Marvel Rivals runs on Unreal Engine 5, but NetEase's implementation is notably well-optimized for a 2024 UE5 title — it ships without Lumen or Nanite as default rendering paths, instead using a more traditional deferred pipeline that keeps CPU overhead manageable on mid-range hardware. The game leans heavily on DLSS 3 (including Frame Generation), FSR, and XeSS to push frame rates on a wide hardware range. VRAM demand is moderate: 4 GB at 1080p, 6 GB at 1440p, and around 10 GB at 4K Ultra textures. The biggest performance variables are effect quality during six-player team fights and shadow quality via UE5's virtual shadow map system. Reflex integration keeps latency competitive even at lower frame counts. Most users will find the largest headroom in effect quality and shadow quality rather than in geometry or texture budgets.
Below is a per-setting breakdown: what each option does, how much it costs, and the value we recommend — tuned to keep the image looking right while reclaiming frames. Want the exact numbers for your GPU? Open the optimizer →
Recommended settings for Marvel Rivals
Reference rig: RTX 4080 at 1440p, balanced preset. Values are accurate to Marvel Rivals's in-game options.
Texture Quality
High
Low cost
Typical impact 0-5% · 4% fps cost
In Marvel Rivals, we recommend Texture Quality at High (4% fps cost).
Controls the maximum mipmap resolution loaded for surface textures. Higher levels stream larger texture maps (2K/4K) from disk into VRAM via the texture streaming pool. The GPU samples these during fragment shading using the currently bound sampler state. The FPS cost is minimal when VRAM is sufficient because texture fetch latency is hidden by the cache hierarchy, but exceeding VRAM capacity triggers page-faulting and hitching as textures are swapped between system RAM and VRAM.
In Marvel Rivals: Marvel Rivals streams UE5 texture mip data into a shared VRAM pool. At Ultra, the engine loads full 4K surface maps for hero skins, environmental geometry in maps like Tokyo 2099 and Yggsgard, and ability decals — expect around 8-10 GB VRAM consumption at 4K. On cards with 6 GB or less, Ultra at 1440p can cause streaming hitches mid-match rather than a flat FPS penalty. High is the sweet spot for 8 GB cards; the visual delta from High to Ultra is subtle on hero models in motion.
Shadow Quality
High
Heavy
Typical impact 8-25% · 12% fps cost
In Marvel Rivals, we recommend Shadow Quality at High (12% fps cost).
Controls shadow map resolution, filtering method, and cascade count for dynamic shadows. The engine renders the scene from each light source perspective into depth-only shadow map textures. Higher settings increase shadow map resolution (1024 to 4096 texels), add more cascaded shadow map splits for the directional light (improving near-field resolution), and enable softer PCF or PCSS filtering which requires more depth comparison samples per pixel during the lighting pass.
In Marvel Rivals: UE5 virtual shadow maps are used. Higher settings improve shadow resolution on destructible environments but cost 15-20% GPU.
Effect Quality
High
Heavy
Typical impact 3-15% · 12% fps cost
In Marvel Rivals, we recommend Effect Quality at High (12% fps cost).
Controls the visual fidelity of gameplay effects including explosions, weapon impacts, ability VFX, and environmental interactions. Higher settings increase particle emitter counts per effect, use higher-resolution flipbook or mesh particles instead of simple sprites, enable GPU particle simulation via compute shaders, and add dynamic lighting from effects (each explosion spawning a temporary point light). The cost is highly variable — intense combat with multiple overlapping effects can produce 4-8x overdraw from layered transparent particles.
In Marvel Rivals: Marvel Rivals has extremely particle-heavy hero abilities (Storm, Thor, Doctor Strange). Ultra effect quality during team fights can tank FPS 30%+ on mid-range GPUs.
Post-Process Quality
High
Low cost
Typical impact 3-10% · 6% fps cost
In Marvel Rivals, we recommend Post-Process Quality at High (6% fps cost).
Controls the overall quality of the fullscreen post-processing effect stack including tone mapping, color grading (LUT application), bloom (bright-pass filter with multi-stage Gaussian blur), lens flare, auto-exposure (luminance histogram compute shader), and screen-space lens distortion. Higher settings run these effects at full resolution, use larger blur kernels for bloom, and enable additional effects. The total cost is the sum of multiple fullscreen passes — each reading and writing the entire framebuffer.
