{"key":"local-fit","locale":"de","title":"Welches offene Modell zu Ihrer Hardware passt — 2026","url":"https://fedi.software/de/ai/infographics/local-fit","fields":[{"title":"Modell","key":"model","format":"link","visible":true,"unit":null},{"title":"Parameter","key":"params","format":"text","visible":true,"unit":null},{"title":"NVIDIA GeForce RTX 3060 12 GB","key":"hw:rtx-3060-12","format":"fit","visible":true,"unit":null},{"title":"NVIDIA GeForce RTX 4060 Ti 16 GB","key":"hw:rtx-4060-ti-16","format":"fit","visible":true,"unit":null},{"title":"NVIDIA GeForce RTX 3090","key":"hw:rtx-3090","format":"fit","visible":true,"unit":null},{"title":"NVIDIA GeForce RTX 4090","key":"hw:rtx-4090","format":"fit","visible":true,"unit":null},{"title":"NVIDIA GeForce RTX 5090","key":"hw:rtx-5090","format":"fit","visible":true,"unit":null},{"title":"NVIDIA RTX PRO 6000 Blackwell","key":"hw:rtx-pro-6000","format":"fit","visible":true,"unit":null},{"title":"2× NVIDIA GeForce RTX 3090","key":"hw:2x-rtx-3090","format":"fit","visible":true,"unit":null},{"title":"3× NVIDIA GeForce RTX 3090","key":"hw:3x-rtx-3090","format":"fit","visible":true,"unit":null},{"title":"4× NVIDIA GeForce RTX 3090","key":"hw:4x-rtx-3090","format":"fit","visible":true,"unit":null},{"title":"Apple M4 Max (16-core CPU, 40-core GPU)","key":"hw:apple-m4-max","format":"fit","visible":true,"unit":null},{"title":"Apple M3 Ultra","key":"hw:apple-m3-ultra","format":"fit","visible":true,"unit":null},{"title":"AMD Ryzen AI Max+ 395 (Strix Halo)","key":"hw:strix-halo","format":"fit","visible":true,"unit":null},{"title":"NVIDIA DGX Spark (GB10)","key":"hw:dgx-spark","format":"fit","visible":true,"unit":null},{"title":"AMD EPYC 7002/7003 (Rome/Milan), 8× DDR4-3200","key":"hw:epyc-ddr4-8ch","format":"fit","visible":true,"unit":null},{"title":"AMD EPYC 9004 (Genoa), 12× DDR5-4800","key":"hw:epyc-ddr5-12ch","format":"fit","visible":true,"unit":null}],"rows":[{"_id":"qwen25-3b-instruct","model":{"id":"qwen25-3b-instruct","text":"Qwen2.5 3B Instruct","href":"https://fedi.software/de/ai/compare?m=qwen25-3b-instruct","hf":"Qwen/Qwen2.5-3B-Instruct"},"params":"3.1B","hw:rtx-3060-12":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":84,"max":"F16"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":67,"max":"F32"},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":218,"max":"F32"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":235,"max":"F32"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":417,"max":"F32"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":417,"max":"F32"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":168,"max":"F32"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":136,"max":"F32"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.1,"tok_s":115,"max":"F32"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":5.1,"tok_s":111,"max":"F32"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":5.1,"tok_s":167,"max":"F32"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":5.1,"tok_s":52,"max":"F32"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":5.1,"tok_s":56,"max":"F32"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":5.1,"tok_s":15,"max":"F32"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":5.1,"tok_s":34,"max":"F32"}},{"_id":"llama-3-2-3b-instruct","model":{"id":"llama-3-2-3b-instruct","text":"Llama 3.2 3B Instruct","href":"https://fedi.software/de/ai/compare?m=llama-3-2-3b-instruct","hf":"meta-llama/Llama-3.2-3B-Instruct"},"params":"3.2B","hw:rtx-3060-12":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":74,"max":"BF16"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":59,"max":"BF16"},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":192,"max":"BF16"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":207,"max":"BF16"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":368,"max":"BF16"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":368,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":148,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":120,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":5.9,"tok_s":101,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":5.9,"tok_s":98,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":5.9,"tok_s":147,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":5.9,"tok_s":46,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":5.9,"tok_s":49,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":5.9,"tok_s":14.3,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":5.9,"tok_s":32,"max":"BF16"}},{"_id":"gemma-3-4b-it","model":{"id":"gemma-3-4b-it","text":"Gemma 