COLLECTION / RECIPES

Recipes

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

12 records

Apple M1 Max 64GB1 × 64 GBQwen3-Coder-30B-A3B-InstructQ4_K_Mollamareference2Kcontext72.3 tok/smeasuredcandidate ollama evidence measured

Apple M3 Ultra 512GB1 × 512 GBQwen3-Coder-30B-A3B-InstructQ4_K_Mllama.cppreference8Kcontext100 tok/smeasuredcandidate llama.cpp evidence measured

Apple M5 Max 128GB1 × 128 GBQwen3-Coder-30B-A3B-Instruct4bitmlxreferenceUnknown contextcontext140.7 tok/smeasuredcandidate mlx evidence measured

Apple M5 Pro 64GB1 × 64 GBQwen3-Coder-30B-A3B-Instruct4bitmlxreferenceUnknown contextcontext101 tok/smeasuredcandidate mlx evidence measured

GeForce RTX 50901 × 32 GBQwen3-Coder-30B-A3B-InstructQ6_Klmstudioreference2Kcontext263 tok/smeasuredcandidate lmstudio evidence measured

GeForce RTX 50901 × 32 GBQwen3-Coder-30B-A3B-InstructINT4vllmreferenceUnknown contextcontext287.7 tok/smeasuredcandidate vllm evidence measured

Intel Arc Pro B701 × 32 GBQwen3-Coder-30B-A3B-InstructGGUF UD-Q4_K_XLllama.cppreference4Kcontext108.1 tok/smeasuredcandidate llama.cpp evidence measured

Intel Arc Pro B702 × 32 GBQwen3-Coder-30B-A3B-InstructUD-Q4_K_XLllama.cppreference2Kcontext93.3 tok/smeasuredcandidate llama.cpp evidence measured

Radeon RX 7900 XTX1 × 24 GBQwen3-Coder-30B-A3B-InstructQ4_K_Mllama.cppreference1Kcontext46.9 tok/smeasuredcandidate llama.cpp evidence measured

Radeon RX 7900 XTX1 × 24 GBQwen3-Coder-30B-A3B-InstructQ4_K_Mllama.cppreference2Kcontext43.8 tok/smeasuredcandidate llama.cpp evidence measured

Ryzen AI Max+ 3951 × 128 GBQwen3-Coder-30B-A3B-InstructQ4_K_Mllama.cppreference4Kcontext70.5 tok/smeasuredcandidate llama.cpp evidence measured

Ryzen AI Max+ 3951 × 128 GBQwen3-Coder-30B-A3B-InstructQ5_K_Mllama.cppreference16Kcontext80.5 tok/smeasuredcandidate llama.cpp evidence measured