In Marvel Rivals: Marvel Rivals' post-process stack includes UE5's tone mapper, auto-exposure luminance histogram, bloom (multi-stage Gaussian across bright hero energy effects), and lens distortion on certain map transitions. Higher settings run the full chain at native resolution with wider bloom kernels — the bloom pass is noticeably heavy given how many bright-emission hero abilities exist simultaneously on screen. Dropping from Ultra to Medium runs some passes at half resolution and reduces bloom sample count. The FPS saving is modest (3-6%) but the auto-exposure aggressiveness also stabilizes, which some players prefer in competitive contexts.
Reflection Quality
Medium
Low cost
Typical impact 3-20% · 6% fps cost
In Marvel Rivals, we recommend Reflection Quality at Medium (6% fps cost).
Controls the method and fidelity of surface reflections. Low settings use pre-baked cubemap probes — a single texture lookup per pixel. Medium enables screen-space reflections (SSR) that ray-march through the depth buffer to find reflected geometry. High uses higher-resolution SSR with more march steps. Ultra may enable planar reflections (re-rendering the scene from a mirrored viewpoint) or RT reflections (hardware-accelerated rays). The cost escalation from cubemaps to SSR to RT is dramatic — cubemaps are nearly free, SSR costs 3-8%, and RT reflections cost 15-25%.
In Marvel Rivals: At Off and Low, Marvel Rivals falls back to static cubemap probes — a single texture fetch per reflective surface. Medium and above enable screen-space reflections (SSR) which ray-march the depth buffer on surfaces like the water in Atlantis and polished floors in the S.H.I.E.L.D. Helicarrier map. Ultra may enable limited RT reflections depending on hardware. SSR adds 5-10% GPU cost and can produce obvious edge cutoffs on fast-moving hero geometry. For competitive play, Low or Off is recommended — the reflective surfaces are largely environmental and don't affect ability readability.
Volumetric Fog
Medium
Low cost
Typical impact 5-18% · 6% fps cost
In Marvel Rivals, we recommend Volumetric Fog at Medium (6% fps cost).
Renders physically-based 3D fog that interacts with lighting, shadows, and participating media density. The engine allocates a 3D froxel (frustum-voxel) volume texture — typically 160x90x64 or higher — and ray-marches through it from each pixel, accumulating scattered light and extinction at each step. Each froxel samples the shadow map to determine direct illumination, applies the Henyey-Greenstein phase function for anisotropic scattering, and accumulates density from noise textures or analytical fog volumes. The cost is substantial because every visible pixel requires a full volumetric integration.
In Marvel Rivals: Marvel Rivals uses a froxel-based volumetric fog implementation in UE5, visible in maps like the mist-shrouded regions of Yggsgard and smog layers in Tokyo 2099. The engine allocates a 3D froxel texture and integrates lighting and shadow occlusion per-voxel each frame. Higher settings increase froxel grid resolution and the number of integration steps. Moving from High to Off recovers around 5-10% frame time and has a meaningful visual impact on atmospheric depth, but competitive players typically prefer the cleaner sightlines that lower fog density provides.
Ambient Occlusion
Medium
Low cost
Typical impact 3-12% · 5% fps cost
In Marvel Rivals, we recommend Ambient Occlusion at Medium (5% fps cost).
Computes soft shadowing in crevices and where surfaces meet by estimating how much ambient light is occluded at each pixel. SSAO samples the depth buffer in a hemisphere around each pixel, testing for nearby occluders. HBAO+ uses ray-marching along the depth buffer horizon. GTAO uses a multi-directional horizon search with cosine-weighted integration for physically correct results. Each method runs as a fullscreen compute or pixel shader pass — higher quality modes increase sample count from 4 (SSAO) to 32+ (GTAO Ultra), directly scaling the per-pixel ALU cost.
In Marvel Rivals: Marvel Rivals uses a screen-space ambient occlusion pass (SSAO/GTAO-style) to add contact shadows beneath hero feet, in architectural crevices, and around map props. At High, the engine dispatches a fullscreen compute pass with more hemisphere samples per pixel, increasing ALU pressure. In a hero shooter context, AO primarily affects environmental geometry and hero ground contact — the visual contribution to combat clarity is low. Dropping to Low or Off saves 3-8% GPU, and with DLSS or FSR active, the temporal accumulation in those upscalers partially compensates for lost AO subtlety.
