Problem description & steps to reproduce Summary When running llama-server with the Vulkan backend on Intel Arc A770, sustained parallel translation workload can make the whole Windows machine become ...