9 comments

[ 4.4 ms ] story [ 32.1 ms ] thread
(comment deleted)
Looks like it'll be in the ball pack of 100 PetaFLOPS (~97.7 PFLOPS). Which should triple the Tianhe-2 (current #1) in computing power.
Power9 is an interesting choice vs. x86. Anyone know the reasoning?
It's better for massive parallelization, which is normally what you want for large scale data crunching. More threads per core than Xeon.
There are probably other reasons, but the NVLink interconnect is expected to show up on Power systems long before Intel, putting the GPUs much, much closer to system CPU and memory.

This thing is going to be a beast.

Let's hope IBM does a better job than they did on Blue Waters
I'm guessing ORNL isn't going to move the target as much as NCSA did for Blue Waters.
Not a Cray... That's gotta hurt. They've been a Cray shop for years now.