follow dev80_1, 1. 16×32×32×48统一格子MG套件+4min构建+1.42x调参; 2. Hierarchical VRAM→RAM + BatchedLocalSchur W10 + Cheap Jacobi; 3. V100 1.74s L1 vs 2.21s BiStabCG (1.27x), 8×8×8×16 1.42x (43 vs 94 iters); 4. 多卡 P100*2 PASS; [opencode].
follow dev80_1, 1. 16×32×32×48统一格子MG套件+4min构建+1.42x调参; 2. Hierarchical VRAM→RAM + BatchedLocalSchur W10 + Cheap Jacobi; 3. V100 1.74s L1 vs 2.21s BiStabCG (1.27x), 8×8×8×16 1.42x (43 vs 94 iters); 4. 多卡 P100*2 PASS; [opencode].