AMD vs. Nvidia Inference Benchmark: Performance per Cost / Million Tokens