NVIDIA has given a new meaning to supercomputing by inventing GPU (Graphics Processing Unit) http://en.wikipedia.org/wiki/Graphics_processing_unit and releasing a PSC (Personal Supercomputer) by integrating GPUs. A GPU is a processing unit comprising of multiple cores working simultaneously.
GPU is explained by Mythbusters here:
NVIDIA has launched Tesla personal supercomputers in single processor, desktop and rack-mount units which support Windows 32bit (on Tesla C870 and Tesla D870 only), Windows 64bit, Linux 32bit and 64bit (RedHat Enterprise Linux 4&5, SUSE 10.1,10.2 and 10.3). The Tesla personal supercomputer is scalable from one to one thousand GPUs. Each GPU contains 128 to 240 multi-threaded processors. The processors are scalar thread with full integer and IEEE 754 single-precision floating point support. Initially the memory access was 76.8GB/sec but it has now reached 408GB/sec in Tesla S1070.
High end processing tasks are achieved on personal supercomputer by the C language compiler and standard numerical libraries for Fast Fourier Transform and Basic Linear Algebra Subroutines available for it. NVIDIA CUDA is the software which makes this all done and is available freely from NVIDIA website http://www.nvidia.com/object/cuda_get.html for Windows XP(32bit & 64bit), Windows Vista (32bit & 64bit), Linux (32 bit & 64bit) and Macintosh.
The personal supercomputer products released so far are Tesla C870 computing processor, Tesla D870 desktop GPU system, Tesla S870 GPU computing system, Tesla S1070 and Tesla C1060 processor.
Tesla C870 computing processor contains one GPU, 1.5 GB dedicated memory and can be connected via full-length dual open PCI Express x16 slot.
Tesla D870 desktop GPU system contains 2 GPUs, 3GB memory and can be connected to a host via cabling to a low power PCI Express x8 or x16 adapter card. This desktop system is ideal for office environment because of its quiet operation (40dB).Tesla S870 GPU computing system contains 4 GPUs with 6GB memory. It also connects to a host via cabling to a low power PCI Express x8 or x16 adapter card. It is available in a 19 inch 1 unit rack-mount chassis.
Tesla S1070 GPU computing system contains 4 GPUs with 16GB memory. It connects to a host via cabling to PCI Express x8 or x16. It contains 960 processing cores (240 cores per processor). The frequency of each core is 1.44GHz.
Tesla C1060 processor contains 1GPU with 4GB GDDR3 memory. It connects to a host via cabling to PCI Express x16. It contains 240 processing cores. The frequency of each core is 1.3GHz.
NVIDIA personal computers are widely accepted by the market.
OptiTex (Israel) has taken the design industry to an entirely new level with its 3D CAD/CAM design technology. They are using the NVIDIA® CUDA™ software development environment to reconstruct their cloth simulation engine’s data and algorithms to run on GPUs. The GPU computing solution has enabled developers to remove bottlenecks in the CPU environment and deliver up to a 10 times performance increase.
The National Center for Atmospheric Research (NCAR) are using NVIDIA® CUDA™ for Microphysics, a crucial but computationally intensive component of WRF (Weather Research & Forecasting Model). Similarly Manfold.net has implemented NVIDIA in the field of GIS. According to Dimitri Rotow, product manager, Manifold.net “It is not an exaggeration to say that NVIDIA CUDA technology could be the most revolutionary development in computing since the invention of the microprocessor. It’s fast, inexpensive and loaded with potential. NVIDIA CUDA is so important that all Manifold users should insist that the computer hardware they procure is CUDA-enabled.”
Techniscan has ported its proprietary inverse scattering algorithm for medical sciences from a FORTRAN and MPI system to NVIDIA® CUDA™
NVIDIA personal super computers are available under the price range of $10,000 and are thus easily affordable to scientists and researchers.