ROCm
AMD's open GPU-computing software stack, its counterpart to NVIDIA's CUDA ecosystem.
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
~15–30%lower hardware cost for AMD vs NVIDIA — the prize that funds the ROCm tax
37–66%MI300X realized inference performance vs H100/H200 despite ~1.3x paper FLOPS — the ROCm goodput gap