NVIDIA GeForce RTX 3090, AI & Machine Learning Benchmarks & Specs

24GB · AI Score 7.7/100 · first-party measured on 12 AI workloads

7.7 AI Score ✓ Measured

Every number on this page is first-party: NVIDIA GeForce RTX 3090 was run on our pinned 12-workload AI suite on 2026-07-11, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA GeForce RTX 3090 delivers about 144.9 tokens/sec. Stepping up to Qwen3 32B it holds roughly 37.95 tok/s. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 24GB. For image generation, SDXL runs at 3.73 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 4 of the 12 workloads won't fit on 24GB at the tested precision, Llama 3.3 70B, FLUX.1-dev, FLUX.1 Kontext, Qwen-Image-Edit. We publish those as hard gates rather than quietly dropping to a smaller quant. NVIDIA GeForce RTX 3090 isn't a retail purchase for most people. It's rented by the hour. You can run this exact card on RunPod.

Bench notes: from the person who ran it

Still the used-market AI story. 1.34 tokens/watt is genuinely decent for Ampere, and the 24GB runs 8 of my 12 workloads including Qwen3 32B. Held 99% of its 350W rating on video gen without drama. If you want cheap VRAM, this is it. Quick note on the setup: all my AI benchmarking was done on rented cloud GPUs, I used all three of Vast.ai, RunPod and Modal depending on which had the card, and they all have their pros and cons. Same pinned harness on every run, and everything here got double-checked before it went up.

AI & Machine Learning benchmark results

Text Generation tok/s 28

Llama 3.2 1B632.21

Qwen3 0.6B583.17

LFM2.5 2.6B314.96

MiniCPM5 2B296.77

gpt-oss-20b231.97

Granite 4.1 3B227.46

Nemotron 3 Nano 4B222.39

Qwen3-Coder 30B A3B219.07

Qwen3 30B A3B Instruct 2507213.49

Qwen3 4B206.67

Qwen3 30B A3B202.26

DeepSeek Coder 7B Instruct v1.5164.74

Workload	Result	Telemetry	Data
Qwen3 0.6B	583.17 tok/s	178 W36°CQ4_K_M	✓ Measured
Llama 3.2 1B	632.21 tok/s	162 W41°CQ4_K_M	✓ Measured
MiniCPM5 2B	296.77 tok/s	203 W66°CQ4_K_M	✓ Measured
LFM2.5 2.6B	314.96 tok/s	218 W65°CQ4_K_M	✓ Measured
Granite 4.1 3B	227.46 tok/s	222 W64°CQ4_K_M	✓ Measured
Nemotron 3 Nano 4B	222.39 tok/s	224 W67°CQ4_K_M	✓ Measured
Qwen3 4B	206.67 tok/s	2.7 GB peak154 W37°C1.34 tok/WQ4_K_M	✓ Measured
DeepSeek Coder 7B Instruct v1.5	164.74 tok/s	236 W63°CQ4_K_M	✓ Measured
Qwen2.5-7B	156.51 tok/s	296 W45°CQ4_K_M	✓ Measured
Qwen2.5-Coder 7B	156.52 tok/s	294 W45°CQ4_K_M	✓ Measured
Llama 3 8B	146.34 tok/s	234 W62°CQ4_K_M	✓ Measured
Llama 3.1 8B	144.9 tok/s	4.9 GB peak179 W41°C0.81 tok/WQ4_K_M	✓ Measured
Qwen3 8B	137.59 tok/s	244 W69°CQ4_K_M	✓ Measured
Nemotron Nano 9B v2	108.86 tok/s	248 W63°CQ4_K_M	✓ Measured
Ornith 1.5 9B	124.77 tok/s	264 W66°CQ4_K_M	✓ Measured
Gemma 4 12B	85.77 tok/s	262 W70°CQ4_K_M	✓ Measured
Qwen2.5-Coder 14B	78.79 tok/s	8.7 GB peak178 W48°C0.44 tok/WQ4_K_M	✓ Measured
Qwen3 14B	81.11 tok/s	266 W70°CQ4_K_M	✓ Measured
gpt-oss-20b	231.97 tok/s	254 W45°CQ4_K_M	✓ Measured
Gemma 4 26B A4B	154.33 tok/s	231 W66°CQ4_K_M	✓ Measured
Qwen3.6 27B	42.51 tok/s	295 W66°CQ4_K_M	✓ Measured
Qwen3.8 27B	41.93 tok/s	295 W66°CQ4_K_M	✓ Measured
Qwen3 30B A3B	202.26 tok/s	204 W68°CQ4_K_M	✓ Measured
Qwen3 30B A3B Instruct 2507	213.49 tok/s	182 W64°CQ4_K_M	✓ Measured
Qwen3-Coder 30B A3B	219.07 tok/s	230 W41°CQ4_K_M	✓ Measured
Gemma 4 31B	39.2 tok/s	295 W66°CQ4_K_M	✓ Measured
Qwen3 32B	37.95 tok/s	18.7 GB peak136 W51°C0.28 tok/WQ4_K_M	✓ Measured
Llama 3.3 70B	✕ Won't fit needs ~46 GB	VRAM-gated at this precision	✓ Measured

