A LocalLLaMA user asked for a detailed comparison of Qwen-3.8 27b at 8bpw and Qwen-3.8-FN at 4bpw, saying they have not found one that fits their needs.
The post says the user runs the models on dual RTX 3090s at int8A16 and wants a high-precision comparison.
The user is considering selling their current setup for a DGX Spark, which they say can run QFN at nvfp4 with a large context window. They use Hermes with OpenWebUI for a personal assistant. The post does not provide a comparison or report on the DGX Spark in use.