GPU Value Score Explained: What 100 Points Means, and Four Things It Cannot See
A Value Score of 100 means a graphics card delivers roughly 30 frames per second at 1440p ultra for every $100 it costs. That is the whole claim. It is a price-to-performance ratio with an age adjustment, scaled to a 0–100 range, and it is deliberately blind to nearly everything else about the card.
Which is why the numbers surprise people. Across the 77 cards we have both a benchmark and a price for, the median score is 39. Only 6 score 80 or higher. And NVIDIA GeForce RTX 5090 — the fastest card on this site at 171 fps — scores 28, placing it 51st of 77. None of that is a bug. It is what a ratio does.
Short version.
- Score = 1440p ultra fps ÷ price, multiplied by an age factor, scaled so 30 fps per $100 reaches 100.
- Median is 39. Above 70 is genuinely strong; under 25 almost always means an old card at a price nobody should pay.
- It says nothing about how fast a card is. RTX 5090 and RX 6500 XT score within 1 point of each other and one is 9.5× faster.
- Use it to build a shortlist. Decide between the survivors on VRAM, resolution and the price you can actually pay today.
What the score measures
The formula takes one performance input and one price input. Performance is the geometric mean of 15 games at 1440p ultra with no upscaling, from Tom's Hardware GPU hierarchy. Price is the best figure we hold for the card — a retail price where we have observed one, otherwise the manufacturer's list price. Divide the first by the second, multiply by an age factor between 0.6 and 1.0, then scale so that 30 fps per $100 lands on 100 points and clip anything above.
That makes the score directly readable rather than abstract. A score of 100 is about 30 frames per $100. A score of 50 is about 15. A score of 39 — the median card — is roughly 11.6 frames per $100. Our methodology page holds the code-level detail; this article is about what the number does and does not license you to conclude.
What a score of 100 actually means
It means "at or above the practical ceiling", not "perfect". The 30 fps-per-$100 scale point was chosen because it sits just above the best ratio in our data, and exactly 1 card currently exceeds it: NVIDIA GeForce RTX 5060 Ti 16 GB at 33.5 fps per $100, which clips to 100.
Worth stating plainly, because it is a real limitation: because the scale clips, two cards above the ceiling would both display 100 and you could not tell them apart. If that happens the ceiling gets raised, not the cards re-ranked. The raw fps-per-$100 figure is published on every card page for exactly this reason — it never clips.
The distribution is worth looking at before you judge any individual score. 13 of 77 cards sit within five points of the median, 10 score under 20, and the 6 cards above 80 are genuinely unusual. The scale is not graded on a curve. A score of 60 is not mediocre — it is better than about three quarters of the market.
The score is not a performance ranking
This is the single most common misreading, and the reason we get email about the NVIDIA GeForce RTX 5090. A ratio has no scale in it. Two cards land on the same score when their frames-per-dollar match, no matter how many frames either of them produces.
The chart has no trend line because there is no trend. At almost any score you can find a fast expensive card and a slow cheap one sitting on top of each other. The extreme case here is NVIDIA GeForce RTX 5090 at 171 fps and AMD Radeon RX 6500 XT at 18 fps: a 9.5× performance difference, and their Value Scores are 28 and 27.
So if your requirement is a frame-rate target — 120 fps at 1440p, 60 fps at 4K — the Value Score cannot answer it and will actively mislead you if you let it. Filter by measured frame rate first, then use the score to choose among the cards that clear your bar. That ordering matters more than anything else in this article.
Same score, different purchase
The blindness is not confined to the extremes of the range. NVIDIA GeForce RTX 5050 scores 76 and AMD Radeon RX 9070 scores 74 — 2 points apart, both released in 2025 or later so the age factor is identical for both. Everything separating them is price and performance.
And what separates them is a lot: 122 fps against 57, 2.1× the frames, with 16 GB of VRAM against 8 GB, for $549 MSRP against $249 MSRP. Both are efficient purchases — that is precisely what the matching scores are telling you. Only one of them plays a 2026 title at 1440p with textures turned up. The score answers "is this priced sensibly for what it does", never "is what it does enough for me".
The age multiplier, and how much it moves a card
Cards released in 2024 or later count at full weight. 2022–2023 is multiplied by 0.9, 2020–2021 by 0.75, and 2019 or earlier by 0.6. This is the most contestable part of the formula, so here is the argument for it and then the size of its effect, so you can decide whether you accept it.
Why an older card at the same ratio is not the same buy
The frame rate is the same; the ownership is not. An older card typically means no hardware AV1 encoding, weaker or entirely absent upscaling support, ray tracing that is nominally present and practically unusable, driver attention that has moved to newer architectures, and fewer years of remaining useful life before the next engine generation leaves it behind. None of that appears in an average frame rate. The multiplier is our attempt to price it in with one number rather than five caveats.
What it costs in practice
The largest single deduction on the site is AMD Radeon RX 5500 XT: 56 on price and performance alone, 33 as published, a 22-point cut. In rank terms the biggest mover is NVIDIA GeForce RTX 3060, which sits 12th of 77 on raw frames per dollar and 24th once its release date is applied — 12 places.
Aggregated by generation, the median score by age bucket runs 72 for 2024 or newer (18 cards), 46 for 2022–2023 (23 cards), 38 for 2020–2021 (14 cards), 23 for 2019 or older (22 cards). Read that carefully, though: most of that gap is not the multiplier. Older cards also genuinely return fewer frames per dollar at the prices they still sell for, and the two effects are tangled together. The chart above isolates the multiplier; the bucket medians do not.
