NVIDIA Helps Create “Personal Supercomputers”

Over the course of the past year, NVIDIA has been actively pushing their GPUs as excellent performers outside of gaming, and for most of us, this was a rather new concept. We quickly found out, though, that GPUs were indeed superb performers when executed in a certain manner.

It goes without saying that where GPGPU programming is concerned, we’re still in its infancy, but over the course of the past year, I’ve slowly become more of a believer in what the technology can accomplish. If you care at all about the progress of research in the medical field, then just taking a look at how much more a GPU can accomplish with Folding@home when compared to a normal CPU will help make you a believer.

“Personal Supercomputer” is a term that’s been thrown around quite a bit in the past, but when dealing with scenarios that can execute off of a GPU architecture, then it’s far easier to accomplish. To help make this point even clearer, NVIDIA has published a release that unveiled their “Personal Supercomputer” platform, which numerous vendors are currently supporting. Each configuration will vary, but the fastest ones will include the Tesla C1060. The largest configurations in a single-chassis/rack unit will deliver upwards of 3.732 TFLOPS of computing power.

By using their GPUs, NVIDIA claims, you can achieve “Cluster Class” performance with 1/100th of the available space, and at 1/10th the overall power consumption. Those claims are huge, but believable. I’m no coder, and I certainly don’t handle any supercomputers, but I’m still not entirely confident that GPUs can replace CPUs for any kind of SC computing, but if you’re lucky enough that GPUs will work out just fine, the performance benefits are nothing short of jaw-dropping.

“GPUs have evolved to the point where many real world applications are easily implemented on them and run significantly faster than on multi-core systems,” said Prof. Jack Dongarra, director of the Innovative Computing Laboratory at the University of Tennessee and author of LINPACK. “Future computing architectures will be hybrid systems with parallel-core GPUs working in tandem with multi-core CPUs.”