Gemma4 31B with MTP enabled is faster and I feel a bit stronger at coding. Either one can run in 32GB VRAM or unified RAM with some tuning (3 bit weights, 8 bit

Gemma4 31B with MTP enabled is faster and I feel a bit stronger at coding. Either one can run in 32GB VRAM or unified RAM with some tuning (3 bit weights, 8 bit kv cache)