Skip to content

Methodology

How we test & verify

The reference rig, retest cadence, cross-checked sources, and verification workflow behind every ToolWools calculator. Each tool links here from its byline so you can audit the process behind the number in one click.

Reference rig

Every tier score anchors at 100/100 to this build. All FPS bands and VRAM allocations are measured or modelled against it and scaled from there.

CPU
Intel Core i9-14900K (24C/32T)
GPU
NVIDIA GeForce RTX 4090 24 GB
Memory
64 GB DDR5-6400 CL32 (dual channel)
Storage
Samsung 990 Pro 2 TB NVMe (DirectStorage-ready)
OS
Windows 11 24H2 with latest drivers

What each model does, in plain terms

Seven engines power the calculators on this site. Each one is documented here with its inputs, the shape of its formula, what it anchors to, and how far it can be off.

Bottleneck model

Inputs
CPU model, GPU model, target resolution, target frame rate.
How it is calculated
Each part gets a normalised throughput score from our measured database, then the two scores are compared at the requested resolution. Because raster load rises faster with pixels than draw-call load does, the GPU score is weighted up as resolution climbs and the CPU score is weighted up as the frame-rate target climbs. The imbalance is the signed gap between the two weighted scores, expressed as a percentage of the stronger part.
Anchor point
A perfectly matched pair at 1440p/60 reads 0%. Positive means the GPU is the limiter, negative means the CPU is.
Known error band
Direction of the imbalance is reliable; magnitude carries roughly +/-8 points until PC settings are published.
Open the tool this model powers

VRAM allocation model

Inputs
Resolution, texture preset, ray-tracing state, upscaler mode, frame generation on or off.
How it is calculated
A measured base pool for the resolution is added to preset-specific texture, shadow and geometry overheads, then multiplied by a ray-tracing factor and adjusted for the upscaler's internal render resolution. Frame generation adds a flat buffer cost because it holds extra frames in memory.
Anchor point
Base pools come from GTA V Enhanced, RDR2 and Cyberpunk 2077 captures at matched settings on the reference rig.
Known error band
Allocation is not the same as strictly required VRAM. Treat results as the point where stutter risk starts, +/-1 GB.
Open the tool this model powers

Frame-rate model

Inputs
GPU, CPU, resolution, preset, upscaler, frame generation.
How it is calculated
The GPU's relative performance index sets a raster ceiling for the resolution and preset, the CPU's index sets a separate simulation ceiling, and the lower of the two becomes the average. 1% lows are derived from the CPU ceiling because that is what dominates frame-time spikes in dense open worlds.
Anchor point
The reference rig is pinned to the frame rate Rockstar has shown for the console build scaled to PC preset equivalents.
Known error band
The widest band on the site: roughly +/-15% pre-launch, because GTA 6 has no published PC preset table yet.
Open the tool this model powers

Readiness scoring

Inputs
Full part list: GPU, CPU, RAM, storage type, OS.
How it is calculated
Each subsystem earns points against a weighted rubric — GPU carries the most weight, then CPU, then RAM capacity, then storage class — and the weighted total becomes a 0-100 score with a letter grade. Any subsystem below its hard floor caps the overall grade regardless of the others.
Anchor point
The reference rig scores 100/100 by definition.
Known error band
Scores are ordinal, not predictive: a 78 will run better than a 62, but neither is a promised frame rate.
Open the tool this model powers

GPU tier placement

Inputs
GPU raster index, VRAM capacity, ray-tracing index, resolution filter.
How it is calculated
Raster performance, VRAM headroom and ray-tracing capability are blended into a single score, then the score is bucketed into S through F bands per resolution. A card with strong raster but thin VRAM drops a band at 4K and holds its band at 1080p, which is why the same GPU can move as you change the filter.
Anchor point
Band edges are fixed numbers, not a forced distribution — a tier can be empty.
Known error band
Placement is stable within one band. Cards sitting near an edge may move once real PC settings ship.
Open the tool this model powers

Connection grading

Inputs
Region, access medium (fibre through satellite), device, plus your measured download, upload, ping, jitter and packet loss.
How it is calculated
The latency target is your region's typical wired floor plus the documented penalty for your access medium, so the same 45 ms result passes in Latin America and fails in Western Europe. Five axes are then graded pass, marginal or fail against fixed published thresholds and weighted into a 0-100 score, latency heaviest. Any outright failure caps the letter grade, because one broken axis is not offset by four good ones.
Anchor point
Regional floors are typical wired best-case round trips to the nearest major edge; axis thresholds are fixed numbers printed on the page so a grade can be reproduced by hand.
Known error band
Thresholds are exact; the regional floors carry roughly +/-10 ms because routing differs by ISP. Nothing is measured in your browser.
Open the tool this model powers

Economy and payout model

Inputs
Money method, session length, business type, Shark Card tier.
How it is calculated
Each method carries a gross payout, a cooldown, and a setup cost, so net per hour is gross minus costs divided by real elapsed time including cooldowns. Business ROI compounds daily yield against the purchase price to produce a payback period, and card value is dollars of in-game currency per real dollar.
Anchor point
Payouts are modelled from GTA Online's published values, scaled by the inflation pattern between GTA IV and GTA V Online.
Known error band
Directional only until GTA 6 Online launches. Ranking between methods is the useful output, not absolute numbers.
Open the tool this model powers

