CPU-GPU Workload Distribution during Throughput-Oriented LLM Inference on Single-GPU Systems | Synapse