Skip to main content
NVIDIA GeForce RTX 3090
NVIDIA · GPU · Ampere

NVIDIA GeForce RTX 3090 — Benchmarks & Specs

24 GB VRAM350W TDP$1,499 MSRP2020
Buy NVIDIA GeForce RTX 3090 on Amazon →

*Price sourced from Amazon.com. Price and availability subject to change.

Bottom line: how fast is the NVIDIA GeForce RTX 3090?

At 1440p (Ultra), the NVIDIA GeForce RTX 3090 averages 156 fps in Call of Duty: Black Ops Cold War with a 130 fps 1% low, per PCBench. For local LLM inference it generates 135.7 tokens/sec running qwen3:35b-a3b at Q4_K_XL under llama.cpp, per GitHub (thc1006). In PassMark G3D Mark it scores 26,548 pts, per PassMark. Its 24 GB of VRAM is the binding constraint for local inference: that capacity fits 32B-parameter models at Q4 without offloading to system RAM.

Every figure above is a row in the tables below, and each row links out to the review or public benchmark database the number was taken from. SpecPicks aggregates published measurements; it does not report first-party benchmark runs.

The NVIDIA GeForce RTX 3090 is a graphics card from the Ampere family released in 2020 from NVIDIA. Key on-paper specs include 24 GB of GDDR6X VRAM, 350W TDP. It launched with a $1,499 MSRP, though street prices typically diverge meaningfully from launch pricing — see the linked product cards below for current Amazon listings. Data on this page draws on 8+ Amazon listings, 9 synthetic benchmark results, 12 community AI inference reports, 14 measured game frame-rate results, aggregated from public benchmark databases (TechPowerUp, PassMark, Geekbench, Cinebench) and the LocalLLaMA community. Read this page when shopping the NVIDIA GeForce RTX 3090, comparing it against other graphics cards in your build, or sizing it for a specific workload (gaming at 1080p/1440p/4K, productivity benchmarks, or local LLM inference).

Gaming Performance (measured FPS)

Average and 1% low frame rates by game, resolution, and quality preset. Bars are scaled against the fastest result on this page.

Measured gaming frame rates for the NVIDIA GeForce RTX 3090 by game, resolution, and quality preset. “1% low” is the frame-time floor that determines perceived smoothness. Each row links to its original review or benchmark database.
Game Resolution Settings Relative Avg FPS 1% low Source
Call of Duty: Black Ops Cold War 1080p Ultra 211 fps 184 fps PCBench 2023-01-01
Call of Duty: Black Ops Cold War 1440p Ultra 156 fps 130 fps PCBench 2021-01-01
Watch Dogs: Legion 1080p Ultra 115 fps 95 fps PCBench 2023-01-01
Watch Dogs: Legion 1080p RT Ultra RT on DLSS Quality 97 fps 76 fps PCBench 2023-01-01
Assassin's Creed Valhalla 1080p Ultra 96 fps 85 fps PCBench 2021-01-01
Call of Duty: Black Ops Cold War 4K Ultra 94 fps 73 fps PCBench 2023-01-01
Call of Duty: Black Ops Cold War 4K RT Ultra RT on DLSS Quality 94 fps 71 fps PCBench 2023-01-01
Watch Dogs: Legion 1440p Ultra 92 fps 76 fps PCBench 2023-01-01
Assassin's Creed Valhalla 1440p Ultra 83 fps 74 fps PCBench 2022-01-01
Watch Dogs: Legion 1440p RT Ultra RT on DLSS Quality 78 fps 68 fps PCBench 2023-01-01
Watch Dogs: Legion 1440p Ultra RT on DLSS Quality 78 fps 68 fps PCBench 2021-01-01
Watch Dogs: Legion 1080p RT Ultra RT on 76 fps 64 fps PCBench 2023-01-01
Call of Duty: Black Ops Cold War 4K RT Ultra RT on 64 fps 41 fps PCBench 2023-01-01
Cyberpunk 2077 4K RT Ultra RT on DLSS Performance 60 fps — TweakTown 2020-12-08

AI Inference Performance

Tokens per second under each model + quantization. Higher = faster generation. Bars compare runs across the same model.

