Pick two graphics cards and the resolution you actually use. The first result summarizes their raster-score gap, rounding the lead to a whole percent and calling gaps below 2 percent evenly matched. Add the two prices you can buy them for and the tool calculates each card’s selected-resolution index per price, plus the price where card B would tie card A on that value measure.
No price is assumed. Prices always move around.
The performance data comes from the Tom’s Hardware GPU benchmarks hierarchy, published June 24, 2026. It is one 11-game geometric mean at native resolution, with separate 1080p Ultra, 1440p Ultra, and 4K Ultra columns. The tool does not merge old test suites, invent missing scores, or supply market prices.
A worked comparison without an invented price
At 1440p, the GeForce RTX 5070 Ti has an index of 69.8 and the Radeon RX 9070 XT has an index of 69.7. Using those stored scores, (69.8 - 69.7) / 69.7 is about 0.14 percent, so the performance headline calls them evenly matched.
Price can still separate them. If the RTX 5070 Ti’s price is P, the RX 9070 XT reaches the same index per price at P × 69.7 / 69.8, or about 0.9986 × P. Enter the current prices you found to replace that variable with a real buying comparison. The calculator never treats a manufacturer list price or an old observation as today’s price.
How to read the performance result
Start with your resolution. Scores are not FPS.
The percentage compares the two scores inside the selected resolution column. A 20 percent lead means the higher aggregate index is 20 percent above the lower aggregate index. It does not mean every game will run 20 percent faster, and it does not state either card’s absolute frame rate. Each resolution is normalized independently so the RTX 5090 equals 100 in that column. Compare card A with card B at 1440p, for example. Do not compare one card’s 1440p score with its 4K score to calculate an FPS loss; those two numbers have different normalization bases. The table can also show a reversal—one card leads at 1080p while the other leads at 4K—but the aggregate does not identify VRAM capacity, bandwidth, cache, drivers, or another factor as the cause.
A reversal is an observation, not a diagnosis. Cross-column arithmetic is not valid either.
How the price comparison works
Value remains a separate calculation.
Price A and price B are optional. When both are positive numbers, the tool calculates:
card A value = selected-resolution score A / price Acard B value = selected-resolution score B / price Bcard B break-even price = price A × score B / score A
Use one shared currency. Nothing here converts currencies.
Keep taxes, shipping, rebates, warranty, and condition on the same basis too. A pre-tax new-card listing and a delivered used-card price are not a clean comparison. The calculator shows the exact two ratios even when they are close; the performance headline’s “evenly matched” band does not hide the value arithmetic.
The break-even number answers one narrow question: given card A’s entered price and the two selected-resolution scores, what price would give card B the same score per currency unit? Below that price, B has the higher ratio. Above it, A has the higher ratio.
What the index leaves out
The benchmark stops there. That boundary is deliberate.
These are native-resolution raster results. Ray tracing, DLSS, FSR, XeSS, and frame generation are outside the score. So are cooler noise, board dimensions, power-supply requirements, and CPU limits. Check measurements for the exact game and settings you plan to run, then verify that the specific board fits your case and power budget.
I would use this index to size the raster gap, then make the purchase from game-level measurements and the constraints of the actual build.
Capacity is separate too. VRAM capacity is displayed beside the comparison but does not change any formula. The capacity rows were checked against the manufacturers’ published specifications: NVIDIA GeForce graphics cards, AMD Radeon desktop graphics, and Intel’s Arc B-series specifications. Capacity alone does not explain a score or establish how long a card will remain useful.
Matchups worth running
After choosing a card, check the processor pairing with the PC Bottleneck Calculator or price the rest of the system with the PC Builder.
Sources and calculation scope
- Performance scores: Tom’s Hardware GPU benchmarks hierarchy, June 24, 2026 raster table; 11-game geometric mean, native resolution, normalized separately at each resolution.
- VRAM capacity: primary specification pages from NVIDIA, AMD, and Intel.
- Prices: entered by the reader. The site does not fetch, store, or infer a current price in this tool.
- Value: selected-resolution normalized index divided by the entered price. It is a comparison ratio from this one raster suite, not literal frames per currency unit.