APPLICATIONS

With the advent of technology advancements in the IT industry and with graphic card manufacturers optimizing their hardware for parallel programming, we are in a period where the domination of traditional serial programming comes to an end. Using parallel processors allows the reduction of computing times, hardware costs and electric consumption. An increasing number of companies propose GPU based products and services in order to harness their enormous computing power. Graphic card manufacturers have already launched new interfaces for their cards to facilitate GPU-Computing or Stream programming (CTM for AMD and CUDA for Nvidia). However, these solutions present a number of limitations (no double precision, compatibility with only a specific graphic card, require good GPU knowledge…) GPU programming is therefore, still reserved for a select few. In order to allow any developer to program their applications on the graphic card, GPU-Tech today launches its libraries and API Ecolib. The user can then divide the computing time of applications by 10 or more without changing his working habits. Two means are proposed by GPU-Tech: • Automatically port parts of the code to GPU using the “Ecolib” libraries • Program the entire code in C++ in an optimised way on the GPU using the “Ecolib” API. Thanks to the innovative technology by GPU-Tech you can for the first time program your applications in double precision on the graphic card. Furthermore, the applications programmed using the Ecolib API and libraries are compatible with all GPUs Pixel Shaders 2.x, 3.0 and 4.0 whichever the manufacturer. GPU-Tech is in collaboration with AMD and Nvidia to ensure optimized performances on these cards. Finally, the Ecolib libraries presently offer the best performances in GPU-Computing. Five Ecolib libraries are presently proposed: • Computing of options and VaR • Generation of random numbers • Basic Linear algebra (BLAS) • Resolution of linear system equations (LAPACK) • Signal processing (FFT) and image processing These regroup algorithms used in numerous sectors like finance, oil and gas, aeronautics, life sciences, meteorology, acoustics… GPU-Tech also proposes personalised development to companies wishing to optimise their own applications on the GPU. GPU-Tech ports the application directly in assembler language on the card. GPU-Tech has put up several benchmarks (BLAS, Monte Carlo, Black & Scholes) of its “Ecolib” libraries on their website with notably 90 Gflops for SGEMM on a GeForce 8800 GTS and 30 Gflops for DGEMM on an ATI HD 2900 XT. In the graphic industry, GPU-Tech has already developed a 3D rendering engine on GPU using the Ray tracing and Photon mapping methods. RTSquare renders out photo realistic images in rendering times much faster than other applications of this kind.