MI355X: FP4 vs FP8 Precision Comparison

Near the low end of the 12–66 tok/s/user interactivity band, at 25 tok/s/user on GLM 5/5.1 (MI355X): FP4 runs 1169 tok/s/GPU at $0.35/M tokens, FP8 runs 987 at $0.42/M. FP4 is 21% cheaper per token; FP4 delivers 18% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 39 tok/s/user on GLM 5/5.1 (MI355X), FP4 delivers 525 tok/s/GPU at $0.78 per million tokens; FP8 delivers 809 tok/s/GPU at $0.51. FP8 is 53% cheaper per token; FP8 delivers 54% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 255 tok/s/GPU for $1.62 per million tokens at 53 tok/s/user on GLM 5/5.1 (MI355X); FP8 posts 447 tok/s/GPU for $0.92. FP8 is 76% cheaper per token; FP8 delivers 76% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab. (Numbers reflect the default 1k/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.)