Intel is feeling intense pressure from
the ARM SoC market, where significant gains in SoC designs intended for cellphones and tablets have resulted in better performance-per-watt for recent and planned designs.
In the past 15 years, multiprocessing has gone from a semi-obscure branch of computing to totally mainstream, with even your cellphone likely containing multiple cores. For some tasks, multiprocessing is totally awesome. For others, less so. Linked article focuses kind of on "porting stuff from Windows" which is just one aspect.
But, ultimately, bet on smaller processors in the long run.
In ~2000, I had a large customer who happened to be the largest ISP in the region. They had a server room which was, if I recall correctly, three rows of five racks on each side, 30 racks, 8 servers per rack. Intel 440BX gear, largely. 256-512MB average RAM per unit, in aggregate that would be less than 128GB RAM. Aggregate hard drive storage capacity was maybe 10TB, which included a massive 5TB array of 73GB drives that filled an entire rack. Average CPU was an Intel PIII 550 (ranged from 233's on up to 1GHz)... or 132GHz worth of CPU.
So, an observation. Today I can buy 128GB of RAM and stuff it in a single server, about $1000. 10TB of space? Three 4TB hard drives, $450. But 132GHz worth of CPU, that's 48 cores. I can stick 24 cores in a server (2 x E5-2697v2) at a cost of about $6100. Yiiikes.
Which of those things is out of line and not scaling well? Hint: CPU.
The real problem is that sticking lots of cores on a single socket is hard and expensive, and making them fast, even worse. IF you have an application that can be parallelized, maybe not just across cores but even across servers, then small processors are a win. So Intel is trying to figure out some middle ground, because basically they stand to suffer in the long run if computers (and especially servers) move towards fleets of cheap SoC chips.
But ... there's a flip side to this. The Samba implementation of CIFS isn't particularly good on slower hardware, and ECC is very important for proper operation of ZFS. So FreeNAS works better on systems with fewer fast cores and ECC. That's a good match for Xeon, and potentially for the higher end Rangeley stuff. But I wonder what we'll actually end up seeing, since some of that would appear to directly conflict with Intel's lower end i3 and Xeon E3 offerings.
So... yes, worth keeping an eye on. If you can wait 3-6 months, we'll probably have a much better idea.
