PCIe Gen3 SmartNICs seem to have lower throughput than PCIe Gen2 and PCIe Gen1 SmartNICs
Are you using a free-running clock configuration?
When a server is configured with multiple SmartNICs from different hardware generations in a free-running clock configuration, throughput may appear to differ among the SmartNICs. This is a result of clock differences, as the clocks may have a very noticeable drift among each other. Since the SmartNIC clock is used for time-stamping packets, this can cause various problems, such as differences in throughput measurements. For instance, for a given, stable input, one SmartNIC can show a lower throughput than another.
Consider setting up time synchronization between the SmartNICs, which will keep all SmartNIC clocks synchronized to the configured source.