVs can introduce some “GoPro effect” distortion toward the edges of the view if too high on a particular aspect ratio or resolution.

Enable VSync: Locks the framerate to the monitor's vertical refresh rate (often 60Hz, 120Hz, and 144Hz). This reduces tearing from mismatched frametime delivery, but can create stuttering in scenarios where the GPU cannot keep up with the display's demand for new frames (at 60Hz, that's every 16ms; at 120Hz, every 8ms, and so on). Learn more about this in our V-Sync dictionary entry.

Enable Triple Buffering: Loads more frames into the framebuffer on the GPU. This can aid in smoothness of frame delivery, but greatly impacts performance and VRAM consumption. We strongly recommend disabling this for best framerate, which is generally more important to a first-person shooter than smoothness of frame delivery.

Limit FPS: Lock the FPS to the refresh rate or to 30FPS. We strongly recommend disabling this.

Graphics Quality: Settings available: Low, Medium, High, Ultra, Epic. Adjusts all “advanced settings” automatically. We tested each preset's performance below.

Render Scale: Scales total game graphics render resolution up or down (supersampling and downsampling). The scale is:

  • Low – 50% resolution
  • Medium – 75% resolution
  • High – 100% (this is what you generally want, and what we used in test)
  • Ultra – 150% (from 1080, this becomes 1440 render resolution)
  • Epic – 200% (from 1080, this becomes 4K render resolution)

We strongly recommend leaving this option to “High” (100%), as that will be the most “normal” to what you're usually experiencing. Scaling to higher scales will produce near-equivalent performance deltas as scaling actual resolution to 1440p or 4K (or downward). This is the quickest place to gain FPS, just like changing resolution is the quickest place to gain FPS – effectively the same thing, just one natively scales the entire screen, the other supersamples / downsamples.

Local Fog Detail: Changes LOD of fog / cloud / smoke FX that exist naturally in maps. We are unsure if this also impacts fog / smoke detail related to player character abilities.

Shadow Detail: Changes LOD of shadows. Lowering this setting produces a “marching ants” effect, where shadows can look pixelated and less smooth. High shadow detail settings produce smoother character shadows that more accurately mirror their host object.

Model Detail: LOD of models (unsure if meshes are also affected). Lower model detail will impact most 3D objects in the game. Lower model detail creates “blocky” and less smooth 3D objects and characters.

Refraction Quality: Refractive light interaction with surfaces which distort light rays (glass, semi-transparent objects).

Local Reflections: Seems to impact reflection quality in mirrors or similar objects. Available only in toggled off/on form.

Ambient Occlusion: Lighting and shading effect that creates apparent depth by manipulating shading where surfaces meet or overlap. The easiest example, as always, is underbrush: AO creates an undershadow on foliage and plant life (leaves, usually) which adds to apparent depth. This can have massive performance impact in some games. We tested this very briefly in Overwatch.

Dynamic Ambient: Unsure on what this setting impacts at this time. Seems like application of AO to moving objects.

Texture Quality: Affects texture resolution. Scales from Low, Medium, to High. Lower texture qualities reduce texture resolution and apparent depth created by grittier textures. This traditionally lowers VRAM consumption. 2GB GPUs are disallowed from running “high” texture qualities.

Texture Filtering Quality: Anisotropic Filtering or similar texture filtration technique. See our dictionary entry (linked just prior to this text) for more information.

Dynamic Reflections: Seems to impact mirrors (and similar objects – windows, semi-transparent textures) and their quality of character reflection.

Translucent Shadow Detail: Unsure at this time. Our guess is that this impacts LOD on shadows that interact with characters presently using abilities which apply translucency effects (like stealth or teleportation skills).

Effects Detail: Impacts particle FX and other active FX that relate to abilities. Lower settings will create a “flat” feeling effect, but aid performance.

