News for hackers. Stuff that matters.
Re:NanoGPT by alfor on Wednesday January 11, 2023 @07:27AM (#34337636)
The cost of moving data from one gpu to the next will destroy performance.

The system are moving in the opposite direction (look at Dojo architecture or TensTorrent)

The silver lining is that the cost of training will fall substantially with those architecture that are not based in reusing gpu.

There are no comments on this comment yet.