Overview Specs Price Reviews Alternatives

Nvidia

Nvidia GeForce RTX 3090 (Usada)

Name: Nvidia GeForce RTX 3090 (Usada)
Brand: Nvidia

Specs not independently verified

Starting at$600

Design score

8.5/10

How we score →

Find retailers

Design & Build8.5

StatusSpecs Pending

Verdict

The second-hand market's favourite: 24GB of VRAM around $600, and the basis of most cheap multi-GPU rigs.

Where it wins, where it doesn't

Pros

24GB VRAM at roughly $600 — unmatched cost per gigabyte
Pairs well for multi-GPU rigs (72GB across three cards)
Mature, well-understood CUDA support

Cons

No warranty and unknown usage history
High power draw that compounds across multiple cards
Two generations behind on compute and efficiency

Ideal forBudget-constrained local AI buildsResearchers assembling multi-GPU inference rigsAnyone whose constraint is VRAM rather than speed

Editorial note

One of the more useful ironies in consumer hardware: a card two generations old is the reigning choice for local AI, purely because VRAM aged better than compute. Twenty-four gigabytes for roughly $600 is a price per gigabyte nothing current touches, and because these cards pair well, three of them get you to 72GB for less than a single top-tier card. That is how a large share of home inference rigs are actually built. The risks are the ones you would expect from used silicon: no warranty, unknown history — a card that spent two years mining is not the card in the photograph — and power draw that adds up fast across three of them. Budget for a serious PSU and accept that one of them will eventually fail. For a fixed budget and a memory-bound workload, nothing else comes close.

In-Depth Review

One of the more useful ironies in consumer hardware: a card two generations old is the reigning choice for local AI, purely because VRAM aged better than compute. 24GB for roughly $600 is a price per gigabyte nothing current touches.

Why it defines cheap multi-GPU rigs

These cards pair well. Three of them get you to 72GB for less than a single top-tier card — which is how a large share of home inference rigs are actually built. Mature, well-understood CUDA support means the software just works.

The risks are used-silicon risks

No warranty, unknown history. A card that spent two years mining is not the card in the listing photo.
Power draw adds up fast across three cards — budget for a serious PSU.
Two generations behind on compute and efficiency, so tokens/second per card trails current parts.

Accept that one card will eventually fail, and buy a spare-friendly configuration.

Who should buy

Budget-constrained local AI builds, researchers assembling multi-GPU inference rigs, and anyone whose constraint is VRAM rather than speed. If you want a warranty and low power draw, buy new.

Frequently Asked Questions

Who is Nvidia GeForce RTX 3090 (Usada) for?↓

Nvidia GeForce RTX 3090 (Usada) is a fit for budget-constrained local AI builds, Researchers assembling multi-GPU inference rigs and Anyone whose constraint is VRAM rather than speed.

What are the drawbacks of Nvidia GeForce RTX 3090 (Usada)?↓

The trade-offs we record are: No warranty and unknown usage history, High power draw that compounds across multiple cards and Two generations behind on compute and efficiency.

What does Nvidia GeForce RTX 3090 (Usada) do well?↓

24GB VRAM at roughly $600 — unmatched cost per gigabyte, Pairs well for multi-GPU rigs (72GB across three cards) and Mature, well-understood CUDA support.

How much does Nvidia GeForce RTX 3090 (Usada) cost?↓

$600, as recorded in this index. Prices move; check the retailer for the current figure.

What is the Fathom Layer design score for Nvidia GeForce RTX 3090 (Usada)?↓

Nvidia GeForce RTX 3090 (Usada) scores 8.5 out of 10. The score is assigned by a human against published criteria and placement is never sold — see the methodology page for how it is built.

Alternatives to consider

See all alternatives →