3 4B","href":"https://fedi.software/de/ai/compare?m=gemma-3-4b-it","hf":"google/gemma-3-4b-it"},"params":"4.3B","hw:rtx-3060-12":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":67,"max":"BF16"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":54,"max":"BF16"},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":175,"max":"BF16"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":189,"max":"BF16"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":335,"max":"BF16"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":335,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":135,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":109,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":6.8,"tok_s":92,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":6.8,"tok_s":89,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":6.8,"tok_s":134,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":6.8,"tok_s":42,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":6.8,"tok_s":45,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":6.8,"tok_s":13.8,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":6.8,"tok_s":31,"max":"BF16"}},{"_id":"llama-31-8b-instruct","model":{"id":"llama-31-8b-instruct","text":"Llama 3.1 8B Instruct","href":"https://fedi.software/de/ai/compare?m=llama-31-8b-instruct","hf":"meta-llama/Llama-3.1-8B-Instruct"},"params":"8B","hw:rtx-3060-12":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":37},"hw:rtx-4060-ti-16":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":29,"max":"UD-Q8_K_XL"},"hw:rtx-3090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":95,"max":"BF16"},"hw:rtx-4090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":103,"max":"BF16"},"hw:rtx-5090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":182,"max":"BF16"},"hw:rtx-pro-6000":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":182,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":73,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":59,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":50,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":49,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":73,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":23,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":24,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":10.3,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"UD-Q6_K_XL","gb":9.9,"tok_s":23,"max":"BF16"}},{"_id":"qwen3-8b","model":{"id":"qwen3-8b","text":"Qwen3 8B","href":"https://fedi.software/de/ai/compare?m=qwen3-8b","hf":"Qwen/Qwen3-8B"},"params":"8.2B","hw:rtx-3060-12":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":31},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":25,"max":"UD-Q8_K_XL"},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":80,"max":"BF16"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":87,"max":"BF16"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":154,"max":"BF16"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":154,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":62,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":50,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":11.4,"tok_s":42,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":11.4,"tok_s":41,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":11.4,"tok_s":62,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":11.4,"tok_s":19.2,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":11.4,"tok_s":21,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"UD-Q6_K_XL","gb":10.2,"tok_s":10.2,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":11.4,"tok_s":21,"max":"BF16"}},{"_id":"gemma-3-12b-it","model":{"id":"gemma-3-12b-it","text":"Gemma 3 