4 ms·

Microscaling formats for machine learning: https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf https://www.opencompute.org/doc

• ipsum2 2y ago
  This is a big deal for increasing performance past fp8. We'll see Nvidia/AMD/Intel etc adopt this in their products in the next year.