Romley's arrived - and for those who've been running prev. Sandy Bridge chipsets, there's not a huge suprise, other than to point out the fact that platform latency at with high message rates is greatly improved and peak message rates are stunning.
Here's some performance data showing the uplift achieved moving from Westmere to Romley using the 6122F NICs (our super faster new NIC). This is for a single thread / stream running the sfnt-stream benchmark . As you can see on a Westmere latency starts to ratchet up at around 2Mpps, but stays very flat on the Romley (we used a 3.1Ghz with slow DDR1333 RAM). Still sub-5us at 4Mpps.
The Romley platform also shows off the very high message rates achievable with the 6122F - we measured a huge 20Mpps RX with both ports at socket-level/user-space using 4 threads active.
Sinking line-rate min packet size is also very achievable - welcome to the world of affordable high precision packet capture ...!