Disclaimer: This post has more to do with the technical aspect of the new supercomputer, so my feelings won’t get hurt if you feel like zoning out on this one. However, if floating point operations per second (FLOPS) and processing speeds excite you, keep reading.
Freed-Hardeman University now has a supercomputer, but what makes it so super? The power behind this computer lies in its graphics cards. A CPU is the main processing unit in a regular computer, such as a Macbook or a desktop. For applications such as calculating stock prices, reconstructing MRIs, or, in our case, folding proteins, precise floating point numbers are required. These applications call for many calculations involving these numbers. Unfortunately, computing floating numbers is not a highly optimized operation for a conventional CPU. Graphics cards are specialized processing units that are specifically optimized to compute floating numbers. If you need to compute a lot of floating numbers, graphics cards provide high performance at relatively low cost. This graph compares the computing performance of one Nvidia Tesla C2050 graphics card vs. an Intel i7 processor while running a benchmark in which they were required to multiply matrices of various complexity.
Floating point performance (single precision) for multiplying two matrices using an Intel Core i7-975 Extreme CPU and an NVIDIA Tesla C2050 GPU. [http://wiki.accelereyes.com/wiki/index.php/Jacket_Floating_Point_Performance_(GFlops)]
The typical i7 desktop processor can do 42.56 GigaFLOPS. The graphics cards in our supercomputer can do 1.03 TeraFLOPS each. Our system has two computing nodes, each with 4 graphics cards, plus a third node containing two CPUs to control everything. The graphics computing power of the supercomputer is 8.24 TFLOPS. If we do the math, we find that the supercomputer can do nearly 200 times as many floating point calculations per second as the average desktop processor.
To further put things in perspective, our supercomputer pales in comparison to the fastest supercomputer in the world today which can compute 1759 TFLOPS1. However, the fastest supercomputer in the world back in 2000 could only compute 7.3 TFLOPS2. Our supercomputer beats the best supercomputer in the world ten years ago. That sounds pretty super, doesn’t it?
Until next time,