12B","href":"https://fedi.software/de/ai/compare?m=gemma-3-12b-it","hf":"google/gemma-3-12b-it"},"params":"12.2B","hw:rtx-3060-12":{"state":"vram","quant":"UD-Q5_K_XL","gb":11.7,"tok_s":32},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":17.8},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":58,"max":"UD-Q8_K_XL"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":62,"max":"UD-Q8_K_XL"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":111,"max":"BF16"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":111,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":44,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":36,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":15.7,"tok_s":30,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":15.7,"tok_s":30,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":15.7,"tok_s":44,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":15.7,"tok_s":13.8,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":15.7,"tok_s":14.8,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_1","gb":10.8,"tok_s":10.2,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":15.7,"tok_s":17.1,"max":"BF16"}},{"_id":"qwen3-14b","model":{"id":"qwen3-14b","text":"Qwen3 14B","href":"https://fedi.software/de/ai/compare?m=qwen3-14b","hf":"Qwen/Qwen3-14B"},"params":"14.8B","hw:rtx-3060-12":{"state":"vram","quant":"Q4_K_M","gb":11.8,"tok_s":30},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q6_K","gb":15,"tok_s":18},"hw:rtx-3090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":46,"max":"UD-Q8_K_XL"},"hw:rtx-4090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":49,"max":"UD-Q8_K_XL"},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":88,"max":"UD-Q8_K_XL"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":88,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":35,"max":"BF16"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":29,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":18.5,"tok_s":24,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":18.5,"tok_s":23,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":18.5,"tok_s":35,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":18.5,"tok_s":10.9,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":18.5,"tok_s":11.7,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"UD-Q3_K_XL","gb":10.4,"tok_s":10.1,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":18.5,"tok_s":14.6,"max":"BF16"}},{"_id":"gpt-oss-20b","model":{"id":"gpt-oss-20b","text":"gpt-oss-20b","href":"https://fedi.software/de/ai/compare?m=gpt-oss-20b","sub":"MoE","hf":"openai/gpt-oss-20b"},"params":"20.9B / 3.6B","hw:rtx-3060-12":{"state":"offload","quant":"MXFP4","gb":13.8,"tok_s":53},"hw:rtx-4060-ti-16":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":55},"hw:rtx-3090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":161.8,"measured":{"url":"https://github.com/ggml-org/llama.cpp/discussions/15396","date":"2025-08"}},"hw:rtx-4090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":194},"hw:rtx-5090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":344},"hw:rtx-pro-6000":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":344},"hw:2x-rtx-3090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":138},"hw:3x-rtx-3090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":112},"hw:4x-rtx-3090":{"state":"vram","quant":"MXFP4","gb":13.8,"tok_s":95},"hw:apple-m4-max":{"state":"unified","quant":"MXFP4","gb":13.8,"tok_s":80},"hw:apple-m3-ultra":{"state":"unified","quant":"MXFP4","gb":13.8,"tok_s":115.5,"measured":{"url":"https://github.com/ggml-org/llama.cpp/discussions/15396","date":"2025-08"}},"hw:strix-halo":{"state":"unified","quant":"MXFP4","gb":13.8,"tok_s":59},"hw:dgx-spark":{"state":"unified","quant":"MXFP4","gb":13.8,"tok_s":62},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"MXFP4","gb":13.8,"tok_s":17.2},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"MXFP4","gb":13.8,"tok_s":39}},{"_id":"mistral-small-3-2-24b-instruct","model":{"id":"mistral-small-3-2-24b-instruct","text":"Mistral Small 3.2 24B","href":"https://fedi.software/de/ai/compare?m=mistral-small-3-2-24b-instruct","hf":"mistralai/Mistral-Small-3.2-24B-Instruct-2506"},"params":"24B","hw:rtx-3060-12":{"state":"offload","quant":"Q3_K_M","gb":15.2,"tok_s":10.2,"max":"UD-IQ2_M"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"UD-Q3_K_XL","gb":15.6,"tok_s":18.4},"hw:rtx-3090":{"state":"vram","quant":"Q6_K","gb":23.1,"tok_s":37},"hw:rtx-4090":{"state":"vram","quant":"Q6_K","gb":23.1,"tok_s":40},"hw:rtx-5090":{"state":"vram","quant":"Q8_0","gb":28.8,"tok_s":56},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":28.8,"tok_s":56,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":28.8,"tok_s":22,"max":"UD-Q8_K_XL"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":28.8,"tok_s":18.2,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":28.8,"tok_s":15.3,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":28.8,"tok_s":14.9,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":28.8,"tok_s":22,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q5_K_M","gb":20.5,"tok_s":10.3,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q5_K_M","gb":20.5,"tok_s":11,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":18.1,"tok_s":6.9,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":28.8,"tok_s":10.3,"max":"BF16"}},{"_id":"gemma-4-26b-a4b-it","model":{"id":"gemma-4-26b-a4b-it","text":"Gemma 