Hardware / Apple /Apple M3 Ultra (80-core GPU)

Apple · Apple Silicon

Apple M3 Ultra (80-core GPU)

for example Mac Studio (2025)

Runs 175 models fully in memory, 9 more with CPU offload. 1 of those sit close enough to 512 GB that our size estimate is what decides them.

Memory

512 GB

Bandwidth

819 GB/s

Runs

175

01

Specifications

Memory

512 GB

Bandwidth

819 GB/s

FP16 compute

56.8 TFLOPS

TDP

140 W

MSRP

—

Street price

—

Released

Mar 12, 2025

Architecture

M3

02

What it runs

Best quant per model — sorted by fit score.

ModelParamsBest quantContextSpeedVerdict

Qwen3.6 35B A3B36BQ4_K_M262K268 tok/sestFits in memory Gemma 4 26B A4B26.5BQ4_K_M262K201 tok/sestFits in memory Nemotron 3 Nano 30B A3B31.6BQ4_K_M262K268 tok/sestFits in memory Ministral 3 3B 25123.8BQ4_K_M131K249 tok/sestFits in memory Granite 4.0 Micro3.2BQ4_K_M66K251 tok/sestFits in memory Qwen3 VL 30B A3B Thinking31.1BQ8_0262K153 tok/sestFits in memory Qwen3 Next 80B A3B Thinking81.3BQ8_0262K153 tok/sestFits in memory Llama 3.2 3B Instruct3.2BQ8_0131K169 tok/sestFits in memory Qwen3.5-35B-A3B36BQ8_0262K153 tok/sestFits in memory Qwen3 VL 30B A3B Instruct31.1BQ5_K_M262K228 tok/sestFits in memory Qwen3 Next 80B A3B Instruct81.3BQ4_K_M262K268 tok/sestFits in memory Qwen3 30B A3B Thinking 250730.5BQ5_K_M66K228 tok/sestFits in memory Qwen3 Coder 30B A3B Instruct30.5BQ4_K_M262K268 tok/sestFits in memory Qwen3 30B A3B Instruct 250730.5BQ5_K_M262K228 tok/sestFits in memory Gemma 3 4B4.3BQ8_0131K126 tok/sestFits in memory Llama 3.1 8B Instruct8BQ4_K_M131K118 tok/sestFits in memory Niagara 19m Batch.en20MQ8_0~262K26979 tok/sestFits in memory Niagara 38m Batch.en38MQ8_0~262K14199 tok/sestFits in memory ARK ASR 0.6B1.3BQ5_K_M~262K620 tok/sestFits in memory ARK ASR 3B4.1BQ4_K_M~262K231 tok/sestFits in memory Audio8 ASR 0.1B0.3BQ4_K_M~262K3152 tok/sestFits in memory Higgs Audio v3 STT2.7BQ8_0~262K200 tok/sestFits in memory Cohere Transcribe 03 20262.1BQ4_K_M~262K450 tok/sestFits in memory Distil Large v3.50.8BQ4_K_M~262K1182 tok/sestFits in memory Lite Whisper Large v3 Acc1.4BQ4_K_M~262K676 tok/sestFits in memory Owsm CTC v3.1 1B1.1BQ4_K_M~262K860 tok/sestFits in memory Owsm CTC v3.2 ft 1B1BQ8_0~262K540 tok/sestFits in memory Owsm CTC v4 1B1BQ8_0~262K540 tok/sestFits in memory Hubert Large Ls960 ft0.3BQ8_0~262K1686 tok/sestFits in memory MMS 1b All1BQ5_K_M~262K806 tok/sestFits in memory Wav2vec2 Large 960h Lv60 Self0.3BQ5_K_M~262K2519 tok/sestFits in memory Hojo ASR V15.2BQ4_K_M~262K183 tok/sestFits in memory Granite 4.0 1b Speech2.3BQ5_K_M~262K351 tok/sestFits in memory Granite Speech 3.3 2b3BQ5_K_M~262K269 tok/sestFits in memory Granite Speech 4.1 2b2.3BQ4_K_M~262K411 tok/sestFits in memory Granite Speech 4.1 2b NAR2.3BQ8_0~262K235 tok/sestFits in memory STT 2.6b en2.6BQ8_0~262K208 tok/sestFits in memory Phi 4 Multimodal Instruct5.6BQ4_K_M~262K169 tok/sestFits in memory Voxtral Mini 3B 25075BQ4_K_M~262K189 tok/sestFits in memory Voxtral Mini 4B Realtime 26024.4BQ4_K_M~262K215 tok/sestFits in memory Canary 180m Flash0.2BQ8_0~262K2998 tok/sestFits in memory Canary 1b1BQ5_K_M~262K806 tok/sestFits in memory Canary 1b Flash1BQ4_K_M~262K946 tok/sestFits in memory Canary 1b v21BQ4_K_M~262K946 tok/sestFits in memory Canary Qwen 2.5b2.6BQ5_K_M~262K310 tok/sestFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6BQ8_0~262K843 tok/sestFits in memory Nemotron Speech Streaming en 0.6b0.6BQ5_K_M~262K1343 tok/sestFits in memory Parakeet CTC 0.6b0.6BQ4_K_M~262K1576 tok/sestFits in memory Parakeet CTC 1.1b1.1BQ5_K_M~262K733 tok/sestFits in memory Parakeet RNNT 0.6b0.6BQ8_0~262K899 tok/sestFits in memory Parakeet RNNT 1.1b1.1BQ4_K_M~262K860 tok/sestFits in memory Parakeet TDT CTC 110m0.1BQ8_0~262K4905 tok/sestFits in memory Parakeet TDT 0.6b v20.6BQ8_0~262K899 tok/sestFits in memory Parakeet TDT 0.6b v30.6BQ8_0~262K899 tok/sestFits in memory Parakeet TDT 1.1b1.1BQ8_0~262K491 tok/sestFits in memory Nyra Health CrisperWhisper1.6BQ8_0~262K337 tok/sestFits in memory Whisper Large v31.5BQ8_0~262K360 tok/sestFits in memory Whisper Large v3 Turbo0.8BQ5_K_M~262K1008 tok/sestFits in memory MOSS Transcribe Diarize0.9BQ4_K_M~262K1051 tok/sestFits in memory MOSS Transcribe Preview 2B2.4BQ4_K_M~262K394 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