How the PCPartGuide bottleneck calculator works

Last updated 2026-09-02

In one sentence: every part in your build gets a frame-rate ceiling for the game and resolution you chose; the lowest ceiling is the frame rate you see, and the gap between the two lowest ceilings is your bottleneck.

This page documents the model, the data behind it, how confident you should be in a result, and what the calculator cannot see. Data version 2026.09, last updated 2026-09-01.

What the calculator estimates, and what it does not

The calculator estimates three things for a CPU + GPU pair at a chosen resolution: which part limits the frame rate, by how much, and roughly how many frames per second you will see in a given game. It also flags the two most common secondary limits, memory (capacity, channels and speed) and the display (refresh rate or an in-game frame cap).

It does not measure your PC. It does not know your driver version, Windows power plan, background load, RAM timings, overclocks, cooling, the exact scene you play, or the settings preset you use beyond what is described below. Treat every result as a well-informed estimate that you can confirm in five minutes with a performance overlay; the usage guide shows how.

The data: two indices

Each processor carries a gaming index on a 0 to 100 scale, where 100 is the AMD Ryzen 7 9800X3D, the fastest gaming processor in the dataset. The index represents relative frame rates at 1080p with a graphics card fast enough that the processor is the only limit, averaged across a broad set of modern games. Processor performance changes very little with resolution, so one index is enough.

Each graphics card carries three indices, one per resolution (1080p, 1440p and 4K), where 100 is the NVIDIA GeForce RTX 5090 at that resolution. Three indices are needed because cards scale differently with pixel count: memory bandwidth, cache and VRAM capacity all matter more as the resolution rises.

Both sets of indices are derived from published benchmark aggregates: primarily the relative-performance summaries in TechPowerUp reviews, Tom's Hardware CPU and GPU hierarchies, and Hardware Unboxed scaling tests, cross-checked against launch reviews from multiple outlets. They are derived, not copied: we reconcile sources measured with different test rigs and game sets into one consistent scale, then apply ordering checks (for example, an X3D chip must not fall below its non-X3D sibling, and a 5080 must sit between a 5090 and a 5070 Ti at every resolution) before a dataset version is published. Every part record lists the sources it was derived from.

The ceiling model

For a chosen game and resolution:

  1. CPU ceiling = the game's CPU-bound frame rate at index 100 × the processor's index ÷ 100, × the RAM factor.
  2. GPU ceiling = the game's GPU-bound frame rate at index 100 for that resolution × the graphics card's index ÷ 100.
  3. Display ceiling = your monitor's refresh rate, if you set one, or the game's built-in frame cap, if it has one.
  4. Estimated FPS = the lowest of those ceilings. The part that owns it is the limiter.
  5. Bottleneck percentage = (higher of CPU and GPU ceiling − lower) ÷ higher × 100. The side with the lower ceiling is the bound side.

Bands: under 10% is balanced, 10 to 25% mild, 25 to 45% significant, over 45% severe. GPU-bound results are the normal state of a gaming PC and are described as such; the calculator only recommends a graphics card upgrade when the estimated frame rate is the thing you want to change.

Worked example. A Ryzen 5 5600X (index 58) with an RTX 4070 (indices 56 / 49 / 40) in a typical modern AAA game, whose reference frame rates are 250 (CPU-bound) and 348 / 222 / 120 (GPU-bound at the three resolutions), with a 16 GB dual-channel DDR4-3200 kit (RAM factor 1.0). CPU ceiling: 250 × 0.58 = 145 fps at every resolution. GPU ceilings: 195, 109 and 48 fps. At 1080p the processor is the limiter (145 fps, a 26% CPU bottleneck); at 1440p and 4K the card is (109 and 48 fps). Swap the card for an RTX 5090 and the processor limits at every resolution below 4K; swap the processor for a Ryzen 7 9800X3D and the RTX 4070 is the limiter everywhere.

Game profiles

Each game in the list carries its own CPU-bound and GPU-bound reference frame rates, a RAM requirement, and, where relevant, a frame cap. They describe how the game behaves at the settings noted next to it: competitive titles at their usual reduced settings, everything else at high settings with no ray tracing and no upscaling. The "typical modern AAA game" profile is a blend used for the headline numbers on part and pair pages. Reference frame rates come from the same reviews as the indices and are adjusted when games are patched significantly.

RAM and display adjustments

The RAM factor multiplies the CPU ceiling. Capacity below what the game wants applies a 0.72 factor, which is our way of representing the stutter and paging that an average frame rate hides. A single stick (single channel) applies 0.82. Kit speed applies between 0.94 (DDR4-2666) and 1.03 (DDR5-6000 or DDR4-3600) relative to the platform's standard kit. The combined factor is clamped between 0.5 and 1.12. X3D processors are far less sensitive to memory speed in reality; the dataset reflects this through their higher base index rather than a separate factor, so speed adjustments are slightly pessimistic for them. A RAM kit of the wrong generation for the processor is flagged and not applied.

A monitor refresh rate becomes a ceiling only if you set one. In-game frame caps (for example, a 60 fps cap) are applied automatically for the games that have them, and the result says so.

Upgrade suggestions and idle money

The fix card names the cheapest part in the dataset that lifts the limiting ceiling to at least the next ceiling, in the same game and resolution. For processors it prefers the same socket; the page tells you when a new platform would be needed. "Idle money" is the sum, over the non-limiting parts, of each part's typical price multiplied by the share of its ceiling that goes unused. It is an illustration of waste, not a resale value.

Prices are typical street prices in US dollars at the dataset date, shown as classes rather than exact figures because they move weekly. Affiliate links are marked; they do not affect any calculation or recommendation.

Confidence

Each result carries a confidence label. High means both parts launched in 2020 or later and are derived from at least two independent sources. Medium means both parts launched in 2018 or later. Low covers older parts, for which fewer modern reviews exist and which we re-derive from older test suites. In practice, results for mainstream settings land within roughly 10 to 15% of published review averages; unusual settings, heavy mods, ray tracing, frame generation and games patched after the dataset date push that further out.

Known limitations

Data changelog

Who maintains this

PCPartGuide was founded in 2018 by Andre, who has written about PC hardware for more than a decade and has built more gaming systems, workstations and local-AI machines than he cares to count. The site began as a hardware buying-guide blog, spent a period as a parts-comparison platform, and was relaunched in 2026 as a single-purpose bottleneck calculator because that was the question readers asked most. The dataset and the model are maintained by Andre; corrections are welcome by email at support@pcpartguide.com and are usually reviewed within a week. More on the about page.

Corrections policy

If a part index, price class or game profile is wrong, we fix it in the next dataset release and note it in the changelog. If a page contains a factual error outside the dataset, we correct it in place and update the page's modified date. We do not remove pages to hide mistakes.