GPU Benchmark Chart 2026: Real PassMark Scores, Honestly Labelled
Updated 2026-09-22
The short version
If you want the fastest consumer card money can buy and you are not troubled by what money is, it is the RTX 5090. If you want the sensible one, look at where the curve flattens, which happens a long way below the top. The rest of this page shows you exactly where.
Every number below comes from PassMark’s public G3D Mark database. We pulled them on 22 September 2026 and we are telling you that because a benchmark chart without a date is a horoscope.
The chart
PassMark G3D Mark, high-end consumer cards, retrieved 22 September 2026 from videocardbenchmark.net. Higher is faster.
| Graphics card | G3D Mark | Percent of RTX 5090 |
|---|---|---|
| GeForce RTX 5090 D | 42,042 | 108% |
| GeForce RTX 5090 | 39,019 | 100% |
| GeForce RTX 4090 | 38,036 | 97% |
| GeForce RTX 5080 | 35,633 | 91% |
| GeForce RTX 4080 | 34,435 | 88% |
| GeForce RTX 4080 SUPER | 34,200 | 88% |
| GeForce RTX 5070 Ti | 32,329 | 83% |
| GeForce RTX 4070 Ti SUPER | 31,841 | 82% |
| Radeon RX 7900 XTX | 31,457 | 81% |
A note on the oddity at the top: the RTX 5090 D is a China-market variant, and it scores above the standard 5090 in this database. Benchmark databases are full of results like this, which is a useful reminder that these are aggregated user submissions rather than one lab testing one card under controlled conditions.
The part the chart will not tell you
Here is where most benchmark pages stop and start recommending things. We are going to be more annoying than that, because the number above is genuinely useful and genuinely limited, and pretending otherwise is how people overspend.
It is a synthetic score, not a frame rate. G3D Mark measures a fixed set of graphics workloads. Your game is not that workload. A card that leads here can trail in a specific engine, and ray tracing and upscaling both scramble the ordering further.
It is an aggregate of public submissions. Thousands of people running different drivers, different processors, different thermal situations. That makes it excellent for ranking cards roughly and poor for predicting your exact result.
It says nothing about value. There is no price in that table, deliberately. Prices move weekly and vary by region, and a benchmark chart that quotes a stale price is worse than one that quotes none.
It cannot see your monitor. Which brings us to the actually important bit.
Look at the gaps, not the ranking
The ranking is the least interesting thing in that table. The gaps are where the money is.
Look at the top four. The 4090 sits at 97 percent of a 5090. The 5080 reaches 91 percent. By the time you are at the 5070 Ti you still have 83 percent of the flagship’s score, and the price difference between those two cards is not 17 percent of anything.
That is the flattening curve, and it is the single most useful thing on this page. Performance climbs steadily. Price climbs much faster than performance near the top. The last few percent is charged at a spectacular premium because some people will pay it.
Unless you have a specific, identified reason to need the top card, the value is in the middle of this table, not the head of it.
Start with your monitor, seriously
The most common expensive mistake in this hobby is buying a card your display cannot show you.
A 60Hz screen cannot display more than 60 frames per second. Not “it looks similar”, it physically cannot. Buying a flagship to feed a 60Hz 1080p panel is buying a sports car for a car park.
Work out two things before you look at a single price:
- Resolution. 1080p, 1440p, 4K. Each step roughly doubles the pixels and the work.
- Refresh rate. This is the ceiling on frames you will actually see.
“1440p at 144Hz” is something you can shop against. “The best GPU” is not.
The things that quietly ruin the plan
- Your processor. A fast card fed by a slow CPU will not deliver what the chart promises, and this bites hardest at lower resolutions where the card finishes early and waits. Our CPU buying framework covers the other half of this.
- Your power supply. New cards have appetites. Check before you buy, not after, using our PSU sizing guide.
- Your case. High-end cards are physically enormous. Measure. People do not measure.
- VRAM. It does not make a card faster, it sets a ceiling. Exceed it and performance does not degrade politely, it falls over. This matters more at 4K and with texture-heavy games.
How to finish the job properly
Use the chart above to build a shortlist of two or three cards. Then, before you buy:
- Find reviews that test your games, at your resolution, and that publish their test setup. Tom’s Hardware, TechPowerUp, and Gamers Nexus all do.
- Cross-check at least two sources. If they disagree by more than roughly 10 percent, the difference is test conditions, and knowing which matters.
- Check the review date. Driver maturity genuinely moves performance on newer architectures.
A synthetic score narrows the field. Real reviews of your actual games decide it. Anyone selling you certainty from a single number is selling you something.
Source
PassMark G3D Mark scores, retrieved 22 September 2026 from videocardbenchmark.net high-end GPU chart. PassMark aggregates results submitted by users running its PerformanceTest software. Figures move as more results arrive, so treat the date above as part of the data.
☕ Coffee Corner
Compiling, deploying, or waiting on a render? Here's what to brew while you wait.
🏠 Smart Home Picks
Same hobbyist care applied to your network and your front door.