Regional pricing model

Inputs
Country, edition, dated reference exchange rate, consumption-tax rate.
How it is calculated
One confirmed anchor — $79.99 for the US Standard edition — is either converted at a dated reference rate or replaced by the storefront's standing regional price where that market has one. Consumption tax is then extracted from the shelf price in tax-inclusive markets or added on top in the US and Canada, and the receipt total converts back to USD so 34 currencies sit on one comparable axis.
Anchor point
Rockstar's confirmed $79.99 Standard price. Deluxe and Ultimate rungs are modelled at +$20 and +$30, the spacing used for GTA 5 and Red Dead Redemption 2, and are labelled unconfirmed wherever they appear.
Known error band
Regionally-priced markets are indicative until storefronts list; converted markets are exact for the stated rate date and drift with the currency after it.
Open the tool this model powers

Storage and download model

Inputs
Install size estimate, platform, connection speed, overhead allowance.
How it is calculated
Install footprint is the base package plus a decompression working set plus a patch allowance. Download time is package size divided by sustained throughput, where sustained is derated from advertised line speed because CDN and protocol overhead never let you hit the sticker number.
Anchor point
Install size is anchored to GTA V Enhanced and RDR2 footprints adjusted for a larger map and higher-resolution texture sets.
Known error band
Download times are accurate within a few percent once your real sustained speed is entered; install size is +/-15 GB.
Open the tool this model powers

Pre-order channel scoring

Inputs
Platform, region, edition, and which of cancellability, preload timing or resale value you prioritise.
How it is calculated
Each buying channel carries five raw scores out of ten for cancellability, price-drop protection, preload readiness, bonus retention and resale value. Those are adjusted for context — statutory cancellation rights lift digital cancellability, a long platform preload window makes a late-arriving retailer code worse, larger editions amplify bonus-grant reliability in both directions — then multiplied by weights that shift toward the priorities you selected and renormalised to 100.
Anchor point
Raw axis scores come from long-standing platform and retailer policy, not from live listings. No price is read or displayed anywhere in this model.
Known error band
Ranking is the useful output, not the absolute score. Preload lead times are platform norms Rockstar has not confirmed for GTA 6.
Open the tool this model powers

Stated accuracy for every calculator

Different tools deserve different amounts of trust. This is the band we publish for each one, why it is that wide, and the event that will narrow it.

Published error band per ToolWools GTA 6 calculator
ToolBandWhyWhat narrows it
Download time+/-3%Pure arithmetic on a speed you supply; only overhead is estimated.Nothing pending — this is already launch-accurate.
Storage planner+/-15 GBInstall footprint is inferred from comparable Rockstar releases.Rockstar publishing the store listing size.
VRAM checker+/-1 GBAllocation behaviour differs between engine builds and drivers.First PC build with a public texture preset table.
Bottleneck+/-8 pointsPart indices are measured, but GTA 6's CPU-to-GPU load split is not published.Launch-day CPU scaling captures across 6, 8 and 16 core parts.
GPU tier list+/-1 band at edgesBlended scores cluster tightly around band boundaries.Real preset behaviour, especially ray-tracing cost.
FPS calculator+/-15%No published PC preset table exists, so preset cost is modelled from analogues.Launch benchmarks at each preset on the reference rig.
Price by countryExact at the rate dateConversions are arithmetic on a dated rate; regional price levels are inferred from standing storefront tiers.Storefront listings going live with pre-orders on 25 June 2026.
Internet speed advisor+/-10 ms on regional floorsAxis thresholds are fixed, but the latency floor to a regional edge varies by ISP and peering.Rockstar publishing GTA 6 online requirements and lobby regions.
Pre-order decision deskRanking, not a numberChannel terms are documented policy, but each storefront can change them and regional retail practice varies.Rockstar confirming the GTA 6 preload schedule and retailer bonus lists finalising.
Economy calculatorDirectional onlyGTA 6 Online payouts do not exist yet.Online launch and the first payout patch cycle.

What we do not claim

The limits matter as much as the outputs. Anyone who states these things with certainty before launch is guessing.

  • We do not claim to publish official GTA 6 PC system requirements. Rockstar has not released them. Every tier on this site is an estimate built from console targets and comparable RAGE-engine titles, and it is labelled as an estimate everywhere it appears.
  • We do not claim frame-rate precision. A number like 74 fps is the midpoint of a band, not a guarantee. Anyone publishing exact pre-launch frame rates without a band is guessing with more confidence than the data supports.
  • We do not model overclocks, undervolts, custom driver profiles, or mod loads. All results assume stock clocks, a clean install, and current stable drivers.
  • We do not model thermal throttling or laptop power profiles. A mobile GPU in a thin chassis will land under its desktop-indexed score, sometimes badly.
  • We do not track live retail pricing or regional currency moves in real time. Price and edition tools carry the date they were last checked.
  • We do not treat leaked material as evidence. Leaks may be discussed and labelled as unverified, but no leak has ever moved a number in a calculator.

