Hardware / AMD /Radeon RX 7900 XT

AMD · GPU

Radeon RX 7900 XT

Runs 94 models fully in memory, 22 more with CPU offload. 1 of those sit close enough to 20 GB that our size estimate is what decides them.

Memory

20 GB

Bandwidth

800 GB/s

Runs

94

01

Specifications

Memory

20 GB

Bandwidth

800 GB/s

FP16 compute

103 TFLOPS

TDP

315 W

MSRP

$899

Street price

—

Released

Dec 13, 2022

Architecture

RDNA 3

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Ministral 3 3B 25123.8BQ4_K_M131K226 tok/sestFits in memory Granite 4.0 Micro3.2BQ4_K_M66K228 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ4_K_M131K268 tok/sestFits in memory Gemma 3 4B4.3BQ5_K_M131K170 tok/sestFits in memory Niagara 19m Batch.en20MQ8_0~262K24471 tok/sestFits in memory Niagara 38m Batch.en38MQ5_K_M~262K19240 tok/sestFits in memory ARK ASR 0.6B1.3BQ5_K_M~262K562 tok/sestFits in memory ARK ASR 3B4.1BQ5_K_M~262K178 tok/sestFits in memory Audio8 ASR 0.1B0.3BQ5_K_M~262K2437 tok/sestFits in memory Higgs Audio v3 STT2.7BQ4_K_M~131K318 tok/sestFits in memory Cohere Transcribe 03 20262.1BQ8_0~262K233 tok/sestFits in memory Distil Large v3.50.8BQ8_0~262K612 tok/sestFits in memory Lite Whisper Large v3 Acc1.4BQ8_0~262K350 tok/sestFits in memory Owsm CTC v3.1 1B1.1BQ5_K_M~262K665 tok/sestFits in memory Owsm CTC v3.2 ft 1B1BQ8_0~262K489 tok/sestFits in memory Owsm CTC v4 1B1BQ8_0~262K489 tok/sestFits in memory Hubert Large Ls960 ft0.3BQ4_K_M~262K2680 tok/sestFits in memory MMS 1b All1BQ8_0~262K489 tok/sestFits in memory Wav2vec2 Large 960h Lv60 Self0.3BQ5_K_M~262K2285 tok/sestFits in memory Hojo ASR V15.2BQ5_K_M~262K141 tok/sestFits in memory Granite 4.0 1b Speech2.3BQ8_0~131K213 tok/sestFits in memory Granite Speech 3.3 2b3BQ4_K_M~131K286 tok/sestFits in memory Granite Speech 4.1 2b2.3BQ4_K_M~131K373 tok/sestFits in memory Granite Speech 4.1 2b NAR2.3BQ8_0~131K213 tok/sestFits in memory STT 2.6b en2.6BQ4_K_M~262K330 tok/sestFits in memory Phi 4 Multimodal Instruct5.6BQ4_K_M~66K153 tok/sestFits in memory Voxtral Mini 3B 25075BQ4_K_M~262K172 tok/sestFits in memory Voxtral Mini 4B Realtime 26024.4BQ5_K_M~131K166 tok/sestFits in memory Canary 180m Flash0.2BQ4_K_M~262K4765 tok/sestFits in memory Canary 1b1BQ5_K_M~262K731 tok/sestFits in memory Canary 1b Flash1BQ4_K_M~262K858 tok/sestFits in memory Canary 1b v21BQ4_K_M~262K858 tok/sestFits in memory Canary Qwen 2.5b2.6BQ4_K_M~262K330 tok/sestFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6BQ8_0~262K765 tok/sestFits in memory Nemotron Speech Streaming en 0.6b0.6BQ8_0~262K816 tok/sestFits in memory Parakeet CTC 0.6b0.6BQ5_K_M~262K1219 tok/sestFits in memory Parakeet CTC 1.1b1.1BQ5_K_M~262K665 tok/sestFits in memory Parakeet RNNT 0.6b0.6BQ4_K_M~262K1430 tok/sestFits in memory Parakeet RNNT 1.1b1.1BQ5_K_M~262K665 tok/sestFits in memory Parakeet TDT CTC 110m0.1BQ8_0~262K4449 tok/sestFits in memory Parakeet TDT 0.6b v20.6BQ5_K_M~262K1219 tok/sestFits in memory Parakeet TDT 0.6b v30.6BQ4_K_M~262K1430 tok/sestFits in memory Parakeet TDT 1.1b1.1BQ5_K_M~262K665 tok/sestFits in memory Nyra Health CrisperWhisper1.6BQ5_K_M~262K457 tok/sestFits in memory Whisper Large v31.5BQ5_K_M~262K487 tok/sestFits in memory Whisper Large v3 Turbo0.8BQ8_0~262K612 tok/sestFits in memory MOSS Transcribe Diarize0.9BQ8_0~131K544 tok/sestFits in memory Qwen3 ASR 0.6B0.8BQ5_K_M~262K937 tok/sestFits in memory Qwen3 ASR 0.6B HF0.8BQ4_K_M~131K1072 tok/sestFits in memory Qwen3 ASR 1.7B2BQ8_0~262K240 tok/sestFits in memory Qwen3 ASR 1.7B HF2BQ4_K_M~262K421 tok/sestFits in memory Zipformer cr CTC Transducer XL 290M0.3BQ4_K_M~262K2958 tok/sestFits in memory Zipformer Transducer XL 290M0.3BQ8_0~262K1688 tok/sestFits in memory ASR Conformer Largescaleasr0.5BQ5_K_M~262K1523 tok/sestFits in memory Moonshine Streaming Medium0.2BQ5_K_M~262K3046 tok/sestFits in memory Moonshine Streaming Small0.1BQ8_0~262K4894 tok/sestFits in memory Moonshine Streaming Tiny30MQ8_0~262K16314 tok/sestFits in memory Moonshine Tiny30MQ8_0~262K16314 tok/sestFits in memory GLM ASR Nano 25122.3BQ8_0~131K213 tok/sestFits in memory Llama 3.2 1B Instruct1.2BQ5_K_M33K609 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