To understand the architecture of super computers, we must understand the terminology of sockets and cores. A socket is the location where the CPU is placed. Same as a wall socket, an electrical plug is placed into it, a CPU socket allows the CPU (or processor) to plug into it. This allows the CPU to be unplugged at any time, so no soldering is needed.
A core represents the actual processor, where the instructions are read and executed. A single CPU has one core, but within the last few years, Intel and AMD have been making dual core and quad core CPUs. That means that for every dual core processor, there are two CPUs on the chip and for every quad core CPU, there are four processors on the chip (or four cores). This allows the CPU to execute four different instructions simultaneously. Ideally, this would mean the the computer runs four times faster than a CPU chip that has only one core.
With that said, consider a multitude of CPUs connected together that act as one single processing unit. Enter the super computer. A super computer is stereotypically thought of as one machine, but in essence, it is a cluster of processors joined together and programmed to act as one single machine. Same concept as a dual core CPU, but with thousands of more processors.
Currently, super computer process instructions in units of petaflops per second; that is one quadrillion instructions per second (or FLoating-point Operations Per Second -FLOPS). Compared to today’s PC which runs an average of 3.5 GHz or 3.5 billion instructions per second. Fast as that may sound, it doesn’t stand up the super computer, which is over 10,000 times faster.
The number of actual FLOPS is calculated by the formula below: (Note: This formula does not take into account bandwidth, bus speed, memory capacity or other component constraints that may inhibit the actual number calculated)
FLOPS = (sockets * cores / sockets) * clock speed * (FLOPS / cycle)
What is the fastest computer to date?
Currently the number 1 computer in the category of ‘fast’ is the Tianhe-2 super computer, located at China’s National University of Defense Technology. If runs a mind boggling 33.86 petaflops per second. It uses three million Intel Xeon E5-2692v2 12C processors at 2.2GHz each.
The next runner up is Titan, developed at the Oak Ridge National Laboratory in Tennessee and built by Cray Corporation, which is known for decades as the world’s foremost super computer manufacturer, It can run 20,000 trillion calculations per second or 20 petaflops per second, using over 6000 AMD Opteron processors.
As per Datamation.com, one calculation that Titan can process in one second would equate to 1,000 people working on one calculation per second for 60,000 years.