I just enabled L3HW offload on my ccr2116-12g-4s+ and it is magical to see NAT'd IPv4 internet traffic routed at 10G line speed w/ <5% CPU usage. Is there any trade-off for offloading to the switch chip instead of CPU routing, particularly with respect to latency? One of the Mikrotik examples mentioned something like, "gamers on your network won't be happy" and I'm wondering what this might mean. Thank you!

I've got three border CCR2116's and a CRS317 in the core, all doing about 2Gbps of data non-stop (with L3HW offload), and a mix of CRS310's and RB4011's at the edge. At about 300-500Mbps each, the CRS310's with L3HW offload aren't even breaking a sweat (1-3% always) compared to the CPU-bound RB4011's, which range from 5% (150Mbps) to 20% (600Mbps).

I have easily hundreds of gamers on my network, and nobody's complained about latency. The switch chips are designed to switch packets faster than the CPU's can.

Without know what Mikrotik examples are being referred to, it is hard to say.

wild guess: the offloading will allow the line to be saturated, and the offloading engines don't support any type of flow based QoS, so some bulk transfer could "unfairly" dominate the bandwidth.