Would you consider 5t/s usable for a local model?

A user is running a qwen3.8 27b local model, achieving 20t/s when split between a 3060 12GB and 9070xt, or 5t/s using an iGPU and 5400mhz DDR5. Despite the 5t/s being four times slower, the user finds it more usable because it allows them to continue using their system normally, including playing games, as the CPU remains unaffected.

This report highlights a user's preference for a slower 5t/s local model setup over a 20t/s one, prioritizing system usability over raw speed, unlike typical performance benchmarks.