AMD Leverages AI + HRX Loom In Experiment For Quickly Porting 50 LLMs To Client GPUs

AMD engineers recently conducted an experiment to port 50 popular Hugging Face models across 18 architectures to AMD client GPUs in just three to four days using HRX Loom and agentic AI. The testing revealed a 1.5 to 2.1x speed-up for 32K token prompt processing compared to Llama.cpp's ROCm back-end. HRX Loom remains experimental, but supports CPU and NPU execution alongside GPUs. Open article →