Anti-Aliasing Quality: Scales from Low, Medium, to High. These settings represent:

  • Low – FXAA
  • Medium – SMAA, “low quality” (maybe 2x?)
  • High – SMAA, “medium quality” (maybe 4x?)
  • Ultra – SMAA, “high quality” (maybe 8x to 16x?)

Simple Ambient Light: Enabled for lower graphics settings. Replaces more complex lighting with simpler rays / illumination, which aids in framerate performance.

Test Methodology

We tested using our updated 2015 Multi-GPU test bench, detailed in the table below. Our thanks to supporting hardware vendors for supplying some of the test components.

The latest AMD Catalyst drivers (15.11.1) were used for testing. There is no Overwatch optimization at this time. NVidia's 359.00 drivers were used for testing, which also lack official Overwatch support and optimizations. There are currently no SLI profiles for Overwatch. Game settings were configured to "Epic" and "Ultra," following findings that most cards played “Ultra” fairly well, using the presets at 1080p, 1440p, and 4K resolutions. We set Render Scale to 100% at all times (“High”).

Each scenario was tested for 30 seconds identically, then repeated three times for parity.

We used single-player vs. AI mode with 0 AI, using the “Hanamura” map. This was chosen for its replicability and reliability during test. This benchmark course produced extremely consistent and reliable results, with frametimes sitting more tightly clustered than what we're used to.

GN Test Bench 2015NameCourtesy OfCost
Video Card

This is what we're testing!

--
CPUIntel i7-5930K CPUiBUYPOWER
$580
MemoryKingston 16GB DDR4 PredatorKingston Tech.$245
MotherboardEVGA X99 ClassifiedGamersNexus$365
Power SupplyNZXT 1200W HALE90 V2NZXT$300
SSDHyperX Savage SSDKingston Tech.$130
CaseTop Deck Tech StationGamersNexus$250
CPU CoolerNZXT Kraken X41 CLCNZXT$110

Average FPS, 1% low, and 0.1% low times are measured. We do not measure maximum or minimum FPS results as we consider these numbers to be pure outliers. Instead, we take an average of the lowest 1% of results (1% low) to show real-world, noticeable dips; we then take an average of the lowest 0.1% of results for severe spikes.

Video Cards Tested

Disclaimer: (1) This is Beta, (2) Our Settings Comparisons Are Not Final!

Overwatch is in early beta. Blizzard is still taking feedback on basic UI design in its official forums, which sort of emphasizes the unreleased nature of the game. We have full confidence in the accuracy of our tests as they pertain to the current state of the game, but will need to retest Overwatch at official launch to include any possible optimizations made to graphics hardware. This disclaimer serves to inform system builders that graphics optimizations could happen, and those would impact framerate performance positively. It could also go negative – it just depends on what Blizzard's plans are post-beta.

We still get a good initial look.

As for the settings comparisons below, these are not final. We normally conduct fairly intensive graphics optimizations guides that run 3-10 tests per setting, ensuring accuracy and confidence in data. For Overwatch, given that the beta is only open through the 23rd, we ran brief, two-of tests strictly for our internal use. These are conducted silently, before all of our benchmarks, and aid our staff in determining which settings to test and how to build the test matrix. We ultimately save time in this fashion, eliminating tests which may prove redundant or unrealistically load-intensive.

Because Overwatch is new and folks are likely trying to understand how to best play the game on their cards, we've decided to published our unofficial, unfinalized settings comparison. This is made public to hopefully help gamers figure out points of GPU pressure that are established by some settings.

We only tested settings which did not require a restart to check, so there's potentially a lot more under the hood – and there probably are bigger settings to tune – but that's the nature of a limited-run beta. Can't do it all.

Initial Settings Comparisons

Render Scaling: This is effectively the same as changing resolution, so you'll see similar impact on performance. Supersampling helps embolden some textures and shadows, shown in the comparison below:

overwatch-render-scale-compare-1

Performance impact is massive. “Epic” render scale (200%, so at 1080, that's 4K) is about 87% slower than “high” render scale (100%, so resolution = render scale). Unless otherwise identified, all our tests ran “high” (100%) render scale settings.

