NVIDIA A100 80GB PCIe, AI & Machine Learning Benchmarks & Specs

80GB · AI Score 31.7/100 · first-party measured on 12 AI workloads

31.7 AI Score ✓ Measured

Every number on this page is first-party: NVIDIA A100 80GB PCIe was run on our pinned 12-workload AI suite on 2026-07-10, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA A100 80GB PCIe delivers about 161.75 tokens/sec. Stepping up to Qwen3 32B it holds roughly 43.77 tok/s. The full Llama 3.3 70B still runs, at about 22.89 tok/s. For image generation, SDXL runs at 8.4 it/s, and FLUX.1-dev at 1.85 it/s. All 12 workloads fit in 80GB. There is no model in our suite this card has to turn down. NVIDIA A100 80GB PCIe 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 5

Qwen3 4B198.27

Llama 3.1 8B161.75

Qwen2.5-Coder 14B88.36

Qwen3 32B43.77

Llama 3.3 70B22.89

Workload	Result	Telemetry	Data
Qwen3 4B	198.27 tok/s	2.9 GB peak114 W43°C1.74 tok/WQ4_K_M	✓ Measured
Llama 3.1 8B	161.75 tok/s	5.1 GB peak165 W46°C0.98 tok/WQ4_K_M	✓ Measured
Qwen2.5-Coder 14B	88.36 tok/s	8.8 GB peak179 W50°C0.5 tok/WQ4_K_M	✓ Measured
Qwen3 32B	43.77 tok/s	18.9 GB peak151 W53°C0.29 tok/WQ4_K_M	✓ Measured
Llama 3.3 70B	22.89 tok/s	40 GB peak162 W57°C0.14 tok/WQ4_K_M	✓ Measured

Image Generation images/min 3

Stable Diffusion XL16.8

Z-Image Turbo9.3

FLUX.1 dev3.964

Workload	Result	Telemetry	Data
Stable Diffusion XL	16.8 images/min	14.7 GB peak276 W57°C3.6 s/img	✓ Measured
Z-Image Turbo	9.3 images/min	25.8 GB peak293 W61°C6.4 s/img	✓ Measured
FLUX.1 dev	3.96 images/min	36.7 GB peak299 W66°C15.1 s/img	✓ Measured

Image Editing images/min 2

Workload	Result	Telemetry	Data
FLUX.1 Kontext dev	1.84 images/min	37.6 GB peak297 W75°C32.5 s/img	✓ Measured
Qwen-Image-Edit	1.54 images/min	60.2 GB peak297 W74°C39.2 s/img	✓ Measured

Video Generation frames/s 2

Workload	Result	Telemetry	Data
LTX-Video (distilled)	8.51 frames/s	60.1 GB peak287 W73°C11.4 s/clip	✓ Measured
Wan 2.2 5B (720p)	0.61 frames/s	60.3 GB peak289 W78°C80.9 s/clip	✓ 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 2026-07-10 · harness 2.0.0.

NVIDIA A100 80GB PCIe specifications

Architecture	Ampere
CUDA cores	6,912
VRAM	80GB HBM2e
Memory bus	5120-bit
Memory bandwidth	1935 GB/s
Boost clock	1,410 MHz
TDP	300 W
Process	TSMC 7nm
Interface	PCIe 4.0 x16
Release date	2021-06-28
Launch MSRP	$15,000

Verdict, NVIDIA A100 80GB PCIe on real AI workloads

NVIDIA A100 80GB PCIe scores 31.7/100, #15 of 102. It ran all 12 workloads. Every figure here is our own measurement.

Relative performance: where the NVIDIA A100 80GB PCIe lands

100% = this card, AI & Machine Learning headline metric (AI Score). #13 of 21 desktop cards in this vertical.

GPU	%	AI Score
NVIDIA RTX PRO 6000 Blackwell Server Edition	157%	49.9
NVIDIA H100 PCIe	146%	46.4
NVIDIA A100 80GB SXM4	104%	33.1
NVIDIA A800 80GB	104%	33.1
NVIDIA A100 80GB PCIe	100%	31.7
NVIDIA L40S	87%	27.7
NVIDIA L40	62%	19.5
NVIDIA A40	56%	17.7
NVIDIA A100 40GB SXM4	54%	17

One tier up: NVIDIA A800 80GB (+4%)
One tier down: NVIDIA L40S (-13%) · compare head-to-head

← All AI & Machine Learning GPU rankings

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	3.7 min	13.93 Wh	all 2 stages measured
60-second AI short film	4.6 min	18.6 Wh	all 3 stages measured
6-panel comic page	5.9 min	24.31 Wh	all 3 stages measured
Character sheet, 12 poses	7.7 min	33.46 Wh	all 2 stages measured
Full codebase review	11.3 min	33.67 Wh	measured
10 short social clips	15.7 min	70.16 Wh	all 3 stages measured
20 long-form articles	20.4 min	55.11 Wh	measured
40-product photo shoot	24.8 min	118.29 Wh	all 2 stages measured
100-photo restoration batch	54.7 min	268.4 Wh	measured

Rent or buy?

This card is $15,000 to buy. The cheapest listed rate on RunPod is $1.190/hour, but that is the floor: we budget $1.428/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 10,504 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	$1,042	14.4 years
8 hours a day, working on it	2,920	$4,170	3.6 years
24/7, always-on agent	8,760	$12,509	1.2 years

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

$1.190/hr+0.0% since 2026-08-14low $1.190 · high $1.190

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