The TensorFloat-32 results look really impressive, but yikes that is not a good name. "TensorFloat" is extremely confusable with "TensorFlow," and it would really more accurately be called a 19 bit format.
I wouldn't worry about it. Looks like the Anandtech article [1] doesn't either :)
> bfloat16, a format popularized by Intel
[1] https://www.anandtech.com/show/15801/nvidia-announces-ampere...