Same GGUF. Inside vLLM. One Radeon. Paiton runs NEO CODER MAX on the R9700 with up to 10% lower warm request latency than our tested llama.cpp baseline. 128 input / 128 output tokens. One active…