MI355X: FP4 vs FP8 Precision Comparison

Near the low end of the 1–133 tok/s/user interactivity band, at 34 tok/s/user on DeepSeekv4 Pro 0813 1.6T (MI355X): FP4 runs 4294 tok/s/chip at $0.10/M tokens, FP8 runs 20 at $20.4/M. FP4 is 20939% cheaper per token; FP4 delivers 20939% more tok/s/chip. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 67 tok/s/user on DeepSeekv4 Pro 0813 1.6T (MI355X), FP4 delivers 1733 tok/s/chip at $0.24 per million tokens; FP8 delivers 20 tok/s/chip at $20.4. FP4 is 8391% cheaper per token; FP4 delivers 8391% more tok/s/chip. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 851 tok/s/chip for $0.49 per million tokens at 100 tok/s/user on DeepSeekv4 Pro 0813 1.6T (MI355X); FP8 posts 20 tok/s/chip for $20.4. FP4 is 4071% cheaper per token; FP4 delivers 4071% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab. (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.)