High Performance Computing (HPC) with Graphical Processor Units (GPU) are at the very heart of the RenderStream product line of heterogeneous MPU/GPU Personal Super Computers and HPC clusters. Heterogeneous computing strives to let each processor do what it is best at. GPU computing is rapidly being adapted by the scientific community because where it is possible, it potentially speeds computations by orders of magnitude.

RenderStream Personal Supercomputer

RenderStream Personal Supercomputer

The computer graphics industry is also adapting this technology but further development is required in both fields before the true computational benefits of GPU computing will be realized. Several different models of RenderStream’s single and dual Nehalem based PSC family of MPU/GPU heterogeneous computing workstations have been built and delivered using the nVIDIA Tesla C1060. The C1060 is a PCI-e card with a single high performance 933 GFLOP GPU with 4 GB of DDR3 memory that can potentially deliver over 3.6 Tera-FLOP (TFLOP) of computing power in a single workstation. For the single quad core PSC systems we use one to four C1060 Tesla units or for the nonprofessional environment, general processing cards like the GTX200 series including the GTX295 dual GPU card that in a four card system is capable of delivering a very professional speed of over 7 TFLOP in one RenderStream PSC workstation making it potentially ideal for Folding@home.

For our HPC Nehalem-based dual quad heterogeneous MPU/GPU computing clusters we use the Tesla S1070 server. This 4 TFLOP server uses four GPU with 4 GB DDR3/processor that through PCI-e adapters allows two to four GPU to be hosted by a single MPU computer. Our configurations truly deliver unprecedented capability to expand scientific computing (and GPU accelerated rendering when it becomes mainstream) to totally new heights. We currently have designs that deliver more than 42 TFLOP of compute power for prices in equipment and in power consumption that were totally unheard of even a year ago. They can be bigger too we just have not needed them to be so yet. Things are changing fast!

To this point we’ve briefly discussed HPC for the scientific community but we also supply hybrid PSC systems for the rendering community which can use a combination of C1060 and Quadro FX or general GPU cards with single and dual Nehalem-based quad cores that are designed to do amazing things in the not so distant future with GPU accelerated products from various suppliers consisting of all the usual suspects and a bunch of new ones too.

There are numerous languages that are maturing, primarily of which is CUDA and OpenCL. But, and this is a BIG BUT, we need more applications to run on these monster machines and that means we need more brilliant computational folk learning to solve problems on the GPU. To that end nVIDIA has announced a GPU CUDA programming contest:

http://www.topcoder.com/tc?module=Static&d1=sponsors&d2=NVIDIAInstructions

Register and take part. To get everything you need for CUDA go to:

http://www.nvidia.com/object/cuda_learn.htmlTesla

To help this development along RenderStream with offices at 12325 Hymeadow Dr Suite 2-201 Austin, Tx, starting today, August 28th and continuing for the next month, will offer a promotional discount of up to $500 for the next ten PSC developmental systems we sell with  two or more C1060 or $800 for one S1070 integrated with a Twin Dual Quad Server with no more than two systems per customer. In addition, we will give a free nVIDIA Tesla C1060 GPU card to the highest placing RenderStream customer who has placed within the top 5 and who  has purchased a computer from RenderStream.

For more information please visit http://www.renderstream.com/HPC.html or feel free to email us at info@renderstream.com