Ambient Occlusion: AO toggling saw an FPS impact of about 8.48%. That's enough to steal a couple of frames out of performance, but lowering AO does noticeably change the apparent depth of environment objects (like trees, branches, bricks in walls, and other extruded surfaces).

Refractions: Negligible impact. There was almost 0 FPS change in our very limited testing of this setting. We will look deeper into refractions once more maps are released and when we've got more than one weekend to test.

Local Reflections: As above – negligible impact with almost 0 FPS change in our limited testing of this setting.

Texture Filter Quality: Texture filtration sometimes has marginal impact on FPS, though normal application of anisotropic filtration techniques will typically yield nearly 0 FPS change. In the case of Overwatch, we're seeing an FPS impact of about 3.7% between its maximum setting and “off,” with the in-between settings filling the (already small) gap.

Dynamic Reflection: Fairly large impact to FPS. Dynamic reflections on Hanamura impacted our FPS by about 29.19% at max vs. high, with the high vs. medium delta shrinking to 13.2% and the medium vs. low delta at ~14.56%. We'll research this setting further in future tests to determine the reason for its massive performance impact. This is one of the first places to go to improve framerate performance after render scale.

Anti-Aliasing Quality: We need to test this one further. We saw a change of just 1-2FPS across all AA settings, but we suspect that the impact will be larger if we remove other bottlenecks in the system first. Following analysis of dynamic reflections, it is our thought that the GPU may have been choking on that setting, thus limiting the delta presented by AA fluctuation. We will test this further and report back. The time limitation makes it difficult to fully analyze all settings with 100% confidence.

Overwatch Preset Graphics Settings – FPS at Low vs. Med, High, Ultra, Epic

Our first set of charts stem from preset scalability testing, an internal test we always conduct that we've decided to start publishing. This test benchmarks relative performance across a single graphics configuration (GTX 980 Ti ($660), in this case) and shows deltas between framerates at low, medium, high, ultra, and epic. Render scaling was kept at “high” for all these tests (100% scale).

Toward the lowest settings, we run into an issue where the game caps FPS at 300, something we didn't work around for these tests. This means the delta for 'low' could actually be greater, were FPS uncapped.

overwatch-gpu-bench-presets-delta

overwatch-gpu-bench-presets

The first chart shows the delta loss/gain from setting-to-setting. We established “Ultra” as our “go-to,” or 100% performance. We figured, judging by the benchmarks below, that this is where most gamers will attempt to land with their settings. The delta scale looks only at AVG FPS and does not account for 1% low and 0.1% low metrics.

On this graphics configuration, Ultra is approximately 26% faster than Epic, which scales-up just a few of the settings, and is about 30% slower than High. Low will allow for almost anything to play Overwatch, but the lowest-end card we tested was a GTX 750 Ti or R9 270X from AMD, both reasonable performers.

Scale can fluctuate depending on GPU, but this provides an initial look/understanding of settings impact.

4K Epic Overwatch Benchmark – 980 Ti, 390X, 980, 290X, & SLI

overwatch-gpu-bench-4k-epic

4K is pretty heavy-hitting on our GPU suite. Reducing render scale to “medium” (75%) pushes FPS up about 20-30 FPS on the 980 Ti ($660), but it's worth considering a 1440p native resolution at that point (or reducing settings to “Ultra” and “High”).

There is presently no SLI profile, from what we've gathered. SLI performance shows 0 scaling across all resolutions and presets tested. We eliminated the 980 Ti SLI configuration from our other charts, but left it here to show the lack of scaling.

Overwatch is the type of game where you really want a higher framerate. Noticeable tearing and jarring gameplay begin appearing at the 51FPS 4K output of the 980 Ti. Were I playing the game on this configuration, I'd have to reduce settings to get closer to 60 – it's just too jarring to play at 50FPS in this type of game. It is playable at 4K with lower settings, but we'll wait until SLI and CrossFire support are introduced to run a deeper test on 4K with Overwatch.

