MI355X: FP4 vs INT4 Precision Comparison

Throughput at 39 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (MI355X): FP4 hits 1064 tok/s/GPU, INT4 hits 496. Per-million costs land at $0.39 and $0.82 respectively. FP4 is 109% cheaper per token; FP4 delivers 115% more tok/s/GPU. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Around the middle of the 24–85 tok/s/user interactivity band, at 54 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (MI355X): FP4 runs 632 tok/s/GPU at $0.66/M tokens, INT4 runs 334 at $1.22/M. FP4 is 85% cheaper per token; FP4 delivers 89% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 70 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (MI355X), FP4 delivers 541 tok/s/GPU at $0.76 per million tokens; INT4 delivers 230 tok/s/GPU at $1.80. FP4 is 137% cheaper per token; FP4 delivers 135% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact. (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.)