Pinchflat

Local AI on the Framework Desktop: Best Models for 32GB, 64GB, and 128GB

Media Tasks
Actions

Raw Attributes

Source: Framework
  • nfo_filepath:
  • duration_seconds: 2363
  • prevent_culling: false
  • playlist_index: 0
  • media_filepath: /downloads/Framework/2026-09-02 Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB/Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB [mmntN7zIekU].mp4
  • description: What can you run locally on the 32GB, 64GB, and 128GB Framework Desktop configurations with AMD Ryzen AI Max+ 395, also known as Strix Halo? Donato Capitella maps the models, quantizations, context limits, and image/video workflows that make sense at each memory tier. The first half covers unified memory, GGUF model files, quantization, context and KV-cache memory, prompt processing, token generation, mixture-of-experts models, and speculative decoding. The second half gives current model recommendations for each memory tier, including Qwen3.8-27B, Qwen3.5-122B-A10B, DeepSeek V4 Flash 0731, and image and video workflows in ComfyUI. This video is a paid collaboration with Framework. The benchmark figures shown here are configuration-specific previews. Check the live benchmark site for the current results and methodology. Timestamps 00:00 - What fits in 32GB, 64GB, and 128GB? 03:06 - The unified-memory budget 04:30 - Models, files, and inference runtimes 12:29 - Context and the KV cache 13:43 - Performance metrics 22:09 - The 32GB configuration 25:59 - The 64GB configuration 27:57 - The 128GB configuration and DeepSeek V4 Flash 0731 31:22 - Image, video, and audio generation 34:13 - Conclusion and upcoming benchmarks Links & Resources - Framework Desktop: https://frame.work/gb/en/desktop - Strix Halo Toolboxes and tutorials: https://strix-halo-toolboxes.com - Strix Halo llama.cpp toolboxes: https://github.com/kyuz0/amd-strix-halo-toolboxes - Strix Halo ComfyUI toolboxes: https://github.com/kyuz0/amd-strix-halo-comfyui-toolboxes - Published local LLM benchmarks: https://local-llm-benchmarks.dev/ - Strix Halo Discord: https://discord.gg/pnPRyucNrG Text models and GGUF quants - Qwen3.8-27B, Unsloth Dynamic GGUF quants: https://huggingface.co/unsloth/Qwen3.8-27B-GGUF - Qwen3.6-35B-A3B, Unsloth Dynamic GGUF quants: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF - Gemma 4 26B-A4B, Unsloth Dynamic GGUF quants: https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF - Gemma 4 31B, Unsloth Dynamic GGUF quants: https://huggingface.co/unsloth/gemma-4-31B-it-GGUF - Muse Glimmer 30B, official model page: https://huggingface.co/meta-models/Muse-Glimmer-30B - Qwen3.5-122B-A10B, Unsloth GGUF quants: https://huggingface.co/unsloth/Qwen3.5-122B-A10B-GGUF - NVIDIA Nemotron 3 Super 120B-A12B, Unsloth GGUF quants: https://huggingface.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF - DeepSeek V4 Flash 0731, Unsloth Dynamic GGUF quants: https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF - Inkling-Small, Unsloth GGUF quants: https://huggingface.co/unsloth/Inkling-Small-GGUF
  • title: Local AI on the Framework Desktop: Best Models for 32GB, 64GB, and 128GB
  • short_form_content: false
  • id: 1575857
  • subtitle_filepaths: en/downloads/Framework/2026-09-02 Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB/Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB [mmntN7zIekU].en.srt
  • livestream: false
  • last_error:
  • media_downloaded_at: 2026-09-02T18:00:48Z
  • metadata_filepath: /downloads/Framework/2026-09-02 Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB/Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB [mmntN7zIekU].info.json
  • predicted_media_filepath: /downloads/Framework/2026-09-02 Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB/Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB [mmntN7zIekU].mp4
  • uploaded_at: 2026-09-02T17:59:35Z
  • matching_search_term:
  • updated_at: 2026-09-02T18:00:50Z
  • prevent_download: false
  • thumbnail_filepath: /downloads/Framework/2026-09-02 Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB/Local AI on the Framework Desktop： Best Models for 32GB, 64GB, and 128GB [mmntN7zIekU]-thumb.jpg
  • inserted_at: 2026-09-02T17:59:49Z
  • media_id: mmntN7zIekU
  • culled_at:
  • media_size_bytes: 266081313
  • original_url: https://www.youtube.com/watch?v=mmntN7zIekU
  • media_redownloaded_at:
  • source_id: 4
  • uuid: 85c1c9b6-a784-48f4-9acf-4ade9ce00a64
  • upload_date_index: 99
  • tasks:

Nothing Here!