Image Generation images/min 15

SDXL Turbo332.99

DreamShaper XL Lightning47.54

Stable Diffusion 1.538.24

DreamShaper XL Turbo27.3

FLUX.2 klein 4B21.26

Sana 1.6B20.6

PixArt-Sigma XL10.9

SSD-1B7.48

Stable Diffusion XL7.46

Playground v2.54.79

Kolors4.69

Z-Image Turbo3.825

Workload	Result	Telemetry	Data
Stable Diffusion 1.5	38.24 images/min	330 W56°C	✓ Measured
SDXL Turbo	332.99 images/min	144 W53°C	✓ Measured
SSD-1B	7.48 images/min	345 W63°C	✓ Measured
Sana 1.6B	20.6 images/min	327 W41°C	✓ Measured
Stable Diffusion XL	7.46 images/min	15.8 GB peak342 W57°C8 s/img	✓ Measured
DreamShaper XL Lightning	47.54 images/min	333 W54°C	✓ Measured
DreamShaper XL Turbo	27.3 images/min	343 W40°C	✓ Measured
Playground v2.5	4.79 images/min	346 W65°C	✓ Measured
PixArt-Sigma XL	10.9 images/min	327 W49°C	✓ Measured
FLUX.2 klein 4B	21.26 images/min	345 W64°C	✓ Measured
Kolors	4.69 images/min	348 W69°C	✓ Measured
Z-Image	0.63 images/min	348 W71°C	✓ Measured
Z-Image Turbo	3.83 images/min	23.1 GB peak344 W62°C15.6 s/img	✓ Measured
AuraFlow v0.3	1.5 images/min	329 W58°C	✓ Measured
FLUX.1 dev	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured

Image Editing images/min 2

Workload	Result	Telemetry	Data
FLUX.1 Kontext dev	✕ Won't fit needs ~26 GB	VRAM-gated at this precision	✓ Measured
Qwen-Image-Edit	✕ Won't fit needs ~42 GB	VRAM-gated at this precision	✓ Measured

Image to Video clips/min 5

LTX-Video (image to video)2.165

Stable Video Diffusion1.065

Wan 2.2 TI2V-5B (image to video)0.605

Stable Video Diffusion XT0.494

CogVideoX-5B I2V0.15

Workload	Result	Telemetry	Data
Stable Video Diffusion	1.07 clips/min	299 W77°C	✓ Measured
LTX-Video (image to video)	2.17 clips/min	291 W77°C	✓ Measured
Wan 2.2 TI2V-5B (image to video)	0.61 clips/min	233 W67°C	✓ Measured 2 hosts ±1% · CPU offload
Stable Video Diffusion XT	0.49 clips/min	247 W66°C	✓ Measured 2 hosts ±9% · CPU offload
CogVideoX-5B I2V	0.15 clips/min	242 W66°C	✓ Measured 2 hosts ±8% · CPU offload

Video Generation frames/s 5

LTX-Video (distilled)2.17

Wan 2.1 1.3B0.341

CogVideoX-2B0.241

Wan 2.2 5B (720p)0.22

CogVideoX-5B0.083

Workload	Result	Telemetry	Data
Wan 2.1 1.3B	0.34 frames/s	348 W66°C140.1 s/clip	✓ Measured 3 hosts ±8%
LTX-Video (distilled)	2.17 frames/s	9.3 GB peak240 W60°C44.8 s/clip	✓ Measured CPU offload
CogVideoX-2B	0.24 frames/s	348 W66°C198.1 s/clip	✓ Measured 3 hosts ±10%
CogVideoX-5B	0.08 frames/s	327 W78°C592.7 s/clip	✓ Measured CPU offload
Wan 2.2 5B (720p)	0.22 frames/s	16.9 GB peak314 W64°C221.8 s/clip	✓ Measured CPU offload

Image to 3D assets/hour 1

Workload	Result	Telemetry	Data
TRELLIS.2 Image-to-3D	30.1 assets/hour		✓ Measured

