If MARVEL Tōkon: Fighting Souls drops below its 60 fps battle target, do not begin with someone else’s “best settings.” First make the slowdown happen in the same offline scene twice. If it fails there, tune the local graphics branch. If it holds at 59/60 fps there but breaks only in an online match, stop treating lower image quality as a proven rollback fix.
Use this procedure only for Steam App ID 3787240 on the Windows retail release; its evidence is dated August 9, 2026. The earlier Open Beta, PS5, Epic, Linux, account login, and server access sit outside it. The official Steam listing names Windows 11 and 16 GB RAM in both hardware tiers, with an i5-8400 or Ryzen 5 1600X and an RTX 2060 6 GB or RX 6600 XT 8 GB at the minimum tier. A machine below that floor can still run the test, but the result cannot prove that unsupported hardware should sustain the target.
Build one repeatable load scene
A current English Steam community performance guide supplies a useful retail load recipe. Its character and stage rankings come from one PC, so treat them as test inputs rather than an official benchmark. The value is repeatability: you can use the same eight fighters, location, recording, and readout after every change.
Before entering Training Mode, open Task Manager with Ctrl+Shift+Esc. On the GPU page, watch Utilization and Dedicated GPU Memory. Hover over each fighter on the character-select screen once before recording values, giving the game a chance to load the roster assets. You can also open Steam with Shift+Tab, choose Settings, then In Game, and enable its performance monitor.
- At the title screen, open Quick Menu → Options → Button Settings (Battle) and assign both Record and Playback.
- Enter Training Mode at the City Center starting area of New York City – Night.
- Put Carnage, Peni Parker, Iron Man, and Ghost Rider on both teams.
- Start recording for Player 2, make a harmless movement or attack, call all three assists together, and finish the recording.
- Play that recording and time Player 1’s three assists so all eight characters appear at once. Note the frame reading and GPU measurements during the overlap.
If a restart clears the saved team, rebuild these same two teams. Changing the cast would change the load and spoil the comparison.
Let the offline result choose the next branch
The Steam product page describes battle as locked to 60 fps, while the community test uses 59/60 as the practical pass line. A repeatable dip below that line in this offline scene exists before matchmaking, cross-play, or rollback enters the test. Continue with local graphics changes only in that case.
If the scene stays at 59/60, leave the working graphics values alone for now. Play an online match with the performance monitor visible and compare its behavior with the offline recording. That comparison is more useful than another preset: it tells you whether your current graphics setup can reproduce the problem without an online opponent.
Do not compare a quiet menu reading with an eight-character battle, and do not switch stages between tests. The answer comes from the same scene producing a different result, not from a single attractive number elsewhere in the game.
Change one offline variable, then retest
When the recorded scene fails, return to Quick Menu → Options → Graphics Settings. Run Execute Shader Warmup at least once. The Steam guide also recommends leaving Launch Shader Warmup enabled, starting with the manual graphics values low, and moving upward only after the stress scene holds.
Use Display Mode, Resolution, and Anti-aliasing Type as separate experiments. Change only one, exit and restart the game, confirm that the chosen value is still active, then replay the recording. The guide author saw Windowed outperform Fullscreen and Borderless on one Radeon system; that makes Windowed worth testing, not a universal winner. Keep whichever mode survives the identical retest on your PC.
Watch for two signals together: the scene returns to 59/60, and GPU Utilization or dedicated memory no longer reaches its limit during the assist overlap. Then raise one quality value and repeat. If the same scene still dips with manual settings low, shader warmup complete, and the display/resolution/AA variables tested separately, the available evidence does not support inventing another guaranteed setting. Record that lowest-setting result as unresolved rather than calling a random change a fix.
Stop lowering graphics when only online play breaks
An offline pass followed by online slowdown is a different result. The community guide points to the visible Rollback frame(s) indicator and explains that rollback can require additional CPU simulation during an online match. The official pre-launch performance announcement also treated CPU load, frame drops, rollback, and cross-play as separate workstreams. Neither source proves that every retail issue is fixed, or that one graphics option repairs rollback.
Keep the verified offline settings and observe whether the online-only failure rises with the Rollback frame(s) reading. That does not identify a single network, CPU, or game-code cause by itself. It proves the narrower fact that the local GPU stress scene is not reproducing the failure. Return to graphics tuning only if the offline recording also starts missing 59/60.
Make the result survive a restart
The Day-1 guide reports that graphics values may shuffle and saved teams may disappear after relaunch. A one-session improvement is therefore not the finish. Restart once more, reopen Graphics Settings, verify every value you changed, rebuild the test teams if needed, and replay the same eight-character overlap.
You are done when that post-restart scene holds at 59/60 with the verified values. If offline remains stable and only the online comparison fails, the useful result is equally specific: preserve the working graphics setup, keep the rollback observation separate, and stop sacrificing image quality for a fix the offline test cannot validate.