How we verify every claim

Six rules that apply to every calculator on the site. If a change can't clear all six, it doesn't ship.

  • Every tier score and VRAM number is cross-checked against at least two independent benchmark sources (Digital Foundry, Hardware Unboxed, TechPowerUp, Gamers Nexus) before it ships.
  • FPS bands and VRAM base pools are re-validated against GTA V, RDR2 and Cyberpunk 2077 at 1080p, 1440p and 4K after each major driver release.
  • Ray-tracing and frame-generation overheads are re-measured against matched-settings CP2077 and Alan Wake 2 captures on every driver cycle.
  • Model changes are versioned in each tool's Changelog with the exact date, part list, and multiplier deltas — no silent edits.
  • User-reported outliers are re-run on the reference rig before we adjust a value; anecdotal reports alone never move a score.
  • Rockstar target-spec pages are re-scraped on every publish; sources are quoted verbatim with a dated permalink.

What triggers a re-measure

Models are not written once. Four events force a re-run, and the most recent date each one fired is listed here.

  1. Driver release

    Every NVIDIA and AMD stable game-ready driver triggers a re-run of the raster and ray-tracing suites on the reference rig. Any index that shifts more than 2% is republished.

  2. Rockstar spec or trailer update

    New official material is re-read the same day and quoted verbatim with a dated permalink before any model derived from it changes.

  3. Engine or comparable-title patch

    Major GTA V Enhanced, RDR2 or Cyberpunk 2077 patches that alter VRAM behaviour force a re-capture of the base pools those models lean on.

  4. GTA 6 launch day

    Every model is re-anchored against measured launch behaviour. Estimated bands are replaced with measured ones and the estimate labels come off as each tool is confirmed.

From your report to a shipped fix

Corrections follow one path, and reproduction is the gate. Here is exactly what happens after you send one.

  1. 1. You report a mismatch

    Send the exact tool, your inputs, the number the tool gave, and the number you measured, plus how you measured it. A reading without a capture method cannot be reproduced, so it cannot be actioned.

  2. 2. We reproduce it

    The Hardware Desk re-runs your case on the reference rig, or on the closest part in the database when the exact part is not on hand. Reproduction is the gate: anecdote alone never moves a value.

  3. 3. A reviewer signs off

    If the model is wrong rather than your setup being unusual, an independent reviewer checks the proposed delta against public datasets before it ships. Reviewers can block a change; they cannot author one.

  4. 4. It ships dated and credited

    The fix lands in that tool's Changelog with the date and the multiplier delta, and the reporter is credited unless they ask not to be. Typical turnaround is three to seven days, faster for arithmetic errors.

Baseline sources

Each tool page lists its own page-specific sources. These four are the baseline cross-check panel used across every calculator.

Methodology questions, answered

What people ask about how these estimates are built and how far to trust them.

  • How accurate are pre-launch GTA 6 FPS estimates?

    Our frame-rate model carries a band of roughly +/-15% before launch. That width is honest rather than pessimistic: GTA 6 has no published PC preset table, so preset cost is modelled from GTA V Enhanced, RDR2 and Cyberpunk 2077 behaviour on the same hardware. Relative ordering between two GPUs is far more reliable than the absolute number, which is why the tool always shows a band instead of a single figure.

  • What hardware do you test on?

    A fixed reference rig: Core i9-14900K, RTX 4090 24 GB, 64 GB DDR5-6400 CL32, a 2 TB Samsung 990 Pro, on Windows 11 24H2 with current stable drivers. That rig anchors at 100/100 and every other part is scaled against it from a database of 60+ GPUs and 30+ CPUs that is re-validated each driver cycle.

  • Do GPU vendors pay for tier placement?

    No. We take no payment, no free hardware, and no affiliate placement in exchange for a score or a tier. Every card is blended through the same raster, VRAM and ray-tracing formula, and the band edges are fixed numbers published on this page rather than a curve fitted to make the list look balanced.

  • Why do two of your tools give slightly different verdicts for one PC?

    Because they answer different questions. The readiness score is an ordinal rubric across your whole build including RAM and storage, while the bottleneck calculator only compares CPU against GPU at one resolution. A rig can be well balanced and still score modestly overall if its RAM or storage is the weak link, and vice versa.

  • How often do the numbers change?

    Whenever a trigger fires: a stable driver release, a Rockstar spec or trailer update, a major patch to a comparable RAGE-engine title, or a reproduced reader correction. Nothing changes silently — every model edit is dated in the affected tool's Changelog with the part list and the multiplier delta.

  • Can I reuse your data or cite these numbers?

    Yes, with attribution and a link to the specific tool page rather than to the homepage, because each tool carries its own sources and its own error band. If you need the underlying index values for a comparison of your own, ask and we will send the current table with its capture dates.

Meet the people behind these numbers

Bios for the Hardware Desk, our reviewers, and our editorial team live on the About page.

Read About ToolWools