If you reject the penalty outright, the fix is already on the page. Every card carries its unmodified fps-per-$100 figure next to the Value Score. Sort on that column and you are reading the same benchmark and the same price with none of our opinion in between — which is the point of publishing both.
Four things the score cannot see
1. VRAM, except indirectly
The performance input is an average frame rate, and an average is precisely the statistic that VRAM exhaustion damages last. A card short on memory posts a respectable average and then stutters — texture pop-in, frame-time spikes, a hitch every time you turn a corner — in exactly the games you bought it for. Two cards with the same score and different frame buffers are not the same purchase. Our VRAM guide exists because the score genuinely cannot carry this, and because the capacity you need is set by your resolution and how long you plan to keep the card, neither of which is in the formula.
2. The price it is dividing by
Of the 77 scored cards, we hold an observed retail price for 1. Every other score divides by a manufacturer's list price, which is a marketing decision rather than a measurement, and street prices diverge from it in both directions and by a great deal. This is the largest source of error in every number on this site and we would rather say so than bury it. The full version of the problem, including a card that ranks tenth at list price and first at the price it actually sells for.
3. Anything that is not 1440p raster gaming
The input is 1440p ultra without upscaling, so that is the only axis the ranking is valid on. The effect is easy to quantify: the top five cards by Value Score rank 27th to 47th of 51 on absolute 4K frame rate. In productivity workloads the ordering nearly inverts — NVIDIA GeForce RTX 5090 is first of 77 on our productivity index and 51st of 77 on Value Score. If you play at 1080p, target 4K, care mainly about ray tracing, run local models or edit video, our use-case rankings reorder the same data on the axis you actually care about: 4K gaming, ray tracing, local LLM.
Ray tracing deserves a specific note, because the score is accidentally decent at it and that is not the same as being valid for it. Among cards scoring 70 or above, ray-traced frame-rate retention ranges from 42% on Intel Arc B570 to 61% on Intel Arc B580. Two cards with near-identical Value Scores can lose very different amounts of performance the moment you enable it, and nothing in the score warns you which is which.
4. Everything qualitative about owning the card
Fan noise under sustained load, coil whine, thermals in a small case, physical length against your chassis clearance, the power connector it wants, driver stability on your platform, how a given board partner handles an RMA. None of it is in the formula and all of it decides whether you are still happy two years in. Read owner reports alongside the numbers; this is the one area where forum threads beat benchmarks.
How to use the score in five steps
- Set your frame-rate target first. Resolution, refresh rate, the games you actually play. Filter the table by measured fps, not by score.
- Set your VRAM floor next. Driven by resolution and how many years you want the card to last, not by price.
- Now sort the survivors by Value Score. This is the step the score is built for — separating a fair price from a bad one among cards that already meet your requirements.
- Check the price is real. If the card page shows a list price rather than an observed one, look it up before you trust the score attached to it.
- Decide on what the score cannot see. Encoder support, ray tracing if you use it, noise, size, power. Two cards within a few points are effectively tied and these break the tie.
Used that way the score does one job well: it takes 77 cards down to the handful worth researching and flags the ones that are simply overpriced for what they deliver. Used as a leaderboard it will point you at a cheap card that cannot run your games, which is not a saving.
Common questions
Why isn't the fastest card near the top?
Because it is not the best value, and the score only measures value. NVIDIA GeForce RTX 5090 produces 171 fps at 1440p — more than anything else here — for $1999 MSRP, which works out to 8.5 frames per $100 and a score of 28. Flagships are priced for people who want the top of the stack regardless of the ratio, and the score is not the tool for that decision. It is working correctly when it puts them mid-table.
Is a score of 50 bad?
No. The median is 39 and only 19 of 77 cards clear 60. A 50 means roughly 15 frames per $100, which is an ordinary, defensible price. Scores in the teens are the ones to treat as a warning, and they almost always describe an old card still being sold at something close to its original price.
Why cap the score at 100?
Readability. An uncapped ratio scale has no natural top and a single mispriced entry can stretch it until every other card looks bad. The cost of the cap is real — 1 card above the ceiling would be indistinguishable — and the raw fps-per-$100 figure is published alongside the score to cover it.
Does the score change over time?
Yes, and it should. The benchmark figures are stable but prices are not, so the same card can gain or lose ten points in a quarter without anything about the hardware changing. Every card page carries the date its price was last updated. A Value Score is a statement about a price on a date, not a permanent property of the silicon.
Should I trust the age penalty?
Trust it as an opinion, which is what it is — it is the only judgement call in the formula. If you specifically want a used older card and do not need AV1 encoding, modern upscaling or usable ray tracing, the penalty is charging you for things you do not want. Read the raw fps-per-$100 column instead; both numbers are on every page precisely so you can disagree with us and still use the data.
Why 1440p rather than 1080p or 4K?
It is the resolution where the widest span of cards produces meaningful results. At 1080p, everything from mid-range upward runs into CPU limits and the ranking compresses. At 4K, most of the market falls below playable and the ranking becomes a list of five expensive cards. 1440p separates the field. It is still a choice, and if you game at a different resolution the 1080p and 4K rankings are the ones to read.
Performance figures: Tom's Hardware GPU hierarchy — geometric mean of 15 games at 1440p ultra, no upscaling. We do not test hardware ourselves; our methodology documents where every number on this site comes from.