4 26B A4B ","href":"https://fedi.software/de/ai/compare?m=gemma-4-26b-a4b-it","sub":"MoE","hf":"google/gemma-4-26B-A4B-it"},"params":"25.8B / 3.8B","hw:rtx-3060-12":{"state":"offload","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":14.7,"max":"BF16"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"UD-Q3_K_M","gb":16,"tok_s":56},"hw:rtx-3090":{"state":"vram","quant":"UD-Q5_K_S","gb":22.1,"tok_s":129},"hw:rtx-4090":{"state":"vram","quant":"UD-Q5_K_S","gb":22.1,"tok_s":139},"hw:rtx-5090":{"state":"vram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":173},"hw:rtx-pro-6000":{"state":"vram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":173,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":70},"hw:3x-rtx-3090":{"state":"vram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":57,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":48,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":40,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":60,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":29,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":31,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":13.7,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"UD-Q8_K_XL","gb":30.9,"tok_s":31,"max":"BF16"}},{"_id":"gemma-3-27b-it","model":{"id":"gemma-3-27b-it","text":"Gemma 3 27B","href":"https://fedi.software/de/ai/compare?m=gemma-3-27b-it","hf":"google/gemma-3-27b-it"},"params":"27.4B","hw:rtx-3060-12":{"state":"offload","quant":"UD-IQ3_XXS","gb":14.3,"tok_s":12.7,"max":"UD-IQ2_XXS"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q3_K_S","gb":15.6,"tok_s":18.1},"hw:rtx-3090":{"state":"vram","quant":"UD-Q5_K_XL","gb":22.8,"tok_s":38},"hw:rtx-4090":{"state":"vram","quant":"UD-Q5_K_XL","gb":22.8,"tok_s":41},"hw:rtx-5090":{"state":"vram","quant":"UD-Q6_K_XL","gb":27.2,"tok_s":59},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":32.2,"tok_s":49,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":32.2,"tok_s":19.7,"max":"UD-Q8_K_XL"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":32.2,"tok_s":16,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":32.2,"tok_s":13.5,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":32.2,"tok_s":13.1,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":32.2,"tok_s":19.6,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q4_1","gb":20.6,"tok_s":10.1,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q4_1","gb":20.6,"tok_s":10.8,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":20,"tok_s":6.3,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"UD-Q6_K_XL","gb":27.2,"tok_s":10.8,"max":"BF16"}},{"_id":"qwen3-30b-a3b","model":{"id":"qwen3-30b-a3b","text":"Qwen3 30B A3B","href":"https://fedi.software/de/ai/compare?m=qwen3-30b-a3b","sub":"MoE","hf":"Qwen/Qwen3-30B-A3B"},"params":"30.5B / 3.3B","hw:rtx-3060-12":{"state":"offload","quant":"Q8_0","gb":34.8,"tok_s":16.4,"max":"UD-IQ1_M"},"hw:rtx-4060-ti-16":{"state":"vram","quant":"Q3_K_S","gb":15.6,"tok_s":66},"hw:rtx-3090":{"state":"vram","quant":"Q4_K_M","gb":20.9,"tok_s":153.6,"measured":{"url":"https://www.hardware-corner.net/gpu-llm-benchmarks/rtx-3090/","date":"2026-03"}},"hw:rtx-4090":{"state":"vram","quant":"Q5_K_S","gb":23.4,"tok_s":158},"hw:rtx-5090":{"state":"vram","quant":"UD-Q6_K_XL","gb":28.6,"tok_s":232},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":34.8,"tok_s":192,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":34.8,"tok_s":77,"max":"UD-Q8_K_XL"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":34.8,"tok_s":63,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":34.8,"tok_s":53,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":34.8,"tok_s":45,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":34.8,"tok_s":67,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":34.8,"tok_s":33,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":34.8,"tok_s":35,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":34.8,"tok_s":14.3,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":34.8,"tok_s":32,"max":"BF16"}},{"_id":"gemma-4-31b-it","model":{"id":"gemma-4-31b-it","text":"Gemma 