2011 sockets and the respective Extreme Edition processors should have better SLI and Crossfire performance than Haswells in 1150 sockets.

First, and most important is the limited number of PCIE lanes from the Hasewll: 16 that have to be split among the PCIE3 sockets. By default, the first socket can use all 16 lanes if none of the other slots are occupied, but that's not SLI nor Crossfire. For two slots filled, the 16 lanes are split 8/8 and can host 2-way SLI or crossfire. The bandwidth is cut in half, though. With three slots filled, the split is 8/4/4. That's not good enough for SLI, but Crossfire can struggle through with much reduced bandwidth.

In contrast, processors for 2011 sockets have enough PCIE3 lanes for two SLI or Crossfire cards with 16 lanes each and a worst case split of 16/8/8/8 for 4-way SLI or Crossfire. They should work better in bandwidth terms let alone not having a chance with only 3 sockets on default 1150 motherboards.

The lane switch (plex) found on such top-end gaming cards as the Maximus VI Extreme tries to expand the available lanes to drive 4 slots, but has to time-share the lanes to do that. In theory, there should be better performance with more native lanes of the 2100 socket than the plex trying to serve 3 or 4 slots with one slots worth of lanes.

I'm not surprised at the small difference between the choices with 2-way SLI.

I am surprised the plex choip does as well as it does - better at 4-way SLI than the 2011 chip. For the plex chip to be better than 2011 socket processor shows that the split of native lanes to 16/8/8/8 has lower bandwidth than the switched lanes with all 4 cards getting 16. Maybe that's why the plex chip needs such a healthy heatsink -- it's doing a lot of work.

Jeff

It took some time to read this article... 🙂 Thanks Wolverine for posting it! It was interesting...

I would have expected bigger differences between the two platforms... but well, facts are facts...
In benchmarking 3- and 4-way showed some difference as I was expecting, but as I said, I would have expected more...

Now my only question would be the difference with the older LGA2011, specifically with SB-E... those would clock higher under the same cooling, but they're less efficient...

Well given the narrow limits of the test, honestly it didn't surprise me much...

I guess most people are in the 2 card SLI or single card category and even the change from PCIe 2.0 to 3.0 is not affecting performance for most. If you test at PCIe 3.0 only then bandwidth is no where near maxed out in dual SLI config... if your CPU is not limiting performance there will be little difference.

So move up to 3 or 4 way and use a Plex chip and you'll get more of the same.

Of course 3 and 4 way without Plex is another story...and more-so PCIe 2.0 vs PCIe3.0 so to call the two platforms equal, at HIGH MB level, is fair enough (after all the price is similar ;)) it shouldn't but maybe this will mislead some into thinking that lower level 1150 MBs are equal and this is not the case I suspect? (like people tend to get headline impressions)...