Debating getting a DGX Spark to offload some AI compute to a local LLM. Does anyone have better recommendations?