4 31B","href":"https://fedi.software/de/ai/compare?m=gemma-4-31b-it","hf":"google/gemma-4-31B-it"},"params":"31.3B","hw:rtx-3060-12":{"state":"offload","quant":"Q4_K_M","gb":23.2,"tok_s":3.8,"max":"BF16"},"hw:rtx-4060-ti-16":{"state":"offload","quant":"Q3_K_S","gb":18.1,"tok_s":10.4,"max":"UD-IQ2_M"},"hw:rtx-3090":{"state":"vram","quant":"Q4_1","gb":24,"tok_s":37},"hw:rtx-4090":{"state":"vram","quant":"Q4_1","gb":24,"tok_s":40},"hw:rtx-5090":{"state":"vram","quant":"Q6_K","gb":30.1,"tok_s":55},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":37.5,"tok_s":43,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":37.5,"tok_s":17.1,"max":"UD-Q8_K_XL"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":37.5,"tok_s":13.9,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":37.5,"tok_s":11.7,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":37.5,"tok_s":11.3,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":37.5,"tok_s":17,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"IQ4_XS","gb":21.3,"tok_s":10.3,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q4_K_S","gb":22.3,"tok_s":10.3,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":23.2,"tok_s":5.7,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q6_K","gb":30.1,"tok_s":10.1,"max":"BF16"}},{"_id":"qwen3-32b","model":{"id":"qwen3-32b","text":"Qwen3 32B","href":"https://fedi.software/de/ai/compare?m=qwen3-32b","hf":"Qwen/Qwen3-32B"},"params":"32.8B","hw:rtx-3060-12":{"state":"offload","quant":"Q4_K_M","gb":23.4,"tok_s":3.7,"max":"UD-IQ1_M"},"hw:rtx-4060-ti-16":{"state":"offload","quant":"UD-IQ3_XXS","gb":16.6,"tok_s":14.1,"max":"Q2_K"},"hw:rtx-3090":{"state":"vram","quant":"UD-Q4_K_XL","gb":23.7,"tok_s":35},"hw:rtx-4090":{"state":"vram","quant":"UD-Q4_K_XL","gb":23.7,"tok_s":38},"hw:rtx-5090":{"state":"vram","quant":"Q6_K","gb":30.5,"tok_s":51},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":38.5,"tok_s":40,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":38.5,"tok_s":16,"max":"UD-Q8_K_XL"},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":38.5,"tok_s":13,"max":"BF16"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":38.5,"tok_s":11,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":38.5,"tok_s":10.6,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":38.5,"tok_s":16,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"UD-Q3_K_XL","gb":20.1,"tok_s":10.3,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"IQ4_XS","gb":21.4,"tok_s":10.2,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":23.4,"tok_s":5.4,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q5_K_M","gb":26.9,"tok_s":10.8,"max":"BF16"}},{"_id":"qwen3-5-35b-a3b","model":{"id":"qwen3-5-35b-a3b","text":"Qwen3.5 35B A3B","href":"https://fedi.software/de/ai/compare?m=qwen3-5-35b-a3b","sub":"MoE","hf":"Qwen/Qwen3.5-35B-A3B"},"params":"36B / 3B","hw:rtx-3060-12":{"state":"offload","quant":"Q8_0","gb":40,"tok_s":17.9,"max":"UD-Q8_K_XL"},"hw:rtx-4060-ti-16":{"state":"offload","quant":"Q8_0","gb":40,"tok_s":18.7,"max":"UD-Q2_K_XL"},"hw:rtx-3090":{"state":"vram","quant":"Q4_K_S","gb":23.7,"tok_s":191},"hw:rtx-4090":{"state":"vram","quant":"Q4_K_S","gb":23.7,"tok_s":205},"hw:rtx-5090":{"state":"vram","quant":"Q6_K","gb":31.9,"tok_s":274},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":40,"tok_s":220,"max":"BF16"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":40,"tok_s":89},"hw:3x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":40,"tok_s":72,"max":"UD-Q8_K_XL"},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":40,"tok_s":61,"max":"BF16"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":40,"tok_s":51,"max":"BF16"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":40,"tok_s":77,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":40,"tok_s":37,"max":"BF16"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":40,"tok_s":40,"max":"BF16"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":40,"tok_s":15,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":40,"tok_s":34,"max":"BF16"}},{"_id":"llama-3-3-70b-instruct","model":{"id":"llama-3-3-70b-instruct","text":"Llama 3.3 70B