MI355X
AMD's MI350-series AI GPU, the generation whose ROCm support first made a credible multi-thousand-GPU second source, and a clean illustration of precision-driven FLOPS inflation.
Current numbers
~40%MI355X (ROCm/SGLang FP8) cost-per-Mtoken advantage vs B200 (SGLang) at points on the GLM-5 single-node frontier — gap reversed on funded workloads
ROCm 7.xAMD stack with RCCL NCCL-API parity; MI350X/MI355X support (7.0 Sep 2025)
~10.1 PFLOPS / 288 GBAMD MI355X FP4 (no sparsity) and HBM3E capacity; 1,400 W peak board power