Nvidia has demonstrated that its NVFP4 4-bit floating point format — originally intended for inference — can be used for stable, large-scale training of LLMs with minimal accuracy loss, offering significant gains in compute and memory efficiency over both FP8 and BF16.

Qualcomm’s legal victory over Arm not only secures its right to deploy Nuvia’s custom CPU cores under its existing license but also sets a major industry precedent, weakening Arm’s control over licensing as it pivots toward designing its own CPUs.