Beating the InferenceX MI325X Reference: 60% More DeepSeek Throughput
ProsGrow measured 364.6 output tokens per second per GPU on two AMD MI325X GPUs—60% above the published two-GPU, 32-session reference for DeepSeek V4.1 Flash.
Read post : Beating the InferenceX MI325X Reference: 60% More DeepSeek Throughput
