Marvel Rivals PC Stutter and Low FPS Explained: Fix the Ult-Spam CPU Load First, Then Use DLSS and FSR Correctly

Marvel Rivals runs fine in spawn and falls apart mid-teamfight. That pattern is the whole diagnosis: this is not a weak-GPU problem, it is a combat-event spike problem.

Marvel Rivals is a free-to-play 6v6 third-person hero shooter developed and published by NetEase Games in collaboration with Marvel Games, released for Windows, PlayStation 5 and Xbox Series X/S on December 6, 2024. It runs on Unreal Engine 5, features destructible environments that let players reshape the battlefield, and builds fights around combined Team-Up skills rather than single-gun duels.

Most stutter guides treat it like an open-world streaming problem. I think that misses the point, and that is why lowering resolution alone rarely fixes it.

My take: the Twelve-Ult Fireworks Factory

Here is my own interpretation, not NetEase's official statement: Marvel Rivals is a Twelve-Ult Fireworks Factory.

In an open-world UE5 game, the heavy work is streaming a big map. In Marvel Rivals, the map is small and the heavy work arrives all at once: 12 heroes firing particle-heavy abilities, two or three ultimates overlapping, Groot walls and Strange portals changing cover, destruction chunks flying, hit-markers, healing numbers, and server replication confirming all of it on the main thread.

That explains the classic complaint pattern validated in search suggestions — marvel rivals low fps fix, marvel rivals fps drop fix, why is marvel rivals fps so bad on Bing, plus marvel rivals best settings pc and marvel rivals optimization guide demand:

• 120+ fps walking out of spawn, 40 fps when Luna, Strange and Punisher ult at the same choke
• Stutter that spikes when walls break, not when you turn the camera
• A GPU that is not at 99% while the game still hitches — the CPU event queue is the bottleneck

If you only lower render resolution, you polish the pixels while the fireworks factory keeps overloading the foreman. You get a blurrier slideshow. Fix the event load first, then let upscaling do its real job.

A second difference from first-person shooters: the third-person camera renders your own hero plus more of the arena around you. Effects, shadows and post-processing stay on screen longer than in a tight first-person view. That is the Third-Person Tax — settings that look cheap in a corridor shooter cost more here.

What to do first: baseline before settings

Do this once. It removes false GPU blame.

1. Install on an SSD with 30+ GB free. UE5 + destruction + 12-player replication punishes slow streaming. A full HDD or near-full SSD turns every ult into a hitch.
2. Let shaders finish. Launch, enter Practice vs AI, play one full round before judging fps. Early-match hitches after a patch or driver update are often shader compilation, not your final performance.
3. Cap fps slightly below your real average. If teamfights hold ~110, cap at 90 or 100 in-game. An uncapped frame rate that swings 60–160 feels worse than a locked 90 and makes Frame Generation unstable.
4. Use wired or close-range Wi-Fi, close browsers and recording. Marvel Rivals is server-authoritative 6v6. Packet loss looks like stutter. Close Chrome tabs, Discord streams and background uploads while testing.
5. Update GPU drivers, set Windows power to Best Performance, enable Game Mode. On NVIDIA, leave Low Latency to the in-game Reflex setting instead of forcing it in three places.
6. Verify files after big seasonal drops. Seasons add heroes and maps — the official site now lists Alchemax Headquarters, K'un-Lun arenas, Times Square and many more. A corrupted chunk after a 20 GB update causes hitches no setting can fix.

If spawn is smooth but the first big ult fight still drops 40% of fps, you have confirmed the Fireworks Factory pattern. Move to the cut order.

Cut order: tame the factory, not the textures

Change in this order, testing one teamfight at a time in Quick Match or Practice vs AI. Stop when teamfight fps stabilizes.

1. Effects / Particle Quality — cut first. This is the direct ult-spam cost. Medium or Low gains more teamfight fps than any other single slider.
2. Shadows — cut second. Third-person + destructible cover means shadows re-render constantly. Medium or Low, never Epic for competitive play.
3. Post-Processing — cut third. Bloom, lens flares and ambient occlusion look great on hero-select screens and cost dearly when six ults stack.
4. Volumetrics / Fog and Anti-Aliasing — cut fourth. Tokyo 2099 and dark maps like Empire of Eternal Night use heavy atmosphere. Lower these before touching textures.
5. View Distance / Environment Detail — medium. You need to see Iron Man at range, but Epic mostly adds background clutter that the factory has to track.
6. Texture Quality — cut last. Textures cost VRAM, not event CPU. If you have 8 GB+ VRAM, leave textures at High while effects stay Low. Blurry skins never saved anyone from a Luna ult.

Prefer Fullscreen, keep VSync off if you use G-Sync/FreeSync, and do not toggle render settings mid-match without a restart.

Use DLSS and FSR by resolution, not by brand loyalty

DLSS Super Resolution boosts performance by rendering at lower internal resolution and reconstructing with motion data and prior frames, now with a second-generation transformer model for stability and clarity. FSR 3 combines temporal upscaling with Frame Generation via interpolation and optical flow, supporting DirectX 12, Vulkan and Unreal Engine 5.1 through 5.6. Both are tools for a stable base, not rescue ropes for an overloaded CPU.

