COLLECTION / RECIPES

Recipes

One artifact × hardware × engine unit. Validated rows are launch-safe; measured speed sweeps attach to each recipe as evidence.

2,251 records

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1BQ4_K_Mllama.cppreference124.9Kcontext204.9 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1BQ8llama.cppreference124.9Kcontext175.8 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1BQ4_K_Mllama.cppreference124.9Kcontext210.7 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1BQ8_0llama.cppreference124.9Kcontext191.2 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1BQ4_K_Mllama.cppreference124.9Kcontext191.2 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1B4bitmlxreference124.9Kcontext178.7 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBLFM2.5-8B-A1B8bitmlxreference121.1Kcontext153.7 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBllama-3.1-70bQ4_K_Mllama.cppreference127.9Kcontext13.5 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBllama-3.1-70b-instructQ4_K_Mllama.cppreference127.9Kcontext13.5 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBllama-3.1-nemotron-70b-q4kmQ4_K_Mllama.cppreference127.9Kcontext13.5 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBllama-3.3-70b-instruct-q4kmQ4_K_Mllama.cppreference127.9Kcontext13.5 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashiq1mllama.cppreference255.9Kcontext50.8 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashiq1sllama.cppreference255.9Kcontext50.8 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashiq2mllama.cppreference255.9Kcontext51.9 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashiq2xxsllama.cppreference255.9Kcontext52.1 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashUD-IQ1_Mllama.cppreference255.9Kcontext52.8 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashUD-IQ1_Mllama.cppreference255.9Kcontext51.5 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashUD-IQ1_Sllama.cppreference255.9Kcontext50.4 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashUD-IQ2_Mllama.cppreference255.9Kcontext52.6 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBmimo-v2-flashUD-IQ2_XXSllama.cppreference255.9Kcontext52.8 tok/smeasuredcandidate llama.cpp evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBnemotron-3-nano-30b-a3b8bitmlxreference252.1Kcontext95.2 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBnemotron-3-nano-30b-a3bnvfp4mlxreference252.1Kcontext103.4 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 96GB 80-core GPU1 × 96 GBnex-n2-miniQ4_K_Mllama.cppreference255.9Kcontext94 tok/smeasuredcandidate llama.cpp evidence measured