Twice the price for a large but not proportional performance gain, plus double the memory and a 215 W increase in board power. The question is whether the top of the range is a product or a statement.
The fastest consumer graphics card on the market and the only one that treats 4K at high refresh rates as routine. It also asks for a 575 W power budget, a large case and flagship money, which is why it only makes sense at the very top of a build.
A capable 4K card with the full RTX 50 feature set at roughly half the price of the flagship. The 16 GB memory pool is the one specification that invites debate at this price point.
vs
RTX 5090 comes out ahead of RTX 5080.
RTX 5090 is about 35% faster than RTX 5080 at 1440p.
RTX 5080 costs 234% less than RTX 5090.
RTX 5090 has 32 GB of VRAM against 16 GB.
The score blends performance at 1440p (50%), performance per dollar (30%) and memory headroom (20%), each against every card CompareYourGPU tracks. Fewer than three points apart counts as level. How verdicts are calculated.
Biggest differences
Performance, memory, power and value rows, ranked by how far apart RTX 5090 and RTX 5080 are. The full sheet below has every row.
The largest differences between RTX 5090 and RTX 5080
Difference
RTX 5090
RTX 5080
Edge
Performance per $100
1.5
Better: 3.8
RTX 5080147% higher
VRAM
Better: 32 GB
16 GB
RTX 5090100% higher
Memory bandwidth
Better: 1,792 GB/s
960 GB/s
RTX 509087% higher
Board power
575 W
Better: 360 W
RTX 508060% lower
Performance index, 4K
Better: 100
69
RTX 509045% higher
Ray tracing index, 1440p
Better: 100
72
RTX 509039% higher
Full specification comparison
Open any area to see every row. Rows where a card wins are marked; rows where the two match, or where a figure is not published, say so.
Performance indexCompareYourGPU's relative speed rating, where 100 is the fastest card in the database at that resolution. A card at 50 is roughly half as fast.It is an estimate aggregated from public review coverage, not a CompareYourGPU measurement, and it is meant for ranking cards against each other rather than predicting a specific frame rate.CompareYourGPU estimate, 100 = fastest card here.
✓Best in this row: 100
82
Performance indexCompareYourGPU's relative speed rating, where 100 is the fastest card in the database at that resolution. A card at 50 is roughly half as fast.It is an estimate aggregated from public review coverage, not a CompareYourGPU measurement, and it is meant for ranking cards against each other rather than predicting a specific frame rate.CompareYourGPU estimate, 100 = fastest card here.
✓Best in this row: 100
74
Performance indexCompareYourGPU's relative speed rating, where 100 is the fastest card in the database at that resolution. A card at 50 is roughly half as fast.It is an estimate aggregated from public review coverage, not a CompareYourGPU measurement, and it is meant for ranking cards against each other rather than predicting a specific frame rate.CompareYourGPU estimate, 100 = fastest card here.
✓Best in this row: 100
69
Ray tracingA rendering technique that simulates how individual light rays travel and bounce, instead of approximating lighting with pre-computed tricks.It produces much more convincing reflections, shadows and lighting, and it is expensive. Expect to lose a substantial share of your frame rate when you enable it.CompareYourGPU estimate, 100 = fastest card here.
✓Best in this row: 100
72
Measured suite average at 1440pThird-party geometric mean across their game suite, run with vendor upscaling at Quality. A different test from the estimated row below, not a check on it.
✓Best in this row: 142.4 FPS (ComputerBase)
105.4 FPS (ComputerBase)
Estimated FPS at 1440pModelled estimate for a typical modern AAA game at High settings with no upscaling. Not a measurement.
VRAMThe graphics card's own memory. Textures, geometry and frame data live here while a game is running.When a game needs more VRAM than the card has, it either lowers texture quality or stutters. More VRAM mostly buys you higher texture settings and a longer useful life.
✓Best in this row: 32 GB
16 GB
GDDR6 / GDDR6X / GDDR7Successive generations of graphics memory. Each generation moves data faster per lane than the one before.GDDR7 lets a card with a narrow memory bus reach bandwidth figures that used to need a much wider one.
GDDR7
GDDR7
Memory busHow many parallel lanes connect the graphics processor to its memory, measured in bits. A 256-bit bus has twice as many lanes as a 128-bit bus.A narrow bus can bottleneck a card at higher resolutions even when it has plenty of VRAM, because the memory cannot be read fast enough.
✓Best in this row: 512-bit
256-bit
Memory bandwidthHow quickly the graphics card can move data between the graphics processor and its memory, measured in gigabytes per second.Bandwidth is bus width multiplied by memory speed. It usually matters more than raw VRAM capacity for frame rate, especially at 4K.
CUDA coresNVIDIA's name for the small processors that do the bulk of a graphics card's maths.Core counts are only comparable within one manufacturer and one generation. An NVIDIA card with fewer cores than an AMD card can easily be faster.Only comparable within one manufacturer.
✓Best in this row: 21,760
10,752
Compute unitsA block of stream processors on AMD cards, or an Xe core on Intel cards. Each block contains a fixed number of shader units plus its own ray tracing and AI hardware.Within a single AMD or Intel generation, more compute units means a faster card. Across generations it does not.
–Data unavailable
–Data unavailable
RT coresDedicated hardware for calculating how light rays bounce around a scene. AMD calls them ray accelerators, Intel calls them ray tracing units.If you want to play with ray tracing switched on, the quality and count of this hardware matters more than the raw shader count.
✓Best in this row: 170
84
Tensor coresHardware built to run AI models quickly. NVIDIA calls them Tensor cores, AMD calls them AI accelerators, Intel calls them XMX engines.This is the hardware that runs modern upscalers such as DLSS and FSR 4, and it is increasingly what separates good upscaling from mediocre upscaling.
