Hardware / NVIDIA /A100 80GB SXM

NVIDIA · GPU

A100 80GB SXM

Runs 136 models fully in memory, 8 more with CPU offload. 4 of those sit close enough to 80 GB that our size estimate is what decides them.

Memory

80 GB

Bandwidth

2039 GB/s

Runs

136

01

Specifications

Memory

80 GB

Bandwidth

2039 GB/s

FP16 compute

312 TFLOPS

TDP

400 W

MSRP

$15,000

Street price

—

Released

Nov 16, 2020

Architecture

Ampere

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Granite 4.1 8B8.8BQ8_0131K175 tok/sestFits in memory Qwen3.6 35B A3B36BQ4_K_M262K762 tok/sestFits in memory Gemma 4 26B A4B26.5BQ8_0131K326 tok/sestFits in memory Qwen3.5-9B9.7BQ5_K_M262K236 tok/sestFits in memory Nemotron 3 Nano 30B A3B31.6BQ5_K_M262K650 tok/sestFits in memory Ministral 3 14B 251213.9BQ4_K_M262K194 tok/sestFits in memory Ministral 3 8B 25128.9BQ5_K_M262K258 tok/sestFits in memory Ministral 3 3B 25123.8BQ8_0131K404 tok/sestFits in memory Granite 4.0 Micro3.2BQ5_K_M66K609 tok/sestFits in memory Qwen3 VL 8B Thinking8.8BQ8_0131K175 tok/sestFits in memory Qwen3 VL 30B A3B Thinking31.1BQ8_0262K435 tok/sestFits in memory Qwen3 Next 80B A3B Thinking81.3BQ4_K_M131K762 tok/sestFits in memory UI-TARS 7B8.3BQ8_066K185 tok/sestFits in memory Hunyuan A13B Instruct80.4BQ5_K_M66K150 tok/sestFits in memory Llama Guard 4 12B12BQ4_K_M262K224 tok/sestFits in memory Qwen3 14B14.8BQ4_K_M131K182 tok/sestFits in memory Gemma 3 12B12.2BQ5_K_M131K188 tok/sestFits in memory Rocinante 12B12.2BQ5_K_M66K188 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ8_0131K480 tok/sestFits in memory Mistral Nemo12.2BQ5_K_M131K188 tok/sestFits in memory Qwen3.5-35B-A3B36BQ4_K_M262K762 tok/sestFits in memory Qwen3 VL 8B Instruct8.8BQ4_K_M262K306 tok/sestFits in memory Qwen3 VL 30B A3B Instruct31.1BQ5_K_M262K650 tok/sestFits in memory Qwen3 Next 80B A3B Instruct81.3BQ4_K_M131K762 tok/sestFits in memory Qwen3 30B A3B Thinking 250730.5BQ4_K_M66K762 tok/sestFits in memory Qwen3 Coder 30B A3B Instruct30.5BQ8_0262K435 tok/sestFits in memory Qwen3 30B A3B Instruct 250730.5BQ4_K_M262K762 tok/sestFits in memory Qwen3 8B8.2BQ4_K_M131K328 tok/sestFits in memory Gemma 3 4B4.3BQ8_0131K357 tok/sestFits in memory Reka Flash 320.9BQ4_K_M66K129 tok/sestFits in memory Llama 3.1 8B Instruct8BQ8_0131K192 tok/sestFits in memory Niagara 19m Batch.en20MQ8_0~262K76762 tok/sestFits in memory Niagara 38m Batch.en38MQ5_K_M~262K60353 tok/sestFits in memory ARK ASR 0.6B1.3BQ8_0~262K1181 tok/sestFits in memory ARK ASR 3B4.1BQ8_0~262K375 tok/sestFits in memory Audio8 ASR 0.1B0.3BQ5_K_M~262K7645 tok/sestFits in memory Higgs Audio v3 8b STT v28.9BQ8_0~262K173 tok/sestFits in memory Higgs Audio v3 STT2.7BQ8_0~262K569 tok/sestFits in memory Cohere Transcribe 03 20262.1BQ4_K_M~262K1281 tok/sestFits in memory Distil Large v3.50.8BQ4_K_M~262K3363 tok/sestFits in memory Lite Whisper Large v3 Acc1.4BQ8_0~262K1097 tok/sestFits in memory Owsm CTC v3.1 1B1.1BQ8_0~262K1396 tok/sestFits in memory Owsm CTC v3.2 ft 1B1BQ4_K_M~262K2691 tok/sestFits in memory Owsm CTC v4 1B1BQ4_K_M~262K2691 tok/sestFits in memory Hubert Large Ls960 ft0.3BQ4_K_M~262K8408 tok/sestFits in memory MMS 1b All1BQ8_0~262K1535 tok/sestFits in memory Wav2vec2 Large 960h Lv60 Self0.3BQ8_0~262K4798 tok/sestFits in memory Hojo ASR V15.2BQ8_0~262K296 tok/sestFits in memory Granite 4.0 1b Speech2.3BQ4_K_M~262K1170 tok/sestFits in memory Granite Speech 3.3 2b3BQ5_K_M~262K765 tok/sestFits in memory Granite Speech 3.3 8b8.6BQ4_K_M~262K313 tok/sestFits in memory Granite Speech 4.1 2b2.3BQ4_K_M~262K1170 tok/sestFits in memory Granite Speech 4.1 2b NAR2.3BQ5_K_M~262K997 tok/sestFits in memory STT 2.6b en2.6BQ5_K_M~262K882 tok/sestFits in memory Phi 4 Multimodal Instruct5.6BQ8_0~262K274 tok/sestFits in memory VibeVoice ASR HF8.3BQ8_0~262K185 tok/sestFits in memory Voxtral Mini 3B 25075BQ4_K_M~262K538 tok/sestFits in memory Voxtral Mini 4B Realtime 26024.4BQ4_K_M~262K612 tok/sestFits in memory Canary 180m Flash0.2BQ4_K_M~262K14948 tok/sestFits in memory Canary 1b1BQ5_K_M~262K2293 tok/sestFits in memory

A speed marked est is calculated from this card's memory bandwidth and the size of the weights it has to read per word. It is not a measurement.

Something wrong on this page? Tell us