Intel Xeon processors: Difference between 2S/4S and S2S/S4S scalability

Solution 1:

Based on " MPSUP: Multiprocessor Support" in Intel® Xeon® Processor E7-4800/8800 v3 Product Families; Datasheet - Volume 1: Electrical, Mechanical and Thermal (pdf, page 139):

This location contains 2 bits for representing the supported number of physical processors on the bus. These two bits are LSB aligned where 00b equates to nonscalable 2 socket (2S) operation, 01b to scalable 2 socket (S2S), 10 to scalable 4 socket (S4S), and scalable 8 socket (S8S). The Intel® Xeon® E7 v3 processor is a S2S, S4S, or S8S processor. The first six bits in this field are reserved for future use. Writes to this register have no effect.

  • S prefix:
    • Exists: Scalable (based on the language, I believe this means you can use fewer processors)
    • Missing: Nonscalable (you must have the specified number of processors)
  • Number:
    • Number of processors (if scalable, this is maximum number of processors; if nonscalable, this is the required number of processors)
  • S postfix:
    • Stands for "socket"

So, for example, S4S is "scalable four socket" and 2S is "nonscalable two socket".

Solution 2:

The difference between S4S and 4S is the number of QPI-Links. 4S has 2 links per CPU, S4S and S8S have 3.

On a 4S system CPUs form a ring, which means for every CPU there is another CPU with which it can only communicate indirectly, because it is only connected to 2 CPUs via QPI.

On S4S every CPU is directly connected to every other CPU.

For S8S there are also 3 QPI links.

Question is a little older, but maybe my answer is still relevant.