GPU databases have a higher latency, due to the need to copy data to and from the GPU. The data processed has to end up in the host RAM, to be sent to the client - it can't be sent directly from the GPU RAM.
The GPU really only helps when you need a higher throughput - eg. you're analyzing many billions of rows, or are running a relatively compute-intensive query.
The GPU really only helps when you need a higher throughput - eg. you're analyzing many billions of rows, or are running a relatively compute-intensive query.