NVIDIA RTX 4000 (Ada Generation), AI & Machine Learning Benchmarks & Specs

20GB · AI Score 4.4/100 · first-party measured on 12 AI workloads

4.4 AI Score Includes estimates

Every number on this page is first-party: NVIDIA RTX 4000 (Ada Generation) 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 RTX 4000 (Ada Generation) delivers about 66.59 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 20GB. For image generation, SDXL runs at 3.28 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 5 of the 12 workloads won't fit on 20GB at the tested precision, Qwen3 32B, Llama 3.3 70B, FLUX.1-dev, FLUX.1 Kontext and others. We publish those as hard gates rather than quietly dropping to a smaller quant. NVIDIA RTX 4000 (Ada Generation) isn't a retail purchase for most people. It's rented by the hour. You can run this exact card on RunPod.

AI & Machine Learning benchmark results

Text Generation tok/s 17

MiniCPM5 2B183.38

LFM2.5 2.6B170.15

Granite 4.1 3B127.76

Qwen3 4B110.18

Nemotron 3 Nano 4B106.26

DeepSeek Coder 7B Instruct v1.575.46

Llama 3 8B66.88

Llama 3.1 8B66.59

Qwen3 8B65.39

Ornith 1.5 9B58.29

Nemotron Nano 9B v247.48

Gemma 4 12B42.23

Workload	Result	Telemetry	Data
MiniCPM5 2B	183.38 tok/s	73 W51°CQ4_K_M	✓ Measured
LFM2.5 2.6B	170.15 tok/s	78 W50°CQ4_K_M	✓ Measured
Granite 4.1 3B	127.76 tok/s	82 W59°CQ4_K_M	✓ Measured
Nemotron 3 Nano 4B	106.26 tok/s	88 W54°CQ4_K_M	✓ Measured
Qwen3 4B	110.18 tok/s	2.9 GB peak68 W46°C1.63 tok/WQ4_K_M	✓ Measured
DeepSeek Coder 7B Instruct v1.5	75.46 tok/s	100 W61°CQ4_K_M	✓ Measured
Llama 3 8B	66.88 tok/s	102 W61°CQ4_K_M	✓ Measured
Llama 3.1 8B	66.59 tok/s	4.8 GB peak82 W55°C0.81 tok/WQ4_K_M	✓ Measured
Qwen3 8B	65.39 tok/s	103 W61°CQ4_K_M	✓ Measured
Nemotron Nano 9B v2	47.48 tok/s	102 W61°CQ4_K_M	✓ Measured
Ornith 1.5 9B	58.29 tok/s	100 W57°CQ4_K_M	✓ Measured
Gemma 4 12B	42.23 tok/s	105 W60°CQ4_K_M	✓ Measured
Qwen2.5-Coder 14B	36.34 tok/s	8.6 GB peak83 W63°C0.44 tok/WQ4_K_M	✓ Measured
Qwen3 14B	36.81 tok/s	110 W62°CQ4_K_M	✓ Measured
Gemma 4 31B	✕ Won't fit needs ~22 GB	VRAM-gated at this precision	Est.
Qwen3 32B	✕ Won't fit needs ~23 GB	VRAM-gated at this precision	✓ Measured
Llama 3.3 70B	✕ Won't fit needs ~46 GB	VRAM-gated at this precision	✓ Measured

Image Generation images/min 6

Stable Diffusion 1.539.51

Sana 1.6B17.96

PixArt-Sigma XL10.45

Stable Diffusion XL6.56

Playground v2.54.32

Workload	Result	Telemetry	Data
Stable Diffusion 1.5	39.51 images/min	120 W57°C	✓ Measured
Sana 1.6B	17.96 images/min	122 W64°C	✓ Measured
Stable Diffusion XL	6.56 images/min	14.6 GB peak123 W66°C9.2 s/img	✓ Measured
Playground v2.5	4.32 images/min	124 W76°C	✓ Measured
PixArt-Sigma XL	10.45 images/min	123 W69°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

Video Generation frames/s 2

Workload	Result	Telemetry	Data
LTX-Video (distilled)	2.06 frames/s	9.2 GB peak96 W70°C47.1 s/clip	✓ Measured CPU offload
Wan 2.2 5B (720p)	0.18 frames/s	16.6 GB peak116 W79°C269.7 s/clip	✓ Measured CPU offload

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 RTX 4000 (Ada Generation) specifications

Architecture	Ada Lovelace
CUDA cores	6,144
VRAM	20GB GDDR6 ECC
Memory bus	160-bit
Memory bandwidth	360 GB/s
Boost clock	2,175 MHz
TDP	130 W
Process	4nm
Interface	PCIe 4.0 x16
Release date	2023-08-09
Launch MSRP	$1,250

Verdict, capable, but 20GB sets the ceiling

NVIDIA RTX 4000 (Ada Generation) scores 4.4/100, #46 of 102. It ran 6 of 12; 5 exceeded its 20GB. Every figure here is our own measurement.

Relative performance: where the NVIDIA RTX 4000 (Ada Generation) lands

100% = this card, AI & Machine Learning headline metric (AI Score). #13 of 61 desktop cards in this vertical.

GPU	%	AI Score
GeForce RTX 5080	118%	5.2
NVIDIA GeForce RTX 4080	109%	4.8
GeForce RTX 4080 Super	107%	4.7
GeForce RTX 5070 Ti	107%	4.7
NVIDIA RTX 4000 (Ada Generation)	100%	4.4
NVIDIA GeForce RTX 4070 Ti Super	98%	4.3
AMD Radeon RX 7900 XTX	84%	3.7
GeForce RTX 5060 Ti	84%	3.7
NVIDIA GeForce RTX 3080 Ti	80%	3.5

One tier up: GeForce RTX 5070 Ti (+7%) · compare head-to-head
One tier down: NVIDIA GeForce RTX 4070 Ti Super (-2%) · compare head-to-head
Best value alternative ($749): GeForce RTX 5070 Ti (+7%) · compare head-to-head

← All AI & Machine Learning GPU rankings

The silicon

Transistors	35,800 million
Die size	294.5 mm²
Process node	4 nm
Fabricated by	TSMC
Transistor density	121.6 million per mm²

Denser than 94% 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
Full codebase review	27.5 min	38.11 Wh	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), Short social clips (needs Qwen3 32B), 60-second AI short film (needs Qwen3 32B), 24-frame storyboard (needs Qwen3 32B), 6-panel comic page (needs Qwen3 32B), Long-form article batch (needs Llama 3.3 70B).

Rent or buy?

This card is $1,250 to buy. The cheapest listed rate on Vast.ai is $0.175/hour, but that is the floor: we budget $0.210/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 5,952 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	$153	8.2 years
8 hours a day, working on it	2,920	$613	2.0 years
24/7, always-on agent	8,760	$1,840	8.2 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

$0.200/hr+0.0% since 2026-08-14low $0.175 · high $0.200

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