This graph shows that the relaxed memory model gives lower execution times. Specifically, it does so by essentially eliminating the write latency -- shown by the reduction in the grey bars.
It is also interesting that the time spend reading seems to increase slightly in the relaxed model.
@PID_1 I think the processor actually does not eliminate the write latency but hide it by executing subsequent read immediately in order to reduce total latency. According to this slide the write latency is still the same.
I am also wondering why the read time increases slightly here.
@Hil. Exactly. It "hides" the latency. The latency to memory is certainly still there.