Local LLM inference throughput on the NVIDIA GeForce RTX 3090, in generated tokens per second. Higher is better; each row links to the community report or benchmark database it came from.
Model Quantization Relative Tokens/sec VRAM used Source
qwen3:35b-a3b Q4_K_XL llama.cpp 135.7 tok/s 21.0 GB GitHub (thc1006) 2026-04-19
qwen3.6-35b-a3b q3_K_M llama.cpp 120.0 tok/s 23.0 GB HuggingFace — Qwen3.6-35B-A3B Community 2025-04-01
qwen3.5-35b-a3b q4_K_M llama.cpp 112.0 tok/s 22.4 GB Medium — Code Pulse 2026-03-03
llama3.1:8b q4_K_M ollama 95.7 tok/s — LocalScore.ai 2025-01-01
llama3.1:8b q4_K_M llama.cpp 92.0 tok/s — kunalganglani.com LLM Benchmarks 2024-12-01
qwen3:8b q4_K_M llama.cpp 87.5 tok/s — Hardware Corner 2026-03-01
qwen2.5:14b q4_K_M ollama 55.8 tok/s — LocalScore.ai 2025-01-01
qwen2.5:14b q4_K_M llama.cpp 55.0 tok/s — kunalganglani.com LLM Benchmarks 2024-12-01
qwen3:14b q4_K_M llama.cpp 52.1 tok/s — Hardware Corner 2026-03-01
qwen3:0.6b — ollama 47.1 tok/s — LocalLLaMA 2026-04-15
llama3.1:8b FP16 vllm 42.0 tok/s 16.1 GB The Register 2024-08-23
codestral:22b q5_K_M llama.cpp 32.0 tok/s — Kunal Ganglani LLM Benchmarks 2024-01-01

Synthetic Benchmarks

Higher is better. Bars are scaled within each benchmark family (multi-thread, single-thread, etc.) so you can compare like-with-like at a glance.

Synthetic benchmark scores for the NVIDIA GeForce RTX 3090 — higher is better. Each row links to the public database the number was taken from.
Benchmark Relative Score Source
PassMark G3D Mark 26,548 pts PassMark 2026-04-20
PassMark G3D Mark 26,546 points PassMark Software 2026-04-29
3DMark Time Spy 18,824 points 3DMark 2022-06-01
3DMark Port Royal 15,039 points 3DMark 2020-11-30
3DMark Port Royal 13,905 points esports.hwbot.org 2021-03-12
3DMark Steel Nomad 5,340 points 3DMark 2024-08-28
3DMark Steel Nomad 5,117 points UL Benchmarks 2024-05-01
3DMark Speed Way 5,075 points 3DMark 2022-11-24
PassMark G2D Mark 1,069 pts PassMark 2026-04-20

Products Featuring the NVIDIA GeForce RTX 3090

Full Specifications

tdp w350
vram gb24
vram typeGDDR6X
cuda cores10496

NVIDIA GeForce RTX 3090 — Frequently Asked Questions

What is the NVIDIA GeForce RTX 3090 best used for?
NVIDIA GeForce RTX 3090 is positioned as a 24 GB VRAM Ampere-family graphics card. Use it for high-end 4K gaming and local LLM inference. See the synthetic + AI benchmark tables below for measured performance.
When was the NVIDIA GeForce RTX 3090 released, and what was its launch MSRP?
NVIDIA GeForce RTX 3090 launched in 2020 at a $1,499 MSRP. Street prices diverge from launch pricing over a product's lifetime — check the linked Amazon listings on this page for current availability.
Where do the benchmark numbers on this page come from?
Synthetic benchmarks are scraped from public databases (TechPowerUp, PassMark, Geekbench Browser, Cinebench leaderboards). AI inference numbers come from the LocalLLaMA community (Reddit threads, llama.cpp / Ollama discussion logs, and Phoronix when available). Every benchmark row carries an inline source citation — click through to verify the original number.
Can the NVIDIA GeForce RTX 3090 run local LLMs?
Yes — NVIDIA GeForce RTX 3090 has 12 AI inference benchmarks on file (see the AI Inference Performance section above for model + tokens-per-second numbers). With 24 GB VRAM, it fits the popular 32B-parameter open-weight models at Q4 quantization comfortably.
Where can I buy the NVIDIA GeForce RTX 3090?
Amazon listings are linked at the top of this page (in the hero CTA) and in the "Products Featuring this Hardware" section below. SpecPicks earns a small affiliate commission on qualifying purchases.

Buying guides that rank the NVIDIA GeForce RTX 3090's class

This page is the raw performance data. The guides below turn it into a ranked pick for a specific build.

Related GPUs

Editorial guides covering the NVIDIA GeForce RTX 3090

In-depth SpecPicks reviews, build guides, and head-to-heads referencing this graphics card.

More guides & deep dives from the SpecPicks archive

Browse all articles & guides →

More reviews from the SpecPicks archive

Browse all reviews →

More buying guides from SpecPicks

Browse all buying guides →
NVIDIA GeForce RTX 3090
NVIDIA GeForce RTX 3090
View on Amazon →