Hardware / AMD /Radeon RX 9070 XT

AMD · GPU

Radeon RX 9070 XT

Runs 86 models fully in memory, 20 more with CPU offload.

Memory

16 GB

Bandwidth

640 GB/s

Runs

86

01

Specifications

Memory

16 GB

Bandwidth

640 GB/s

FP16 compute

97.3 TFLOPS

TDP

304 W

MSRP

$599

Street price

—

Released

Mar 6, 2025

Architecture

RDNA 4

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Ministral 3 3B 25123.8BQ4_K_M66K181 tok/sestFits in memory Granite 4.0 Micro3.2BQ4_K_M66K182 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ4_K_M131K214 tok/sestFits in memory Gemma 3 4B4.3BQ4_K_M131K160 tok/sestFits in memory Niagara 19m Batch.en20MQ4_K_M~262K34309 tok/sestFits in memory Niagara 38m Batch.en38MQ4_K_M~262K18058 tok/sestFits in memory ARK ASR 0.6B1.3BQ4_K_M~262K528 tok/sestFits in memory ARK ASR 3B4.1BQ4_K_M~262K167 tok/sestFits in memory Audio8 ASR 0.1B0.3BQ8_0~262K1305 tok/sestFits in memory Higgs Audio v3 STT2.7BQ8_0~66K145 tok/sestFits in memory Cohere Transcribe 03 20262.1BQ5_K_M~262K279 tok/sestFits in memory Distil Large v3.50.8BQ5_K_M~262K731 tok/sestFits in memory Lite Whisper Large v3 Acc1.4BQ8_0~262K280 tok/sestFits in memory Owsm CTC v3.1 1B1.1BQ4_K_M~262K624 tok/sestFits in memory Owsm CTC v3.2 ft 1B1BQ5_K_M~262K585 tok/sestFits in memory Owsm CTC v4 1B1BQ4_K_M~262K686 tok/sestFits in memory Hubert Large Ls960 ft0.3BQ8_0~262K1224 tok/sestFits in memory MMS 1b All1BQ8_0~262K392 tok/sestFits in memory Wav2vec2 Large 960h Lv60 Self0.3BQ5_K_M~262K1828 tok/sestFits in memory Hojo ASR V15.2BQ4_K_M~131K133 tok/sestFits in memory Granite 4.0 1b Speech2.3BQ4_K_M~131K298 tok/sestFits in memory Granite Speech 3.3 2b3BQ4_K_M~131K229 tok/sestFits in memory Granite Speech 4.1 2b2.3BQ8_0~131K170 tok/sestFits in memory Granite Speech 4.1 2b NAR2.3BQ4_K_M~131K298 tok/sestFits in memory STT 2.6b en2.6BQ8_0~262K151 tok/sestFits in memory Phi 4 Multimodal Instruct5.6BQ4_K_M~66K123 tok/sestFits in memory Voxtral Mini 3B 25075BQ4_K_M~131K137 tok/sestFits in memory Voxtral Mini 4B Realtime 26024.4BQ5_K_M~66K133 tok/sestFits in memory Canary 180m Flash0.2BQ4_K_M~262K3812 tok/sestFits in memory Canary 1b1BQ4_K_M~262K686 tok/sestFits in memory Canary 1b Flash1BQ4_K_M~262K686 tok/sestFits in memory Canary 1b v21BQ8_0~262K392 tok/sestFits in memory Canary Qwen 2.5b2.6BQ4_K_M~262K264 tok/sestFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6BQ4_K_M~262K1072 tok/sestFits in memory Nemotron Speech Streaming en 0.6b0.6BQ8_0~262K653 tok/sestFits in memory Parakeet CTC 0.6b0.6BQ5_K_M~262K975 tok/sestFits in memory Parakeet CTC 1.1b1.1BQ5_K_M~262K532 tok/sestFits in memory Parakeet RNNT 0.6b0.6BQ8_0~262K653 tok/sestFits in memory Parakeet RNNT 1.1b1.1BQ4_K_M~262K624 tok/sestFits in memory Parakeet TDT CTC 110m0.1BQ8_0~262K3559 tok/sestFits in memory Parakeet TDT 0.6b v20.6BQ8_0~262K653 tok/sestFits in memory Parakeet TDT 0.6b v30.6BQ4_K_M~262K1144 tok/sestFits in memory Parakeet TDT 1.1b1.1BQ5_K_M~262K532 tok/sestFits in memory Nyra Health CrisperWhisper1.6BQ4_K_M~262K429 tok/sestFits in memory Whisper Large v31.5BQ5_K_M~262K390 tok/sestFits in memory Whisper Large v3 Turbo0.8BQ8_0~262K489 tok/sestFits in memory MOSS Transcribe Diarize0.9BQ4_K_M~66K762 tok/sestFits in memory Qwen3 ASR 0.6B0.8BQ5_K_M~262K750 tok/sestFits in memory Qwen3 ASR 0.6B HF0.8BQ5_K_M~66K731 tok/sestFits in memory Qwen3 ASR 1.7B2BQ8_0~262K192 tok/sestFits in memory Qwen3 ASR 1.7B HF2BQ8_0~262K192 tok/sestFits in memory Zipformer cr CTC Transducer XL 290M0.3BQ5_K_M~262K2017 tok/sestFits in memory Zipformer Transducer XL 290M0.3BQ4_K_M~262K2366 tok/sestFits in memory ASR Conformer Largescaleasr0.5BQ8_0~262K816 tok/sestFits in memory Moonshine Streaming Medium0.2BQ4_K_M~262K2859 tok/sestFits in memory Moonshine Streaming Small0.1BQ8_0~262K3915 tok/sestFits in memory Moonshine Streaming Tiny30MQ5_K_M~262K19496 tok/sestFits in memory Moonshine Tiny30MQ5_K_M~262K19496 tok/sestFits in memory GLM ASR Nano 25122.3BQ8_0~131K170 tok/sestFits in memory Llama 3.2 1B Instruct1.2BQ5_K_M33K487 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