Why is L2 ROP Latency Greater Than DRAM Latency

[quanzw99]

According to my intuition, L1 should be the fastest, followed by L2, and then DRAM.

However, in the RTX 2060 configuration line:140, it seems that l2_rop_latency is 194, while dram_latency is 96;
Similarly, in the SM75_RTX2060_S configuration line:174, l2_rop_latency is 160, and dram_latency is 100.

SO I have two questions:
Why can l2_rop_latency be higher than dram_latency?
What is ROP? I checked the GPGPU-Sim official documentation, which explains it as an observed latency, but I still find it confusing.

[JRPan]

I think the real "DRAM latency" that counts how long the request returns, which is what you are looking for, is l2_rop_latency + dram_latency.

So they are not inclusive.