Shrinking the Supercomputer with GPU

This past November I was privileged to work the Dell booth at the Supercomputing Show in New Orleans where we were showcasing, among other things, our c410x PCI expansion chassis for GPU computing. I was amazed at the number of people that I spoke to over the course of a week, nearly non-stop about GPU computing and more specifically, the c410x and how they would be using it in their environments. Equally amazing was the sheer number of people that had “Dr.” in front of their name (I had to throw that in there for good measure). Many of these folks were hands-on users of Supercomputing technologies and all of them were “shopping” for their next Supercomputing system.

As I began to speak with them in detail about their use of GPU computing and the benefits they gained from using it over the existing systems, they immediately responded with two, almost word-for-word responses; hardware footprint and cost. Now I don’t want to make you think that GPU computing is “cheap” computing, but when you look at what the traditional supercomputer is made up of, it revolves around CPU computing, and the massing of huge numbers of processors to get a very high core count.

A single Tesla c2050 or c2070 GPU processor exceeds 400 cores. When you multiply that times sixteen GPU’s (the capacity of a single c410x) you get a whopping amount of processing power that was only available before by using a very large number of CPU-based systems. Footprint? Let’s look at it this way. A single PowerEdge c6100 (Intel-based 2u shared infrastructure system) or a single PowerEdge c6105 (AMD-based 2u shared infrastructure system) can power a fully populated c410x (3u) for a total of 5u. If you use the newly announced PowerEdge c6145 (also 2u) you can power two fully populated c410x’s for a total of 33 Teraflops in just 7u.

Who uses this technology? Anyone needing a rendering farm. Companies in the oil and gas exploration field. Bio-Med or anyone needing to crunch a huge amount of data very quickly with finite results.

Finally, the cost factor. nVidia’s says that “When compared to the latest quad-core CPU, Tesla 20-series GPU computing processors deliver equivalent performance at 1/20th the power consumption and 1/10th the cost.” When everyone is looking to save money in their IT environment without sacrificing performance and quality, the GPU-based Supercomputer offers an amazing value proposition.

To learn more about Dell’s PowerEdge C line of servers for Cloud, HyperScale and High Performance Computing, visit http://www.dell.com/poweredgec or talk to your Dell representative.

1 thought on “Shrinking the Supercomputer with GPU”

  1. This is already hitting home machines also; I put in an nVIDIA GeForce 210 in my tower, and for less than $50 it sports GPU processing for converting video (if one uses Roxio 2011 or vReveal it supports using the GPU for conversion processing, dramatically speeding things up).

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s