MI355X: FP4 vs FP8 Precision Comparison

FP4 posts 4294 tok/s/chip for $0.10 per million tokens at 34 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X); FP8 posts 20 tok/s/chip for $20.4. FP4 is 20936% cheaper per token; FP4 delivers 20939% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 67 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X): FP4 hits 1733 tok/s/chip, FP8 hits 20. Per-million costs land at $0.24 and $20.4 respectively. FP4 is 8391% cheaper per token; FP4 delivers 8391% more tok/s/chip. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Toward the upper edge of the 1–133 tok/s/user interactivity band, at 100 tok/s/user on DeepSeek V4 Pro 1.6T (MI355X): FP4 runs 851 tok/s/chip at $0.49/M tokens, FP8 runs 20 at $20.4/M. FP4 is 4052% cheaper per token; FP4 delivers 4071% more tok/s/chip. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks. (Numbers reflect the default 8k/1k selection for this URL — table and chart below update if you change sequence or model in the controls. Each side uses the best available serving configuration for that precision, which may include speculative decoding such as MTP where recipes exist — the same convention as the other comparison pages.)