How we measured this. Every result comes from our own pinned, reproducible AI suite, 12 workloads: the Qwen3-4B to Llama-70B LLM ladder (llama.cpp, Q4_K_M), SDXL / Z-Image / FLUX-dev generation, FLUX-Kontext / Qwen-Edit editing, and LTX / Wan video, run first-party on rented hardware with under 0.5% run-to-run variance. Peak VRAM, power draw, temperature and tokens-per-watt are captured per workload. “Won’t fit” rows are real data: where a model exceeds the card’s VRAM at the tested precision we record a hard gate rather than silently dropping to a smaller quant. Measured 2026-07-11 · harness 2.0.0-standalone.

NVIDIA GeForce RTX 3090 specifications

Architecture	Ampere (GA102)
CUDA cores	10,496
VRAM	24GB GDDR6X
Memory bus	384-bit
Memory bandwidth	936.2 GB/s
Boost clock	1,695 MHz
TDP	350 W
Process	8nm
Interface	PCIe 4.0 x16
Release date	2020-09-24
Launch MSRP	$1,499

Verdict, capable, but 24GB sets the ceiling

NVIDIA GeForce RTX 3090 scores 7.7/100, #33 of 102. It ran 8 of 12; 4 exceeded its 24GB. Every figure here is our own measurement.

Relative performance: where the NVIDIA GeForce RTX 3090 lands

100% = this card, AI & Machine Learning headline metric (AI Score). #8 of 61 desktop cards in this vertical.

GPU	%	AI Score
NVIDIA RTX 5000 Ada Generation	143%	11
NVIDIA GeForce RTX 4090	138%	10.6
NVIDIA GeForce RTX 3090 Ti	110%	8.5
NVIDIA Titan RTX	106%	8.2
NVIDIA GeForce RTX 3090	100%	7.7
GeForce RTX 5080	68%	5.2
NVIDIA GeForce RTX 4080	62%	4.8
GeForce RTX 4080 Super	61%	4.7
GeForce RTX 5070 Ti	61%	4.7

One tier up: NVIDIA Titan RTX (+6%) · compare head-to-head
One tier down: GeForce RTX 5080 (-32%) · compare head-to-head

Same card, other workloads: NVIDIA GeForce RTX 3090 Gaming benchmarks

← All AI & Machine Learning GPU rankings

The silicon

Transistors	28,300 million
Die size	628.4 mm²
Process node	8 nm
Fabricated by	Samsung
Transistor density	45 million per mm²

Denser than 81% of the 746 cards we have silicon data for. Density is the clearest measure of what a process node bought: a card that gained it without growing the die got its speed from the fab rather than the architecture.

Silicon figures from Wikipedia (CC BY-SA 4.0). Benchmarks on this page are our own. Compare every chip.

What this card can build

Whole-job timings, composed from our measured per-model results on this card.

Workflow	Time	Energy	Basis
24-frame storyboard	8.3 min	37.34 Wh	all 2 stages measured
Full codebase review	12.7 min	37.63 Wh	measured
60-second AI short film	15.2 min	57.1 Wh	all 3 stages measured
Animate a batch of images	33.1 min	128.43 Wh	measured
Short social clips	42.4 min	209.99 Wh	all 3 stages measured

Can't run: Product photo shoot (needs FLUX.1 Kontext dev), Photo restoration batch (needs FLUX.1 Kontext dev), Restore and enlarge photos (needs FLUX.1 Kontext dev), Character sheet, 12 poses (needs FLUX.1 dev), 6-panel comic page (needs FLUX.1 dev), Long-form article batch (needs Llama 3.3 70B).

Rent or buy?

This card is $1,499 to buy. The cheapest listed rate on Vast.ai is $0.143/hour, but that is the floor: we budget $0.172/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 8,735 GPU-hours. Below it you are paying for idle silicon.

How you would use it	GPU-hours a year	Rental cost a year	Time to break even
2 hours a day, hobby	730	$125	12.0 years
8 hours a day, working on it	2,920	$501	3.0 years
24/7, always-on agent	8,760	$1,503	12.0 months

At hobby usage this card is very unlikely to pay for itself before it is superseded. Rent it. Rental figures include a 20% premium over the cheapest listed rate. Ignores electricity, resale and the fact that a rented card can be a newer one tomorrow.

Rental price

$0.143/hr+83.3% since 2026-08-14low $0.056 · high $0.156

Cheapest of the RunPod and Vast on-demand rates we see, sampled daily. Spot and interruptible pricing runs lower.