Foliage Quality
High
Low cost
Typical impact 5-20% · 6% fps cost
In Marvel Rivals, we recommend Foliage Quality at High (6% fps cost).
Controls density, LOD transitions, and rendering quality for non-grass vegetation — trees, bushes, ferns, and vines. Higher settings increase the number of foliage instances, delay the transition from full 3D meshes to billboard imposters, and use higher-poly foliage meshes. In UE5 games using Nanite foliage, this controls the mesh cluster granularity and streaming distance. The primary cost drivers are massive overdraw from layered alpha-tested foliage cards and the high draw call count from thousands of individually-placed foliage instances.
In Marvel Rivals: Several Marvel Rivals maps include foliage — the overgrown ruins of Wakanda and garden areas in certain environment zones use UE5 instanced foliage with alpha-tested leaf geometry. Higher settings increase instance density, push LOD transition distances, and keep 3D foliage meshes visible further before swapping to billboards. The alpha-test overdraw from dense foliage layers is the primary cost. Since foliage is a background environmental element with no competitive visibility impact, Medium is the practical ceiling for performance-focused players — the density difference between High and Ultra is marginal.
View Distance
High
Heavy
Typical impact 5-20% · 8% fps cost
In Marvel Rivals, we recommend View Distance at High (8% fps cost).
Sets the maximum distance at which world geometry, props, and objects are rendered. The engine performs frustum culling and occlusion culling on all objects — increasing view distance dramatically increases the number of objects passing visibility tests, leading to more draw calls submitted to the GPU command processor. In UE5 titles, this also affects Nanite virtual geometry streaming range. The CPU cost of scene traversal and draw call submission often bottlenecks before the GPU at extreme view distances.
In Marvel Rivals: In a hero shooter with contained map layouts, view distance governs how far environmental props, background structures, and LOD-swapped geometry remain at their highest detail tier. Marvel Rivals maps are not open-world but contain background cityscapes (Tokyo 2099 skyscrapers, Midtown Manhattan depth geometry) that scale with this setting. Higher values keep distant background meshes in their dense LOD tiers, increasing draw calls and UE5 scene traversal cost on the CPU. For most maps the competitive sightlines are short enough that Medium causes no gameplay pop-in while saving 5-12% GPU/CPU overhead compared to Ultra.
Anti-Aliasing
TAA High
Low cost
Typical impact 2-15% · 4% fps cost
In Marvel Rivals, we recommend Anti-Aliasing at TAA High (4% fps cost).
Smooths jagged edges (aliasing) on geometric boundaries. FXAA is a single-pass edge-detection blur — cheap but softens the image. TAA accumulates multiple frames using motion vectors, sampling sub-pixel jitter offsets to reconstruct smoother edges — moderate cost with potential ghosting. SMAA uses pattern-matching edge detection with a more intelligent blend. MSAA runs the rasterizer at 2x/4x the sample count, evaluating coverage for each triangle edge — expensive because it multiplies ROP work and render target memory, but produces sharp geometry edges without blur.
In Marvel Rivals: Marvel Rivals offers TAA and TAA High alongside Off. Standard TAA uses UE5's built-in temporal accumulator with sub-pixel jitter and motion vector reprojection — effective at smoothing UE5's fine geometry but introduces characteristic ghosting on fast-moving heroes. TAA High runs the temporal resolve at higher precision with more history weight, reducing shimmer on particle VFX at a modest cost. Turning AA Off produces sharp but heavily-aliased images, particularly on hero silhouettes. In practice, if DLSS or FSR is active, their temporal reconstruction supersedes the TAA selection, making this setting largely redundant when any upscaler is enabled.
NVIDIA DLSS
Off
Low cost
Typical impact -30-80% · no measurable cost
In Marvel Rivals, the recommended preset leaves NVIDIA DLSS off — little visual loss for the frames it returns.
Deep Learning Super Sampling — NVIDIA's AI-based temporal upscaling that runs on dedicated Tensor Core hardware. The engine renders at a lower internal resolution and feeds the reduced-resolution frame, motion vectors, and depth buffer to a neural network that reconstructs a high-resolution output. DLSS 3+ adds optical flow-based frame generation on Ada/Blackwell architectures. The FPS gain comes from rendering fewer pixels — Quality mode renders ~67% of native pixels, Performance ~50%, Ultra Performance ~33%.