Instruct","href":"https://fedi.software/de/ai/compare?m=llama-3-3-70b-instruct","hf":"meta-llama/Llama-3.3-70B-Instruct"},"params":"70.6B","hw:rtx-3060-12":{"state":"offload","quant":"Q4_K_M","gb":46.7,"tok_s":1.3,"max":"UD-Q6_K_XL"},"hw:rtx-4060-ti-16":{"state":"offload","quant":"Q4_K_M","gb":46.7,"tok_s":1.4,"max":"UD-Q6_K_XL"},"hw:rtx-3090":{"state":"offload","quant":"Q4_K_M","gb":46.7,"tok_s":2,"max":"UD-IQ2_XXS"},"hw:rtx-4090":{"state":"offload","quant":"Q4_K_M","gb":46.7,"tok_s":2,"max":"UD-IQ2_XXS"},"hw:rtx-5090":{"state":"vram","quant":"UD-IQ3_XXS","gb":31.8,"tok_s":49},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":79.2,"tok_s":18.8,"max":"UD-Q8_K_XL"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q4_K_M","gb":46.7,"tok_s":16.3,"measured":{"url":"https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference","date":"2024-05"}},"hw:3x-rtx-3090":{"state":"vram","quant":"Q4_1","gb":48.5,"tok_s":10.3,"max":"UD-Q6_K_XL"},"hw:4x-rtx-3090":{"state":"vram","quant":"IQ4_XS","gb":42.1,"tok_s":10,"max":"UD-Q8_K_XL"},"hw:apple-m4-max":{"state":"unified","quant":"UD-Q3_K_XL","gb":39,"tok_s":10.6,"max":"UD-Q8_K_XL"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q5_K_M","gb":54.1,"tok_s":11.2,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q4_K_M","gb":46.7,"tok_s":4.1,"max":"UD-Q8_K_XL"},"hw:dgx-spark":{"state":"unified","quant":"Q4_K_M","gb":46.7,"tok_s":4.4,"max":"UD-Q8_K_XL"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":46.7,"tok_s":2.9,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q4_K_M","gb":46.7,"tok_s":6.6,"max":"BF16"}},{"_id":"qwen3-next-80b-a3b-instruct","model":{"id":"qwen3-next-80b-a3b-instruct","text":"Qwen3 Next 80B A3B Instruct","href":"https://fedi.software/de/ai/compare?m=qwen3-next-80b-a3b-instruct","sub":"MoE","hf":"Qwen/Qwen3-Next-80B-A3B-Instruct"},"params":"81.3B / 3B","hw:rtx-3060-12":{"state":"offload","quant":"Q5_K_M","gb":59.1,"tok_s":24},"hw:rtx-4060-ti-16":{"state":"offload","quant":"Q6_K","gb":67.8,"tok_s":21},"hw:rtx-3090":{"state":"offload","quant":"UD-Q6_K_XL","gb":70.8,"tok_s":26,"max":"UD-TQ1_0"},"hw:rtx-4090":{"state":"offload","quant":"UD-Q6_K_XL","gb":70.8,"tok_s":26,"max":"UD-TQ1_0"},"hw:rtx-5090":{"state":"offload","quant":"UD-Q6_K_XL","gb":70.8,"tok_s":31,"max":"Q2_K_L"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q8_0","gb":87.1,"tok_s":213,"max":"UD-Q8_K_XL"},"hw:2x-rtx-3090":{"state":"vram","quant":"Q4_K_S","gb":47.8,"tok_s":145},"hw:3x-rtx-3090":{"state":"vram","quant":"UD-Q6_K_XL","gb":70.8,"tok_s":84},"hw:4x-rtx-3090":{"state":"vram","quant":"Q8_0","gb":87.1,"tok_s":59,"max":"UD-Q8_K_XL"},"hw:apple-m4-max":{"state":"unified","quant":"Q8_0","gb":87.1,"tok_s":49,"max":"UD-Q8_K_XL"},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":87.1,"tok_s":74,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q8_0","gb":87.1,"tok_s":36,"max":"UD-Q8_K_XL"},"hw:dgx-spark":{"state":"unified","quant":"Q8_0","gb":87.1,"tok_s":39,"max":"UD-Q8_K_XL"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q8_0","gb":87.1,"tok_s":14.8,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":87.1,"tok_s":33,"max":"BF16"}},{"_id":"glm-4-5-air","model":{"id":"glm-4-5-air","text":"GLM 4.5 Air","href":"https://fedi.software/de/ai/compare?m=glm-4-5-air","sub":"MoE","hf":"zai-org/GLM-4.5-Air"},"params":"110B / 17B","hw:rtx-3060-12":{"state":"offload","quant":"Q4_0","gb":65.8,"tok_s":5.3},"hw:rtx-4060-ti-16":{"state":"offload","quant":"Q4_K_S","gb":70.1,"tok_s":5.1},"hw:rtx-3090":{"state":"offload","quant":"Q4_K_M","gb":76,"tok_s":5.6},"hw:rtx-4090":{"state":"offload","quant":"Q4_K_M","gb":76,"tok_s":5.7},"hw:rtx-5090":{"state":"offload","quant":"Q3_K_M","gb":60.2,"tok_s":10.2,"max":"UD-Q5_K_XL"},"hw:rtx-pro-6000":{"state":"vram","quant":"Q5_K_M","gb":86.6,"tok_s":55},"hw:2x-rtx-3090":{"state":"offload","quant":"Q4_0","gb":65.8,"tok_s":10,"max":"UD-IQ2_M"},"hw:3x-rtx-3090":{"state":"vram","quant":"UD-Q4_K_XL","gb":70.8,"tok_s":22},"hw:4x-rtx-3090":{"state":"vram","quant":"Q5_K_M","gb":86.6,"tok_s":15.2},"hw:apple-m4-max":{"state":"unified","quant":"Q5_K_M","gb":86.6,"tok_s":12.8},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":120.5,"tok_s":13.9,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q4_K_M","gb":76,"tok_s":10.6,"max":"UD-Q6_K_XL"},"hw:dgx-spark":{"state":"unified","quant":"Q5_K_M","gb":86.6,"tok_s":10,"max":"UD-Q6_K_XL"},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":76,"tok_s":8,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":120.5,"tok_s":13.1,"max":"BF16"}},{"_id":"openai-gpt-oss-120b","model":{"id":"openai-gpt-oss-120b","text":"gpt-oss-120b","href":"https://fedi.software/de/ai/compare?m=openai-gpt-oss-120b","sub":"MoE","hf":"openai/gpt-oss-120b"},"params":"117B / 