Our use case is atypical, based on what others seem to require. While we serve multiple requests in parallel, our workloads are not 'chat'.