Set your system and your settings — the calculator shows you separately what your CPU can deliver and what your graphics card can deliver. The lower of the two is your frame rate. In ACC, during a race, that is almost always the CPU.
This is a model, not a benchmark of your machine. It works out two ceilings and shows you the lower one. Every constant in it has a source — and where there is none, it says so out loud.
The CPU side rests on a single, very good measurement: since 2026 PurePC has measured ACC in a clearly described worst case — Spa, 50 opponents, heavy rain, measured during the race start, Epic preset, a fixed RTX 5090. There a 9800X3D lands at 139 FPS, a 7800X3D at 113, an i9-14900K at 94, a Core Ultra 9 285K at 81, a 7700X at 74. The decisive part of that measurement: 1080p and 4K give identical numbers (146 vs. 146 on the fastest processor). In this situation the graphics card is completely irrelevant — the CPU sets the limit. Processors PurePC did not test we place using the gaming hierarchies from Tom's Hardware and ComputerBase; those are marked derived.
The GPU side is the weaker leg, and you should know that: there is no published ACC benchmark of any kind for a graphics card newer than the RTX 3080 / RX 6800. No review outlet runs ACC in its GPU test suite — not TechPowerUp, not Notebookcheck, not Tom's Hardware, not ComputerBase. So we anchor on the only clean ACC GPU measurement that exists (simracing-pc.de, March 2021: RTX 3080, RTX 3070 and GTX 1080 at 3440×1440 across all presets and Resolution Scale steps), and scale from there to other cards using Tom's Hardware's performance hierarchy. The model hits the three measured cards to within 3–11%; for everything else it is an extrapolation.
The preset steps come from the same measurement: on the GTX 1080, Low gave 32% more than Epic, Medium 17%, High 10%. On the CPU side, TechSpot/Hardware Unboxed shows a consistent 25% or so between Medium and Epic across four processors.
The resolution curve is not proportional to the pixel count — that is the most common mistake. According to Tom's Hardware's GPU hierarchy (June 2026, 17 cards), 1440p delivers on average 74% of 1080p and 4K around 43%, not the 44% and 25% that pure pixel math would suggest.
What is expressly not proven: there is no published measurement of how many FPS a single visible opponent costs — the two community measurements on this contradict each other by a factor of five, and both are older than ACC v1.8. We model the slider along the full-grid measurement (1 visible opponent = +43% compared with “All”), but that is an approximation. For VR there are no published ACC numbers at all; there we only extrapolate the pixel count, and because of the doubled render pass the real values tend to sit below that.
Sources: PurePC (Ryzen 9950X3D2 review, April 28, 2026) · TechSpot/Hardware Unboxed (Every Ryzen X3D CPU, April 13, 2026 and 9800X3D review, November 6, 2024) · simracing-pc.de (ACC GPU Benchmark, March 6, 2021) · Tom's Hardware GPU Benchmarks Hierarchy and CPU Hierarchy (as of June 2026) · ComputerBase CPU hierarchy (May 12, 2026) · Kunos changelogs v1.2, v1.3.0 and v1.8.0.
engine.ini tweaks: skip them. ACC often overwrites the config itself; most of the old forum tweaks are placebo today, or get reset. The menu already has the settings that count.
The NVIDIA app's “optimal settings”: they overrate ACC — they do not know about the CPU case at the race start. Set power management to “prefer maximum performance” instead, and turn the overlay and instant replay off. On RTX 50 cards in 2026, use a current, clean driver.
Upscaling in ACC is old: DLSS and FSR came into the game with version 1.8 (November 2021) and have never been updated since — there is no XeSS, no frame generation and no ray tracing (Kunos planned ray tracing back then and dropped it again). Which sub-versions exactly are in there, Kunos has never stated; the widely repeated “DLSS 2.2 / FSR 1.0” comes from community documentation, not from the developer. And there is no solid benchmark of how much DLSS gains you in ACC at all — the only real measurements come from VR, and they were negative. In practice that means: “native resolution + Resolution Scale + sharpening” often looks better than the built-in DLSS. That is also why this calculator has no DLSS switch.
Important, because almost every source gets it wrong: ACC does not split “Visible Opponents” 50/50 between ahead and behind. In the v1.3.0 changelog (February 4, 2020) Kunos states explicitly that cars ahead of you have render priority over cars behind you, and that cars in the pit lane are ignored. The widespread formula “10 means 5 ahead and 5 behind” comes from a single forum post from 2019 and is not supported by anything.
Assetto Corsa Competizione is heavily CPU limited, especially at the race start with a full field. The most common beginner mistake is buying a bigger graphics card, which does not fix that at all. If it only stutters in the pack while everything runs smoothly on your own, it is the processor, or rather the game thread — not the GPU.
The calculator on this page estimates both limits separately: what your CPU can deliver in the chosen scenario and what your graphics card manages at your resolution. The lower of the two numbers is your frame rate — and the part behind it is your bottleneck. It is based on published ACC measurements (PurePC, simracing-pc.de) and the GPU hierarchy from Tomʼs Hardware; how exactly it calculates is laid out openly on the page.
The single strongest slider is Visible Opponents: lowering it to 10–30 fixes race start stutter immediately. After that come mirror quality, Shadows and Resolution Scale with temporal upsampling. engine.ini tweaks are mostly placebo today. Current GPUs (RTX 40/50, RX 7000/9000) are often already CPU limited from 1440p upward — and then a faster CPU helps more than a bigger graphics card.
Because ACC is CPU limited, or more precisely game thread limited. Every visible car costs CPU performance. At the start with a full field the frame rate therefore collapses — a bigger GPU does not help, lowering Visible Opponents does.
Visible Opponents. Lowered to 10–30 it fixes race start stutter dramatically. ACC renders the nearest cars; according to the Kunos changelog for v1.3.0, cars ahead of you have render priority, and cars in the pits do not count. After that, Mirror Resolution, Shadows and Resolution Scale help.
Mostly not. ACC often overwrites the config itself, and most of the old forum tweaks are placebo today or get reset. The graphics menu already has the sliders that count.
Both, but the GPU sets the ceiling in a hotlap, while the CPU sets the floor at a full race start. For calm frame times in traffic, CPU clock speed and cache matter most — AMD's X3D processors lead here.