5.1B","hw:rtx-3060-12":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":19.1},"hw:rtx-4060-ti-16":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":19.6},"hw:rtx-3090":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":[26.2,27.5],"measured":{"url":"https://github.com/ggml-org/llama.cpp/discussions/15396","date":"2026-01-20"}},"hw:rtx-4090":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":25},"hw:rtx-5090":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":31},"hw:rtx-pro-6000":{"state":"vram","quant":"MXFP4","gb":65.2,"tok_s":258},"hw:2x-rtx-3090":{"state":"offload","quant":"MXFP4","gb":65.2,"tok_s":36},"hw:3x-rtx-3090":{"state":"vram","quant":"MXFP4","gb":65.2,"tok_s":[41.1,73.3],"measured":{"url":"https://hardware-corner.net/3x-rtx-3090-gpt-oss-120b-test/","date":"2025-10-19"}},"hw:4x-rtx-3090":{"state":"vram","quant":"MXFP4","gb":65.2,"tok_s":71},"hw:apple-m4-max":{"state":"unified","quant":"MXFP4","gb":65.2,"tok_s":60},"hw:apple-m3-ultra":{"state":"unified","quant":"MXFP4","gb":65.2,"tok_s":90},"hw:strix-halo":{"state":"unified","quant":"MXFP4","gb":65.2,"tok_s":50.2,"measured":{"url":"https://quozul.dev/posts/2025-10-27-gpt-oss-120b-on-ryzen-ai-max-395/","date":"2025-10-27"}},"hw:dgx-spark":{"state":"unified","quant":"MXFP4","gb":65.2,"tok_s":60.6,"measured":{"url":"https://github.com/ggml-org/llama.cpp/discussions/16578","date":"2025-11-02"}},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"MXFP4","gb":65.2,"tok_s":15.8},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"MXFP4","gb":65.2,"tok_s":36}},{"_id":"qwen3-5-122b-a10b","model":{"id":"qwen3-5-122b-a10b","text":"Qwen3.5 122B A10B","href":"https://fedi.software/de/ai/compare?m=qwen3-5-122b-a10b","sub":"MoE","hf":"Qwen/Qwen3.5-122B-A10B"},"params":"125B / 10B","hw:rtx-3060-12":{"state":"offload","quant":"UD-IQ4_NL","gb":64.7,"tok_s":10.5},"hw:rtx-4060-ti-16":{"state":"offload","quant":"UD-IQ4_NL","gb":64.7,"tok_s":10.8},"hw:rtx-3090":{"state":"offload","quant":"Q4_K_S","gb":74.9,"tok_s":11.1},"hw:rtx-4090":{"state":"offload","quant":"Q4_K_S","gb":74.9,"tok_s":11.2},"hw:rtx-5090":{"state":"offload","quant":"UD-Q4_K_XL","gb":80.2,"tok_s":11.9},"hw:rtx-pro-6000":{"state":"vram","quant":"UD-Q5_K_XL","gb":95.1,"tok_s":97},"hw:2x-rtx-3090":{"state":"vram","quant":"UD-IQ3_XXS","gb":48,"tok_s":76},"hw:3x-rtx-3090":{"state":"vram","quant":"UD-IQ4_NL","gb":64.7,"tok_s":46},"hw:4x-rtx-3090":{"state":"vram","quant":"UD-Q5_K_XL","gb":95.1,"tok_s":27},"hw:apple-m4-max":{"state":"unified","quant":"UD-Q5_K_XL","gb":95.1,"tok_s":23},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":133.1,"tok_s":24,"max":"BF16"},"hw:strix-halo":{"state":"unified","quant":"Q6_K","gb":104.2,"tok_s":15.1},"hw:dgx-spark":{"state":"unified","quant":"Q6_K","gb":104.2,"tok_s":16.1},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"UD-Q5_K_XL","gb":95.1,"tok_s":10.4,"max":"BF16"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":133.1,"tok_s":19.3,"max":"BF16"}},{"_id":"qwen3-235b-a22b","model":{"id":"qwen3-235b-a22b","text":"Qwen3 235B A22B","href":"https://fedi.software/de/ai/compare?m=qwen3-235b-a22b","sub":"MoE","hf":"Qwen/Qwen3-235B-A22B"},"params":"235B / 22B","hw:rtx-3060-12":{"state":"no","gb":104.5},"hw:rtx-4060-ti-16":{"state":"no","gb":104.5},"hw:rtx-3090":{"state":"no","gb":104.5},"hw:rtx-4090":{"state":"no","gb":104.5},"hw:rtx-5090":{"state":"no","gb":104.5},"hw:rtx-pro-6000":{"state":"offload","quant":"UD-Q4_K_XL","gb":137.2,"tok_s":10.8,"max":"UD-Q2_K_XL"},"hw:2x-rtx-3090":{"state":"offload","quant":"UD-Q2_K_XL","gb":91.1,"tok_s":8,"low":true},"hw:3x-rtx-3090":{"state":"offload","quant":"Q3_K_M","gb":115.5,"tok_s":6.1},"hw:4x-rtx-3090":{"state":"offload","quant":"UD-Q3_K_XL","gb":106.8,"tok_s":11.7,"max":"UD-Q2_K_XL"},"hw:apple-m4-max":{"state":"unified","quant":"UD-Q2_K_XL","gb":91.1,"tok_s":19.4,"low":true},"hw:apple-m3-ultra":{"state":"unified","quant":"Q8_0","gb":253,"tok_s":10.8,"max":"UD-Q8_K_XL"},"hw:strix-halo":{"state":"unified","quant":"UD-Q3_K_XL","gb":106.8,"tok_s":12.2},"hw:dgx-spark":{"state":"unified","quant":"UD-Q3_K_XL","gb":106.8,"tok_s":13},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":145.2,"tok_s":7.2,"max":"UD-Q8_K_XL"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q8_0","gb":253,"tok_s":10.9,"max":"UD-Q8_K_XL"}},{"_id":"glm-4-7","model":{"id":"glm-4-7","text":"GLM 4.7","href":"https://fedi.software/de/ai/compare?m=glm-4-7","sub":"MoE","hf":"zai-org/GLM-4.7"},"params":"358B / 