Our usage scenario is a >10Gbit/sec linux router where the forwarded traffic is loadbalanced across as much cpu cores as possible. At the moment we use Intel X520 nic's. But they are limited to 16 hardware RX interrupts (RSS) per port. This means a dual 12 core machine is not optimally used.
According to the feature summary ( http://www.mellanox.com/related-docs/products/ethernet_nic_brochure_c_2_3.pdf ) a ConnectX 3 nic supports up to 32 RSS queues.
Does this mean the hardware effectively allocates and uses 32 MSI-X RX interrupts on a 32 core system? And to be sure: a dual port nic would allocate 32 RX irq's for each port?.