Comment by neelm

Have you measured the difference in speed in moving data to the GPU? For enterprise AI workflows this is a bottleneck to utilization, so improved speed can help reduce compute costs.