ROCm vs Vulkan vs vLLM benchmarks on dual R9700s

User reports vLLM significantly outperforms ROCm and Vulkan for running Qwen models on dual Radeon R9700 GPUs. Benchmarks show up to 2x speed improvement over llama.cpp backends.