In Marvel Rivals: DLSS 3 is fully integrated in Marvel Rivals on RTX GPUs, including Frame Generation on RTX 40-series. The engine supplies UE5 motion vectors and depth to the DLSS runtime. Quality mode (67% render scale) is recommended for 1440p and 4K — it preserves hero model detail and ability VFX clarity better than Balanced or Performance, where fine particle edges begin to break down. Frame Generation on RTX 40 cards can push framerates past 144 fps at 1440p even with High-quality settings, though the latency overhead makes Reflex essential to pair with it. DLSS largely supersedes the need for native TAA.
AMD FSR
Off
Low cost
Typical impact -25-70% · no measurable cost
In Marvel Rivals, the recommended preset leaves AMD FSR off — little visual loss for the frames it returns.
FidelityFX Super Resolution — AMD's upscaling technology available on all GPUs. FSR 2.0+ uses temporal accumulation similar to TAA — it combines multiple jittered lower-resolution frames using motion vectors and a depth buffer to reconstruct a higher-resolution output via a multi-pass compute shader pipeline. The pipeline includes depth clip detection, motion vector dilation, luminance instability detection, and a reconstruction pass with Lanczos-based resampling. Unlike DLSS, FSR runs on standard compute units rather than dedicated AI hardware, working vendor-agnostically.
In Marvel Rivals: FSR 2 is the recommended upscaler for non-NVIDIA hardware in Marvel Rivals. Running on standard compute units, FSR Quality mode (67% native) delivers a clean image at 1440p though is somewhat more prone to particle-edge shimmer than DLSS during ability-heavy team fights — this is most visible on Doctor Strange portal rings and energy projectiles. FSR works on any GPU including integrated graphics. Ultra Quality mode (77% native) is the practical recommendation for RX 6000/7000 series cards at 1440p, offering near-native clarity with 20-40% frame time reduction depending on settings.
Intel XeSS
Off
Low cost
Typical impact -25-65% · no measurable cost
In Marvel Rivals, the recommended preset leaves Intel XeSS off — little visual loss for the frames it returns.
Intel Xe Super Sampling — a temporal upscaling technology that uses machine learning inference to reconstruct high-resolution frames from lower-resolution input. On Intel Arc GPUs, XeSS runs on dedicated XMX (Xe Matrix Extensions) AI accelerator hardware. On non-Intel GPUs, XeSS falls back to a DP4a (dot product of 4 8-bit integers) shader implementation that runs on standard compute units. The neural network takes the current low-resolution color buffer, motion vectors, depth, and responsive masks as input. Quality mode renders at ~77% of native, Performance at ~50%.
In Marvel Rivals: XeSS in Marvel Rivals uses the temporal reconstruction pipeline with DP4a fallback on non-Intel hardware. On Intel Arc GPUs it runs on XMX AI hardware, approaching DLSS quality for hero model upscaling. On AMD or NVIDIA hardware the DP4a path performs comparably to FSR 2 at matching presets but may exhibit slightly different temporal stability characteristics on fast-moving ability VFX. Quality mode is the recommended starting preset. XeSS is the third-choice upscaler on Rivals — DLSS for NVIDIA, FSR for AMD, XeSS for Intel Arc or as a fallback.
Motion Blur
High
Low cost
Typical impact 1-5% · 2% fps cost
In Marvel Rivals, we recommend Motion Blur at High (2% fps cost).
Applies directional blur to moving objects based on per-pixel motion vectors. The engine writes a motion vector buffer during the G-buffer pass — each pixel stores a 2D velocity derived from the difference between current and previous frame positions. The post-process shader samples the color buffer along each pixel's motion vector, averaging multiple taps to produce directional streaking. The cost is a single fullscreen pass with 8-16 dependent texture fetches per pixel. Many competitive players disable this for image clarity.
In Marvel Rivals: Marvel Rivals generates per-pixel motion vectors during the G-buffer pass and applies a post-process directional blur scaled to velocity magnitude. In a fast-paced hero shooter, motion blur primarily affects rapidly-moving hero abilities and player character during dashes — characters like Spider-Man and Black Panther generate heavy blur during traversal. Most competitive players disable this entirely: Off removes the fullscreen multi-tap blur pass, recovering 1-4% frame time and substantially improving visual clarity during high-mobility ability exchanges. Low applies a reduced-strength blur only on the most extreme velocity pixels.