1440p Epic Overwatch Benchmark – 980, 970, 960 vs. 390X, 290X, 380X

overwatch-gpu-bench-1440-epic

Running “Epic” with “high” render scale, 1440p resolution is markedly less brutal to the GPUs than the 4.6 million higher pixel-count 4K resolution. The game runs reliably and with high framerates from the 390X ($420) and up, including the 970, 980 ($490), and 980 Ti cards. AMD's R9 290X runs close enough to 60FPS that some minor settings tweaks would push it over the top, but low 0.1% frametimes hammer AMD's cards across all of our Overwatch GPU tests. This seems to be an optimization-level problem, either on Blizzard's end or on AMD's drivers.

The brand new R9 380X falls shy of ideal playable range, and would require a settings reduction to play at 1440p. Here's our full review on the R9 380X, for those curious about AMD's new $230 card.

1080p Epic Overwatch Benchmark – 980, 970, 960, 950 vs. 390X, 290X, 380X, 285

overwatch-gpu-bench-1080-epic

1080p / epic is probably what most gamers will be running, if not 1080p / ultra. This shows just how well Blizzard's Overwatch plays on most hardware (at 1080p), though AMD does again suffer from poor 1% low and 0.1% low frame delivery. The drops hit harder on AMD hardware. Optimization should resolve this as the game and its related drivers advance.

Everything from the GTX 960 ($230) and up (including the 380X ($230), GTX 970 ($320), R9 290X, and so forth) can run Overwatch reliably at 60FPS with 1080 / epic. The R9 285 is close enough that we'd consider it within this range, though some settings tweaks would be advised to improve low performance metrics. Keep in mind that the R9 380 is effectively equivalent to the R9 285, though not present.

NVidia's GTX 950, a card that feels all but forgotten, pushes 52FPS average – fairly close, but probably best left to “ultra” rather than “epic.”

1080p Ultra Overwatch Benchmark – 960 vs. 380X, 285, 750 Ti

overwatch-gpu-bench-1080-ultra

This is what we'd consider the “mainstream” test.

The GTX 750 Ti is still shy of ideal playable framerates, demanding a settings drop toward “medium” for better performance. AMD's R9 270X – for which we used a reference card (from AMD) – pushes just under 60FPS, for which we'd deem it a playable solution at these settings. The R7 370 ($140) has similar performance (slightly lower in some titles) and would fall around this same range.

Conclusion: Best Graphics Cards for Overwatch Beta

Disclaimers all stated and settings all explained, it appears that Blizzard's Overwatch beta presently plays fairly well across most GPUs – even the older mainstays, like the R9 270X and R9 285. In its current state, we would recommend the GTX 950 ($160) or R9 380 (see R9 285 in our charts – about the same performance) at the lower price-point; these cards will enable 1080p Overwatch gaming at Ultra settings (a step below 'max').

Builders seeking to make use of cheaper cards, like the GTX 750 Ti or R7 370, should expect settings closer to “medium” and “high” in the game's current state.

Anyone looking to push “epic” settings at 1080p (render scale at 100%) will need a GTX 960, R9 380, or R9 380X, which beats-out the GTX 960 by a few percent.

Moving to 1440p, demand shifts to the R9 390X and GTX 970 at “epic” settings, with the GTX 970 pushing significantly stronger 0.1% low framerates. A drop to “Ultra” would be more inclusive to cheaper GPUs or existing products, like the R9 290X.

4K is fairly unplayable at “epic” settings right now, until multi-GPU profiles are implemented, at least. Anyone with a 4K display will likely be dropping settings down toward “high” for best playability, maybe “ultra” in some spots.

That's it for this one. If you like this type of coverage, please consider supporting us on Patreon. Check our free support forums for one-on-one assistance.

Editorial & Testing: Steve “Lelldorianx” Burke.