Ya I'd be interested to see a comparison of using llamacpp with ssd offloading to compare real speeds.