✓Best in this row: 680
336
Base and boost clockThe guaranteed minimum operating speed and the peak speed the graphics processor will reach when temperature and power allow. AMD also publishes a 'game clock', the speed it expects to sustain while playing.Clock speeds are only comparable within a single architecture. A card clocked lower can easily be faster than one clocked higher.
2,017 MHz
2,295 MHz
Base and boost clockThe guaranteed minimum operating speed and the peak speed the graphics processor will reach when temperature and power allow. AMD also publishes a 'game clock', the speed it expects to sustain while playing.Clock speeds are only comparable within a single architecture. A card clocked lower can easily be faster than one clocked higher.
TDP / TBPThe board power a graphics card is designed to draw, in watts. NVIDIA calls it TGP, AMD calls it TBP; in practice they describe the same thing.It determines the power supply you need, how much heat ends up in your case, and how loud the fans have to work.
575 W
✓Best in this row: 360 W
Recommended PSUThe total power supply wattage the graphics card manufacturer suggests for a complete system built around that card.It accounts for the processor and the rest of the system, not just the graphics card. Undersizing it causes shutdowns under load.Vendor figure where published, otherwise board power plus 300 W of headroom.
Ray tracingA rendering technique that simulates how individual light rays travel and bounce, instead of approximating lighting with pre-computed tricks.It produces much more convincing reflections, shadows and lighting, and it is expensive. Expect to lose a substantial share of your frame rate when you enable it.
4th-generation RT cores
4th-generation RT cores
UpscalingRendering the game at a lower resolution and then reconstructing a higher-resolution image from it. DLSS, FSR and XeSS are the three main implementations.It is the most effective way to gain frame rate for a small image-quality cost, and it is how most people actually reach 4K on mid-range hardware.
DLSS 4 Super Resolution, DLSS Ray Reconstruction, FSR 3.1, XeSS
DLSS 4 Super Resolution, DLSS Ray Reconstruction, FSR 3.1, XeSS
Frame generationThe graphics card creates additional frames between the ones the game actually renders, raising the displayed frame rate.It makes motion look smoother but does not reduce input latency the way real frames do, and it needs a reasonable base frame rate to feel right. Treat it as a smoothness feature, not a performance fix.
DLSS 4 Multi Frame Generation
DLSS 4 Multi Frame Generation
AV1 encodingA modern video codec. Hardware AV1 encoding lets the graphics card compress video efficiently without using the processor.It matters if you stream or record. AV1 delivers noticeably better quality than older codecs at the same bitrate.
Yes
Yes
Video encoder
9th-generation NVENC with AV1 and HEVC encode
9th-generation NVENC with AV1 and HEVC encode
Video decoder
6th-generation NVDEC with AV1, HEVC and H.264 decode
6th-generation NVDEC with AV1, HEVC and H.264 decode
Slot sizeHow many expansion slots on the back of the case the card physically occupies. A 3-slot card blocks two slots below the one it plugs into.It decides whether the card fits your case, along with its length. Larger coolers are usually quieter but need more room.Essentially tied
All six axes are normalised to a 0-100 scale so they can share one chart. Performance and ray tracing are CompareYourGPU’s estimated indices; value, VRAM, efficiency and features are each card’s standing relative to every other card tracked here.
Normalised comparison of RTX 5090, RTX 5080 across six criteria, judged at 1440p.
RTX 5090
RTX 5080
Normalised comparison of RTX 5090, RTX 5080 across six criteria, judged at 1440p.
Criterion
RTX 5090
RTX 5080
Performance
100
74
Ray tracing
100
72
Value
0
30
VRAM
100
80
Efficiency
40
65
Features
100
100
Performance, memory and value
Estimated performance index at 1440p. 100 is the fastest card CompareYourGPU tracks.
A card with a verified current price is shown at that price, a dated snapshot; a card without one is shown at launch MSRP. Street prices move, so check the retailer.
Which one should you buy?
The short version: the RTX 5090 is faster, but the RTX 5080 offers better value at the prices on record. Which of those matters more is the decision.
Based on the specifications, estimated performance indices and prices on record for these cards. How verdicts are calculated.
Highest estimated performance index at 1440p of the cards compared here, at 100. It is the right answer if frame rate is the only thing you are optimising for.
Delivers the most performance per dollar of the cards compared, at $1,973. Choose this if the difference in price matters more to you than the difference in frame rate.
Strongest ray-traced performance here, with an estimated ray tracing index of 100 at 1440p. It also carries DLSS 4 Super Resolution, which is how ray tracing becomes practical.
Best balance of speed, price and memory of the cards compared: 50% performance at 1440p, 30% performance per dollar, 20% memory headroom. Not necessarily the fastest.
RTX 5090 vs RTX 5080: common questions
Is the RTX 5090 twice as fast as the RTX 5080?+
No. It costs twice as much and delivers a substantial but smaller performance advantage. The bigger practical differences are the 32 GB memory pool and the ability to run path-traced games at high resolutions without heavy upscaling.
Can I run an RTX 5090 on my current power supply?+
Only if it is a high-quality 1,000 W-class unit with the right connector. At 575 W the RTX 5090 is the most demanding consumer card here by a wide margin. The RTX 5080 at 360 W is far easier to accommodate.
Does 32 GB of VRAM matter for gaming?+
Not for games alone. No current title needs anywhere near 32 GB. It matters for creator and AI workloads, and it is insurance for the future rather than a benefit you will see today.
Where to go next
Not sure either is right?
Answer seven questions about your monitor, games and budget and let the recommendation engine narrow it down.