Pick one upscaler — never both:

• 1080p: DLSS Quality or FSR Quality. Balanced already looks soft on 24-inch monitors and hides enemy heads in smoke.
• 1440p: DLSS Balanced or FSR Balanced. Best fps-per-clarity trade for most RTX 3060 / RX 6700-class PCs.
• 4K or weak GPU: DLSS Performance or FSR Performance. Ultra Performance only as a last resort — in a hero shooter, losing distant readability loses fights.
• Native AA / DLAA: only if you already hold 120+ fps in teamfights and want cleaner edges. It is anti-aliasing without upscaling, so it costs performance instead of giving it.

If DLSS looks sharper than FSR on your card, use DLSS. If you are on AMD or Intel without RTX upscaling headroom, FSR 3 Quality is the correct equivalent. The brand matters less than matching the mode to your output resolution.

Frame Generation + Reflex: only on a stable 60 fps base

DLSS Frame Generation uses AI to generate frames for higher fps while Reflex Low Latency keeps responsiveness by synchronizing CPU and GPU work. FSR 3 Frame Generation inserts interpolated frames and works in both GPU- and CPU-limited cases, with frame pacing for regular delivery.

In a 6v6 ult-trading game, that last sentence needs a warning: generated frames smooth motion but do not reduce the input-to-server round trip. My rule:

• Enable Frame Generation only if your teamfight base — with upscaling on, Frame Gen off — already holds 60+ fps. Below that, you amplify a shaky base and add latency to duels you already lose.
• Always pair Frame Generation with Reflex On or On + Boost. This is not optional in Marvel Rivals. Reflex is the latency brake for generated frames.
• If aim feels floaty after enabling it, turn Frame Generation off and keep the upscaler. A crisp 90 fps with low latency beats a smeary 140 fps that misses every Star-Lord dash.
• FSR Frame Generation users: same 60 fps gate, enable Anti-Lag style latency reduction where offered, and prefer VRR on.

Competitive players who main hitscan — Hela, Punisher, Widow — should test Frame Gen in Doom Match free-for-all before taking it into Competitive. If tracking feels delayed by even a few ms, the extra number on the counter is not worth it.

Micro-stutter and audio hitches most guides skip

• Disable audio enhancements for your output device. Windows spatial sound + Discord noise suppression + in-game 3D audio can stack into ult-time crackle. Test with enhancements off.
• Turn off background recording: Windows Game Bar recording, NVIDIA ShadowPlay instant replay and similar buffers cost CPU during the exact moments ults spike. Record on demand, not always.
• Limit fps one step further if 1% lows are bad. Average fps lies. If monitoring shows 130 avg but 45 1% low during destruction, your cap is too high. Drop 20 fps and retest.
• Measure the cut order clean first, then add tweaks one by one. Do not stack five optimizer tools while testing.

Stay sharp while your PC is down: two browser fallbacks

Reinstalling shaders or waiting on a seasonal patch is dead aim time. I use two free browser games from EasyHub for isolation practice because they run on weak hardware with no download:

• For hitscan tracking and cooldown timing under pressure, run Hazmob FPS: Online Shooter. Its seven documented modes — Team Deathmatch for warmup, Free for All for pure aim, Domination and Capture the Flag for objective rotation — recreate the Marvel habit of winning fights versus winning the match. Practice the same discipline: trade with a teammate, check the leaderboard instead of tilting, and time utility at chokes like you time ults at convoy points.
• For third-person spacing and peek rhythm, run BuildNow GG. The third-person camera, BoxFight close-range edits and Zone Wars rotations map well to Marvel's cover play: open only enough angle for your crosshair, reset fast, and take high ground before engaging. Ten minutes of Freebuild plus Aim Training keeps crosshair placement alive while Marvel recompiles.

Neither replaces Marvel's Team-Up system — riding Groot as Rocket or charging Iron Man with Hulk Gamma, as described on Steam, has no browser equivalent. They keep your hands honest while the factory gets fixed.

Competitive starting point (adjust from here)

• Display: Fullscreen, VSync off + adaptive sync on, fps cap at stable teamfight value
• Upscaler: DLSS/FSR Quality at 1080p, Balanced at 1440p, Performance at 4K
• Effects Low-Medium, Shadows Low-Medium, Post-Processing Low-Medium, Textures High if VRAM allows
• Reflex On, Frame Generation only if 60+ fps base holds
• Sensitivity: keep one value for a week; FOV wide enough to see flanks without shrinking targets

Three open questions I am still testing

1. Does destruction debris or ult particles dominate the teamfight drop on mid-range 6-core CPUs? My feel says particles, but I have no clean isolated benchmark.
2. Does Reflex On + Boost measurably help projectile heroes more than hitscan in 6v6 chaos, or is the benefit mostly psychological above 100 fps?
3. Do new maps like Times Square and Grand Garden shift the bottleneck back toward streaming, breaking the Fireworks Factory model for specific modes?

If your spawn-to-teamfight drop is small but movement always feels late, suspect network or Reflex settings before pixels. If the drop is huge only when walls break and ults stack, you are living in my Fireworks Factory — cut effects first, match the upscaler to resolution, and gate Frame Generation behind a real 60 fps base.