Hardware / NVIDIA /GeForce RTX 5090

NVIDIA · GPU

GeForce RTX 5090

Runs 116 models fully in memory, 0 more with CPU offload.

Memory

32 GB

Bandwidth

1792 GB/s

Runs

116

01

Specifications

Memory

32 GB

Bandwidth

1792 GB/s

FP16 compute

104.8 TFLOPS

TDP

575 W

MSRP

$1,999

Street price

—

Released

Jan 30, 2025

Architecture

Blackwell

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Granite 4.1 8B8.8BQ8_066K153 tok/sestFits in memory Qwen3.6 35B A3B36BQ4_K_M66K670 tok/sestFits in memory Qwen3.5-9B9.7BQ4_K_M131K244 tok/sestFits in memory Nemotron 3 Nano 30B A3B31.6BQ4_K_M131K670 tok/sestFits in memory Ministral 3 14B 251213.9BQ4_K_M66K170 tok/sestFits in memory Ministral 3 8B 25128.9BQ4_K_M131K266 tok/sestFits in memory Ministral 3 3B 25123.8BQ4_K_M131K622 tok/sestFits in memory Granite 4.0 Micro3.2BQ4_K_M66K628 tok/sestFits in memory Qwen3 VL 8B Thinking8.8BQ4_K_M131K269 tok/sestFits in memory Qwen3 VL 30B A3B Thinking31.1BQ4_K_M66K670 tok/sestFits in memory UI-TARS 7B8.3BQ8_066K163 tok/sestFits in memory Llama Guard 4 12B12BQ4_K_M262K197 tok/sestFits in memory Qwen3 14B14.8BQ5_K_M66K136 tok/sestFits in memory Gemma 3 12B12.2BQ4_K_M131K194 tok/sestFits in memory Rocinante 12B12.2BQ4_K_M66K194 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ5_K_M131K630 tok/sestFits in memory Mistral Nemo12.2BQ5_K_M66K165 tok/sestFits in memory Qwen3.5-35B-A3B36BQ4_K_M66K670 tok/sestFits in memory Qwen3 VL 8B Instruct8.8BQ4_K_M131K269 tok/sestFits in memory Qwen3 VL 30B A3B Instruct31.1BQ4_K_M66K670 tok/sestFits in memory Qwen3 30B A3B Thinking 250730.5BQ4_K_M66K670 tok/sestFits in memory Qwen3 Coder 30B A3B Instruct30.5BQ4_K_M66K670 tok/sestFits in memory Qwen3 30B A3B Instruct 250730.5BQ4_K_M66K670 tok/sestFits in memory Qwen3 8B8.2BQ4_K_M131K288 tok/sestFits in memory Gemma 3 4B4.3BQ4_K_M131K550 tok/sestFits in memory Llama 3.1 8B Instruct8BQ8_0131K169 tok/sestFits in memory Niagara 19m Batch.en20MQ8_0~262K67464 tok/sestFits in memory Niagara 38m Batch.en38MQ5_K_M~262K53042 tok/sestFits in memory ARK ASR 0.6B1.3BQ5_K_M~262K1551 tok/sestFits in memory ARK ASR 3B4.1BQ5_K_M~262K492 tok/sestFits in memory Audio8 ASR 0.1B0.3BQ5_K_M~262K6719 tok/sestFits in memory Higgs Audio v3 8b STT v28.9BQ8_0~131K152 tok/sestFits in memory Higgs Audio v3 STT2.7BQ4_K_M~131K876 tok/sestFits in memory Cohere Transcribe 03 20262.1BQ8_0~262K643 tok/sestFits in memory Distil Large v3.50.8BQ8_0~262K1687 tok/sestFits in memory Lite Whisper Large v3 Acc1.4BQ8_0~262K964 tok/sestFits in memory Owsm CTC v3.1 1B1.1BQ5_K_M~262K1832 tok/sestFits in memory Owsm CTC v3.2 ft 1B1BQ8_0~262K1349 tok/sestFits in memory Owsm CTC v4 1B1BQ8_0~262K1349 tok/sestFits in memory Hubert Large Ls960 ft0.3BQ4_K_M~262K7390 tok/sestFits in memory MMS 1b All1BQ8_0~262K1349 tok/sestFits in memory Wav2vec2 Large 960h Lv60 Self0.3BQ5_K_M~262K6299 tok/sestFits in memory Hojo ASR V15.2BQ5_K_M~262K389 tok/sestFits in memory Granite 4.0 1b Speech2.3BQ8_0~262K587 tok/sestFits in memory Granite Speech 3.3 2b3BQ8_0~262K450 tok/sestFits in memory Granite Speech 3.3 8b8.6BQ8_0~66K157 tok/sestFits in memory Granite Speech 4.1 2b2.3BQ8_0~262K587 tok/sestFits in memory Granite Speech 4.1 2b NAR2.3BQ8_0~262K587 tok/sestFits in memory STT 2.6b en2.6BQ4_K_M~262K910 tok/sestFits in memory Phi 4 Multimodal Instruct5.6BQ8_0~131K241 tok/sestFits in memory VibeVoice ASR HF8.3BQ8_0~262K163 tok/sestFits in memory Voxtral Mini 3B 25075BQ5_K_M~262K403 tok/sestFits in memory Voxtral Mini 4B Realtime 26024.4BQ4_K_M~131K537 tok/sestFits in memory Canary 180m Flash0.2BQ4_K_M~262K13137 tok/sestFits in memory Canary 1b1BQ5_K_M~262K2016 tok/sestFits in memory Canary 1b Flash1BQ4_K_M~262K2365 tok/sestFits in memory Canary 1b v21BQ4_K_M~262K2365 tok/sestFits in memory Canary Qwen 2.5b2.6BQ4_K_M~262K910 tok/sestFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6BQ8_0~262K2108 tok/sestFits in memory Nemotron Speech Streaming en 0.6b0.6BQ8_0~262K2249 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