Comment by supercharger9

Ignoring latency but not throughput, How does this compare in terms of Cost ( cards Acquisition cost and Power needed) with Nvidia GPU for inference?