March 2026 monthly summary: cross-repo improvements addressing consumer Blackwell GPUs (SM 12.0) for Triton GEMM configuration and autotuning. In openxla/xla, a bug fix introduced SM 12.0-specific default configs (sm120.txtpb) and updated selection logic to use architecture-aware choices, eliminating invalid hints and reducing first-compilation overhead on RTX 5090-class devices. In Intel-tensorflow/tensorflow, a new autotuner feature adds SM 12.0 consumer configs to enable hint-based filtering and avoid brute-force search during Triton GEMM compilations. GetDefaultTritonConfigs was updated to distinguish between datacenter Blackwell (SM 10.0) and consumer Blackwell (SM 12.0+), with platform enum adjustments. Validation through unit and execution tests confirmed correct pathing and no regressions; the changes deliver measurable business value by speeding up GEMM workloads and improving GPU utilization on consumer Blackwell hardware.