Upon reaching a certain limit on new TLS handshakes, the handshakes can be efficiently batched and offloaded to GPU.