39.2B","hw:rtx-3060-12":{"state":"no","gb":149.7},"hw:rtx-4060-ti-16":{"state":"no","gb":149.7},"hw:rtx-3090":{"state":"no","gb":149.7},"hw:rtx-4090":{"state":"no","gb":149.7},"hw:rtx-5090":{"state":"no","gb":149.7},"hw:rtx-pro-6000":{"state":"offload","quant":"UD-IQ3_XXS","gb":149.7,"tok_s":7.3,"max":"UD-TQ1_0"},"hw:2x-rtx-3090":{"state":"offload","quant":"UD-IQ1_S","gb":101.8,"tok_s":5.6,"low":true},"hw:3x-rtx-3090":{"state":"offload","quant":"UD-IQ2_XXS","gb":120.5,"tok_s":4.7,"low":true},"hw:4x-rtx-3090":{"state":"offload","quant":"UD-IQ3_XXS","gb":149.7,"tok_s":3.5,"max":"UD-TQ1_0"},"hw:apple-m4-max":{"state":"unified","quant":"UD-TQ1_0","gb":89.1,"tok_s":16.2,"low":true},"hw:apple-m3-ultra":{"state":"unified","quant":"Q4_1","gb":229.2,"tok_s":10,"max":"UD-Q6_K_XL"},"hw:strix-halo":{"state":"unified","quant":"UD-IQ1_M","gb":112.2,"tok_s":9.6,"low":true},"hw:dgx-spark":{"state":"unified","quant":"UD-IQ1_M","gb":112.2,"tok_s":10.3,"low":true},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":221,"tok_s":4.7,"max":"UD-Q8_K_XL"},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q4_1","gb":229.2,"tok_s":10.2,"max":"UD-Q6_K_XL"}},{"_id":"deepseek-r1","model":{"id":"deepseek-r1","text":"DeepSeek R1","href":"https://fedi.software/de/ai/compare?m=deepseek-r1","sub":"MoE","hf":"deepseek-ai/DeepSeek-R1"},"params":"684B / 37B","hw:rtx-3060-12":{"state":"no","gb":321.3},"hw:rtx-4060-ti-16":{"state":"no","gb":321.3},"hw:rtx-3090":{"state":"no","gb":321.3},"hw:rtx-4090":{"state":"no","gb":321.3},"hw:rtx-5090":{"state":"no","gb":321.3},"hw:rtx-pro-6000":{"state":"offload","quant":"UD-IQ1_S","gb":142.3,"tok_s":16.9,"low":true},"hw:2x-rtx-3090":{"state":"no","gb":321.3},"hw:3x-rtx-3090":{"state":"no","gb":321.3},"hw:4x-rtx-3090":{"state":"offload","quant":"UD-IQ1_S","gb":142.3,"tok_s":8,"low":true},"hw:apple-m4-max":{"state":"no","gb":321.3},"hw:apple-m3-ultra":{"state":"unified","quant":"Q3_K_M","gb":321.3,"tok_s":14.9},"hw:strix-halo":{"state":"no","gb":321.3},"hw:dgx-spark":{"state":"no","gb":321.3},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q4_K_M","gb":406.5,"tok_s":5.2},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"Q3_K_M","gb":321.3,"tok_s":13.9}},{"_id":"kimi-k2-5","model":{"id":"kimi-k2-5","text":"Kimi K2.5","href":"https://fedi.software/de/ai/compare?m=kimi-k2-5","sub":"MoE","hf":"moonshotai/Kimi-K2.5"},"params":"1,027B / 32B","hw:rtx-3060-12":{"state":"no","gb":416.9},"hw:rtx-4060-ti-16":{"state":"no","gb":416.9},"hw:rtx-3090":{"state":"no","gb":416.9},"hw:rtx-4090":{"state":"no","gb":416.9},"hw:rtx-5090":{"state":"no","gb":416.9},"hw:rtx-pro-6000":{"state":"no","gb":416.9},"hw:2x-rtx-3090":{"state":"no","gb":416.9},"hw:3x-rtx-3090":{"state":"no","gb":416.9},"hw:4x-rtx-3090":{"state":"no","gb":416.9},"hw:apple-m4-max":{"state":"no","gb":416.9},"hw:apple-m3-ultra":{"state":"unified","quant":"UD-Q2_K_XL","gb":376.8,"tok_s":22,"low":true},"hw:strix-halo":{"state":"no","gb":416.9},"hw:dgx-spark":{"state":"no","gb":416.9},"hw:epyc-ddr4-8ch":{"state":"ram","quant":"Q3_K_S","gb":445.2,"tok_s":7.2},"hw:epyc-ddr5-12ch":{"state":"ram","quant":"UD-IQ2_XXS","gb":328.8,"tok_s":19.7,"low":true}}],"meta":{"sources":[{"name":"Hugging Face","url":"https://huggingface.co/models","license":null,"license_url":null,"modified":false},{"name":"www.nvidia.com","url":"https://www.nvidia.com/en-us/geforce/news/game-on-you-asked-we-answered-qa/","license":null,"license_url":null,"modified":false},{"name":"images.nvidia.com","url":"https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf","license":null,"license_url":null,"modified":false},{"name":"support.apple.com","url":"https://support.apple.com/en-us/121553","license":null,"license_url":null,"modified":false},{"name":"www.amd.com","url":"https://www.amd.com/en/products/processors/laptop/ryzen/ai-300-series/amd-ryzen-ai-max-plus-395.html","license":null,"license_url":null,"modified":false}],"updated":"2026-10-04","notes":["≈ Schätzungen für einen Kontext von 8K: Gewichte + KV-Cache + Laufzeit-Overhead im Verhältnis zum Speicher; Geschwindigkeit aus der Speicherbandbreite. Die realen Werte hängen von Laufzeit und Einstellungen ab."],"license":{"name":"fedi.software","license":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/","scope":"fedi.software data: hardware, builds, glossary, Fit calculations, curation and the compilation; third-party data keeps the licence of its source (sources[])"},"attribution":["Data by fedi.software, CC BY 4.0 — a link to the source page is required.","Model facts: Hugging Face Hub (parameters, architecture, quant files); every model under its own licence."]}}