It’s possible (likely, even) to have a chip fast enough for inference, but not fast enough or with enough memory to do meaningful training runs. Like the curren

It’s possible (likely, even) to have a chip fast enough for inference, but not fast enough or with enough memory to do meaningful training runs. Like the current DGX spark.