Low Latency Packet Capture

Question

I am looking for a solution for capturing live data (TCP and broadcast/multicast over UDP) with little latency impact (e.g. < 50 nanosecond) with timestamping in nano or in some stable clock cycle (e.g. some stable oscillator) given:

1. no additional switch (as no private switch is allowed in the datacenter)

2. no additional server

3. any hardware that can stay inside an 1U server but not outside (optical splitter is okay)

4. only capturing the traffic in one physical optical line

I have tried (onload_)tcpdump and the latency impact is at microsecond level.

Any suggestion is welcome.