MXFP8 improves throughput and peak active memory in a controlled B300 benchmark

The post reports that in a controlled benchmark on four NVIDIA B300 GPUs with uniform expert load, enabling MXFP8 in performance-critical parts increased end-to-end training throughput by ~21% over BF16 baseline and reduced peak active memory from 103 GiB to 95 GiB; most gains came from feed-forward computation and inter-expert data movement, not attention alone.

Kyle Wiggers·Published Oct 1, 2026

Behind the viewpoint

Translations are for reading; original excerpts remain the evidence.

Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs

Original excerpt

With MXFP8 enabled across the parts of the system where it helped most, training throughput was about 21% higher than with BF16, the higher-precision format we used as our baseline, while peak active memory fell from 103 GiB to 95 GiB. Most of the gain came from feed-forward computation and moving data between experts rather than attention alone.

Context

We measured MXFP8’s effect on end-to-end training throughput in a controlled benchmark on four NVIDIA B300 GPUs, with work distributed uniformly across experts.

Open original · Paragraph 21 View source page