004.5 . . « email:[email protected] », , (440027, . , . , 40), , – . , , . CUDA Nvidia, . . : , GPU, CUDA, . USING GPU FOR ACCELERATING EVALUATION OF THEORETICAL PRICE OF THE OPTION CREDIT DERIVATIVE Ovechkin R. M. Penza State University, Penza, Russia (440027, Penza, Krasnaya st., 40), email:[email protected] In the article the new approach has represented for accelerating of calculations based on using graphical processing unit which may be applied for approximate evaluating the one of the most significant analitical parameters - theoretical price of derivative financial product – Option integrated into decission support system in context of credit derivatives trading. Given comparative characteristics of performance of server CPU versus the performance of GPU as well as memory interacting process, the evolution of both types of devices is illustrated. describes method of using technology CUDA by Nvidia corporation which allows to create hi-level programs and run them on GPU. Explained the method of adaptation for complex algorithms for execution. Key words: graphical processor unit, GPU, CUDA, performance acceleration. , . GPU (Graphics Processor Unit) – ( ). . : - GPU , . - GPU ; ( . ). , NVIDIA ALU – , - . , , , . – GPU , ) SIMD . , , . . . , quad- 16 ( HyperThreading). 32 , , , (warp) (threads). , 32 NVIDIA 768 , . GTX280, 1024 , 30 ), (GeForce 30000 . , . . , . ( ) – . , GPU , " " CPU. 32 GPU . , . , GPU. , , . GPU CPU . , , 64- GPU NVIDIA GT200. , , , , , 1 , CPU. – FLoat OPerations (GFLOPS = 109 FLOPS). 1. CPU GPU , , CPU . Intel 3,4 . , GFLOPS, Intel Core i7-975 XE 3,33 NVIDIA Tesla s1070 4000 GFLOPS. : PlayStation3 2000 GFLOPS 1000 GFLOPS. XBOX 360 c . GeForce Nvidia Quadro – 70 , Nvidia Tesla , , GPU- CUDA (Compute Unified Device Architecture). CUDA – NVIDIA, , . CUDA , . 2006 , API . GPGPU (General Purpose Programming using a Graphics Processing Unit). API , . . 2007 , NVIDIA CUDA. NVIDIA , , . , . . CUDA , , C . CUDA , . , GPU : – float, 4 , , CPU . double, 8 . , , GPU double. GPU . , GPU – . , 1680x1050 2 , , . , GPU. ) , , . , . , GPU , CUDA. , , , – . : ( , 2) ( , ). 2. , 3 :« 1», « », « ». – . , 1» , », –« ». 3 3) « 2 . 1», –« » . « », , , , 3. . , , ( – CPU. 4). , , . 4. , , , , GPU, 1. . ., . ComputerWeek2. . // 3. . – 1997. – // 14–15. – . 32–39. . . – 2008. – 4–5. Misra M. Design of Systolic arrays for QR Decomposition / Manoj Misra, Rajat Moona. - 1994. [ ]. – : http://www.cse.iitk.ac.in/users/moona/papers /iccse94.pdf/ 4. Kerr A. GPU Performance Assessment with the HPEC Challenge / Andrew Kerr, Dan Campbell, Mark Richards // HPEC 2008, Lexington, 23-25 September 2008 / Lincoln Laboratory, Massachusetts Institute of Technology. – Lexington, 2008. [ ]. – : http://www.ll.mit.edu/HPEC/agendas/proc08/Day3/58-Day3-Session6-Kerr-abstract.pdf. 5. Karimi K. A Performance Comparison of CUDA and OpenCL / Kamran Karimi, Neil G. Dickson, Firas Hamze // D-Wave Systems Inc. [ ]. – http://arxiv.org/ftp/arxiv/papers/1005/1005.2581.pdf. : . ., ., , », . . ., . , . ., , », . « . « :