13 specifications

NVIDIA GeForce RTX 4080 vs NVIDIA GeForce RTX 5090

On paper, the NVIDIA GeForce RTX 5090 carries more memory (32 GB against 16 GB), the NVIDIA GeForce RTX 5090 has more shading units, and the NVIDIA GeForce RTX 5090 has a wider memory bus. Shading-unit counts are not directly comparable across different architectures, so read that row as context rather than a verdict. Specifications describe what each card is designed to do; how much of that either one holds under continuous load is a separate question.

10 specifications · 5 with a higher side

Full specifications

NVIDIA GeForce RTX 4080

NVIDIA · GeForce RTX 40 · Desktop

Full specifications

NVIDIA GeForce RTX 5090

NVIDIA · GeForce RTX 50 · Desktop

Compute

Architecture

Ada Lovelace

Blackwell (GB202)

Shading units

9,728

21,760

Base clock

2210 MHz

2010 MHz

Boost clock

2510 MHz

2410 MHz

Memory

16 GB

32 GB

Memory type

GDDR6X

GDDR7

Memory bus

256-bit

512-bit

Power and platform

TDP

320 W

575 W

Process

Not verified

5 nm

Launched

2022-11-16

2025-01-30

A check marks the higher figure only on rows where higher is unambiguously better. Rows without one are not ties; they are specifications where “better” has no meaning.

Reading the table

What the difference means.

3 years apart

The RTX 5090 (Blackwell (GB202), 2025) is 3 years newer than the RTX 4080 (Ada Lovelace, 2022). A newer architecture brings more than speed: newer video encoders and decoders, newer upscaling and frame generation, and a longer run of driver updates ahead of it.

Video memory

The RTX 5090 has 32 GB of video memory against 16 GB. Memory does not make a card faster until it runs out, and then the drop is sharp: textures spill into system memory and frame times stutter. High resolutions, high texture settings, heavy mods and running AI models locally are what fill it.

Memory bus

The RTX 5090 moves data over a 512-bit bus of GDDR7; the RTX 4080 uses 256-bit GDDR6X. The bus width and the memory type together set how fast the chip can reach its own memory, which shows at high resolutions where every frame moves more data. A narrow bus with fast memory can match a wide one with slow memory, so neither figure settles it alone.

Shading units

The counts differ, 21,760 against 9,728, but these are different designs, and a shading unit on one is not the same amount of work as on the other. The row cannot be used to rank this pair.

Board power

The RTX 5090 draws up to 575 W against 320 W. That is heat the case has to exhaust and current the power supply has to deliver: allow a supply of about 1100 W for the RTX 5090 with a typical processor, and check the card's power connectors against the supply's. A card that cannot shed its heat does not fail. It lowers its clocks after a few minutes, which is the frame rate that falls the longer you play.

Which to pick, on paper

High resolutions, heavy textures, local AI: RTX 5090; 32 GB against 16 GB. Memory is the wall these workloads hit first.
A small case or a modest power supply: RTX 4080; 320 W against 575 W: less heat, fewer power connectors, quieter fans.

What no table says

How much of its speed a graphics card keeps after the first minute depends on the cooling and the power limit of the machine it is in. If you own one of these, measure it: the test is free and runs in this browser.

Free GPU stress test

Questions

Is the RTX 4080 better than the RTX 5090?: By need, on the specifications: high resolutions, heavy textures, local ai: the RTX 5090; a small case or a modest power supply: the RTX 4080. They are different designs, so the raw counts cannot rank them. Frame rates depend on the game and its settings, and how much speed either card keeps after ten minutes depends on its cooling, which is what a sustained test measures.
Is 16 GB of video memory enough?: For 1080p at sensible settings, yes in most games. At 1440p and above, with high texture settings, or for running AI models, 32 GB gives real headroom. When memory runs out the symptom is stutter, not a lower average.
What power supply does the RTX 5090 need?: Allow about 1100 W with a typical processor, and make sure the supply has the power connectors the card uses. The RTX 4080 at 320 W is easier on both.

More comparisons

Every figure on this table is a design target. None of it is a measurement.

Nothing above describes how a graphics card behaves twenty minutes into a sustained workload, which is where two chips with nearly identical specifications routinely diverge. A boost clock is a ceiling the vendor permits, not a rate the chip sustains, and whether it holds depends on the cooling, the case, the power profile and the room. None of those appear on a specification sheet.

We deliberately do not reduce these figures to a single score. A composite invented from specifications would look authoritative and measure nothing, and choosing the weights would be choosing the winner.

Run the test on your machine How we measure →