Diminishing return considerations when specifying servers

Question

With some of the new 10-core Xeon E7 ,which can work on 20 simultaneous computational threads along with quad-socket motherboards it is possible to specify servers with 80 computational threads and 1 TB of memory. Though it is certainly possible to utilize all of these resources, does anyone know when the laws of diminishing returns start to kick in?

From a Windows Server 2008 R2 perspective, how does the OS handle all of these resources?

Is there a fairly well known maximum that should be considered?

Answer 1

There's no answer to this question. I bet someone asked the same thing when the first quad-core with HT processors hit the market (a few years ago now), where you could easily total 32 threads in a single chassis (note, that that's still only 16 cores).

Now, I wouldn't spec a system with less than 12 cores in total (which with Intel gives 24 threads), but I can achieve that in just dual socket.

I still remember when I specced up a Dual Athlon MP as a workstation back in 2002. All my friends and collegues thought I was insane, and they were very jelous that my Windows XP task manager showed OMFG TWO theads and a full gigabyte of RAM. In a WORKSTATION. My laptop just 9 years later has 8 threads, 8Gb of RAM, 750Gb of HDD and a TV tuner.

If you're buying with a 5-year plan, in 5 years time that 80-thread (which is still "only" 40 cores) and your 1Tb of RAM may very well be closing in as the "norm" for a virtualisation or database server.