N A G P A R I S280101

N A G P A R I S280101

• 1.
  GPU OVERVIEW IN FINANCIALSERVICES ALASTAIR HOUSTON COMPUTE FSI SALES MANAGER
• 2.
  Agenda Nvidia and HPC markets GPU Overview CUDA and OpenCL Current FS deployments © NVIDIA Corporation 2009
• 3.
  CUDA Runs onNVIDIA GPUs … Over 80 Million CUDA GPUs Deployed GeForce® TeslaTM Quadro® Entertainment High-Performance Computing Design & Creation © NVIDIA Corporation 2009
• 4.
  146X 36X 18X 50X 100X Medical Imaging Molecular Dynamics Video Transcoding Matlab Computing Astrophysics U of Utah U of Illinois, Urbana Elemental Tech AccelerEyes RIKEN 50x – 150x 149X 47X 20X 130X 30X Financial simulation Linear Algebra 3D Ultrasound Quantum Chemistry Gene Sequencing Oxford Universidad Jaime Techniscan U of Illinois, Urbana U of Maryland © NVIDIA Corporation 2009
• 5.
  Options Pricing, RiskModeling, Algorithmic Trading Options pricing use Monte Carlo (MC) simulations Random Number Generators (RNG) are key to MC Up to 100x speed-up in RNGs using CUDA 25-60x overall speedup in Monte Carlo simulations © NVIDIA Corporation 2009
• 6.
  Co-Processing CPU GPU The Right Processor for the Right Tasks © NVIDIA Corporation 2009
• 7.
  The Performance GapWidens Further 8x double precision ECC L1, L2 Caches 1 TF Single Precision 4GB Memory NVIDIA GPU © NVIDIA Corporation 2009 X86 CPU
• 8.
  Introducing the ‘Fermi’Architecture The Soul of a Supercomputer in the body of a GPU 3 billion transistors DRAM I/F DRAM I/F DRAM I/F Over 2× the cores (512 total) 8× the peak DP performance DRAM I/F DRAM I/F HOST I/F ECC L2 L1 and L2 caches Giga Thread DRAM I/F DRAM I/F ~2× memory bandwidth (GDDR5) Up to 1 Terabyte of GPU memory DRAM I/F DRAM I/F DRAM I/F Concurrent kernels Hardware support for C++ © NVIDIA Corporation 2009
• 9.
  NVIDIA Compute Products Board Level Products 1U Server Product 1 Tesla GPU 4 Tesla GPUs Workstation Product Data Center Product OEM Product © NVIDIA Corporation 2009
• 10.
  CUDA C andOpenCL Momentum Over 100,000,000 installed CUDA- Architecture GPUs GPU Computing Applications Over 60,000 GPU Computing Developers (1/09) Windows, Linux and MacOS Platforms C OpenCL DirectX FORTRAN Python, supported Compute Java, … With CUDA Extensions Over 60,000 developers 1st GPU demo Microsoft’s GPU Microsoft’ SW supplied by: Compute Kernels GPU Computing spans Shipped 1st OpenCL Computing API • The Portland Group Driver API Bindings Consumer applications Running in Production Driver Supports all CUDA- CUDA- • NCSA release since 2008 to HPC Strategic developers Architecture GPUs SDK + Lib’s + Visual Lib’ since G80 (DX10 and using NV SW today Profiler and Debugger future DX11 GPUs) 200+ Universities teaching the CUDA Architecture and GPU Computing NVIDIA GPU with the CUDA Parallel Computing Architecture © NVIDIA Corporation 2009
• 11.
  NVIDIA Nexus Nexus is a GPU application development suite that integrates directly into Visual Studio. A C/CUDA source debugger for both the CUDA runtime and driver API New C/CUDA performance analysis/trace tools © NVIDIA Corporation 2009
• 12.
  FSI CUSTOMER DEPLOYMENTS ©NVIDIA Corporation 2009
• 13.
  Case Study: EquityDerivatives 15 15x Faster 1 2 Tesla S1070 16x Less Space 500 CPU Cores $24 K 10x Lower Cost $250 K 2.8 KWatts 13x Lower Power 37.5 KWatts Source: BNP Paribas, March 4, 2009 © NVIDIA Corporation 2009
• 14.
  Case Study: SecurityPricing 2 hours 8x Faster 16 hours 48 Tesla S1070 10x Less Space 8000 CPU Cores Source: Wall Street & Technology, September 24, 2009 © NVIDIA Corporation 2009